drbd_nl.c revision bf885f8a6772fb48409dd505a09d974a5e621f22
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 44static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int); 45static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *); 46static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *); 47 48/* see get_sb_bdev and bd_claim */ 49static char *drbd_m_holder = "Hands off! this is DRBD's meta data device."; 50 51/* Generate the tag_list to struct functions */ 52#define NL_PACKET(name, number, fields) \ 53static int name ## _from_tags(struct drbd_conf *mdev, \ 54 unsigned short *tags, struct name *arg) __attribute__ ((unused)); \ 55static int name ## _from_tags(struct drbd_conf *mdev, \ 56 unsigned short *tags, struct name *arg) \ 57{ \ 58 int tag; \ 59 int dlen; \ 60 \ 61 while ((tag = get_unaligned(tags++)) != TT_END) { \ 62 dlen = get_unaligned(tags++); \ 63 switch (tag_number(tag)) { \ 64 fields \ 65 default: \ 66 if (tag & T_MANDATORY) { \ 67 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \ 68 return 0; \ 69 } \ 70 } \ 71 tags = (unsigned short *)((char *)tags + dlen); \ 72 } \ 73 return 1; \ 74} 75#define NL_INTEGER(pn, pr, member) \ 76 case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \ 77 arg->member = get_unaligned((int *)(tags)); \ 78 break; 79#define NL_INT64(pn, pr, member) \ 80 case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \ 81 arg->member = get_unaligned((u64 *)(tags)); \ 82 break; 83#define NL_BIT(pn, pr, member) \ 84 case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \ 85 arg->member = *(char *)(tags) ? 1 : 0; \ 86 break; 87#define NL_STRING(pn, pr, member, len) \ 88 case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \ 89 if (dlen > len) { \ 90 dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \ 91 #member, dlen, (unsigned int)len); \ 92 return 0; \ 93 } \ 94 arg->member ## _len = dlen; \ 95 memcpy(arg->member, tags, min_t(size_t, dlen, len)); \ 96 break; 97#include "linux/drbd_nl.h" 98 99/* Generate the struct to tag_list functions */ 100#define NL_PACKET(name, number, fields) \ 101static unsigned short* \ 102name ## _to_tags(struct drbd_conf *mdev, \ 103 struct name *arg, unsigned short *tags) __attribute__ ((unused)); \ 104static unsigned short* \ 105name ## _to_tags(struct drbd_conf *mdev, \ 106 struct name *arg, unsigned short *tags) \ 107{ \ 108 fields \ 109 return tags; \ 110} 111 112#define NL_INTEGER(pn, pr, member) \ 113 put_unaligned(pn | pr | TT_INTEGER, tags++); \ 114 put_unaligned(sizeof(int), tags++); \ 115 put_unaligned(arg->member, (int *)tags); \ 116 tags = (unsigned short *)((char *)tags+sizeof(int)); 117#define NL_INT64(pn, pr, member) \ 118 put_unaligned(pn | pr | TT_INT64, tags++); \ 119 put_unaligned(sizeof(u64), tags++); \ 120 put_unaligned(arg->member, (u64 *)tags); \ 121 tags = (unsigned short *)((char *)tags+sizeof(u64)); 122#define NL_BIT(pn, pr, member) \ 123 put_unaligned(pn | pr | TT_BIT, tags++); \ 124 put_unaligned(sizeof(char), tags++); \ 125 *(char *)tags = arg->member; \ 126 tags = (unsigned short *)((char *)tags+sizeof(char)); 127#define NL_STRING(pn, pr, member, len) \ 128 put_unaligned(pn | pr | TT_STRING, tags++); \ 129 put_unaligned(arg->member ## _len, tags++); \ 130 memcpy(tags, arg->member, arg->member ## _len); \ 131 tags = (unsigned short *)((char *)tags + arg->member ## _len); 132#include "linux/drbd_nl.h" 133 134void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name); 135void drbd_nl_send_reply(struct cn_msg *, int); 136 137int drbd_khelper(struct drbd_conf *mdev, char *cmd) 138{ 139 char *envp[] = { "HOME=/", 140 "TERM=linux", 141 "PATH=/sbin:/usr/sbin:/bin:/usr/bin", 142 NULL, /* Will be set to address family */ 143 NULL, /* Will be set to address */ 144 NULL }; 145 146 char mb[12], af[20], ad[60], *afs; 147 char *argv[] = {usermode_helper, cmd, mb, NULL }; 148 int ret; 149 150 snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev)); 151 152 if (get_net_conf(mdev)) { 153 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) { 154 case AF_INET6: 155 afs = "ipv6"; 156 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6", 157 &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr); 158 break; 159 case AF_INET: 160 afs = "ipv4"; 161 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4", 162 &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr); 163 break; 164 default: 165 afs = "ssocks"; 166 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4", 167 &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr); 168 } 169 snprintf(af, 20, "DRBD_PEER_AF=%s", afs); 170 envp[3]=af; 171 envp[4]=ad; 172 put_net_conf(mdev); 173 } 174 175 /* The helper may take some time. 176 * write out any unsynced meta data changes now */ 177 drbd_md_sync(mdev); 178 179 dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb); 180 181 drbd_bcast_ev_helper(mdev, cmd); 182 ret = call_usermodehelper(usermode_helper, argv, envp, 1); 183 if (ret) 184 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n", 185 usermode_helper, cmd, mb, 186 (ret >> 8) & 0xff, ret); 187 else 188 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n", 189 usermode_helper, cmd, mb, 190 (ret >> 8) & 0xff, ret); 191 192 if (ret < 0) /* Ignore any ERRNOs we got. */ 193 ret = 0; 194 195 return ret; 196} 197 198enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev) 199{ 200 char *ex_to_string; 201 int r; 202 enum drbd_disk_state nps; 203 enum drbd_fencing_p fp; 204 205 D_ASSERT(mdev->state.pdsk == D_UNKNOWN); 206 207 if (get_ldev_if_state(mdev, D_CONSISTENT)) { 208 fp = mdev->ldev->dc.fencing; 209 put_ldev(mdev); 210 } else { 211 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n"); 212 nps = mdev->state.pdsk; 213 goto out; 214 } 215 216 r = drbd_khelper(mdev, "fence-peer"); 217 218 switch ((r>>8) & 0xff) { 219 case 3: /* peer is inconsistent */ 220 ex_to_string = "peer is inconsistent or worse"; 221 nps = D_INCONSISTENT; 222 break; 223 case 4: /* peer got outdated, or was already outdated */ 224 ex_to_string = "peer was fenced"; 225 nps = D_OUTDATED; 226 break; 227 case 5: /* peer was down */ 228 if (mdev->state.disk == D_UP_TO_DATE) { 229 /* we will(have) create(d) a new UUID anyways... */ 230 ex_to_string = "peer is unreachable, assumed to be dead"; 231 nps = D_OUTDATED; 232 } else { 233 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate"; 234 nps = mdev->state.pdsk; 235 } 236 break; 237 case 6: /* Peer is primary, voluntarily outdate myself. 238 * This is useful when an unconnected R_SECONDARY is asked to 239 * become R_PRIMARY, but finds the other peer being active. */ 240 ex_to_string = "peer is active"; 241 dev_warn(DEV, "Peer is primary, outdating myself.\n"); 242 nps = D_UNKNOWN; 243 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE); 244 break; 245 case 7: 246 if (fp != FP_STONITH) 247 dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n"); 248 ex_to_string = "peer was stonithed"; 249 nps = D_OUTDATED; 250 break; 251 default: 252 /* The script is broken ... */ 253 nps = D_UNKNOWN; 254 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff); 255 return nps; 256 } 257 258 dev_info(DEV, "fence-peer helper returned %d (%s)\n", 259 (r>>8) & 0xff, ex_to_string); 260 261out: 262 if (mdev->state.susp_fen && nps >= D_UNKNOWN) { 263 /* The handler was not successful... unfreeze here, the 264 state engine can not unfreeze... */ 265 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE); 266 } 267 268 return nps; 269} 270 271static int _try_outdate_peer_async(void *data) 272{ 273 struct drbd_conf *mdev = (struct drbd_conf *)data; 274 enum drbd_disk_state nps; 275 276 nps = drbd_try_outdate_peer(mdev); 277 drbd_request_state(mdev, NS(pdsk, nps)); 278 279 return 0; 280} 281 282void drbd_try_outdate_peer_async(struct drbd_conf *mdev) 283{ 284 struct task_struct *opa; 285 286 opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev)); 287 if (IS_ERR(opa)) 288 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n"); 289} 290 291enum drbd_state_rv 292drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force) 293{ 294 const int max_tries = 4; 295 enum drbd_state_rv rv = SS_UNKNOWN_ERROR; 296 int try = 0; 297 int forced = 0; 298 union drbd_state mask, val; 299 enum drbd_disk_state nps; 300 301 if (new_role == R_PRIMARY) 302 request_ping(mdev); /* Detect a dead peer ASAP */ 303 304 mutex_lock(&mdev->state_mutex); 305 306 mask.i = 0; mask.role = R_MASK; 307 val.i = 0; val.role = new_role; 308 309 while (try++ < max_tries) { 310 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE); 311 312 /* in case we first succeeded to outdate, 313 * but now suddenly could establish a connection */ 314 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) { 315 val.pdsk = 0; 316 mask.pdsk = 0; 317 continue; 318 } 319 320 if (rv == SS_NO_UP_TO_DATE_DISK && force && 321 (mdev->state.disk < D_UP_TO_DATE && 322 mdev->state.disk >= D_INCONSISTENT)) { 323 mask.disk = D_MASK; 324 val.disk = D_UP_TO_DATE; 325 forced = 1; 326 continue; 327 } 328 329 if (rv == SS_NO_UP_TO_DATE_DISK && 330 mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) { 331 D_ASSERT(mdev->state.pdsk == D_UNKNOWN); 332 nps = drbd_try_outdate_peer(mdev); 333 334 if (nps == D_OUTDATED || nps == D_INCONSISTENT) { 335 val.disk = D_UP_TO_DATE; 336 mask.disk = D_MASK; 337 } 338 339 val.pdsk = nps; 340 mask.pdsk = D_MASK; 341 342 continue; 343 } 344 345 if (rv == SS_NOTHING_TO_DO) 346 goto fail; 347 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) { 348 nps = drbd_try_outdate_peer(mdev); 349 350 if (force && nps > D_OUTDATED) { 351 dev_warn(DEV, "Forced into split brain situation!\n"); 352 nps = D_OUTDATED; 353 } 354 355 mask.pdsk = D_MASK; 356 val.pdsk = nps; 357 358 continue; 359 } 360 if (rv == SS_TWO_PRIMARIES) { 361 /* Maybe the peer is detected as dead very soon... 362 retry at most once more in this case. */ 363 __set_current_state(TASK_INTERRUPTIBLE); 364 schedule_timeout((mdev->net_conf->ping_timeo+1)*HZ/10); 365 if (try < max_tries) 366 try = max_tries - 1; 367 continue; 368 } 369 if (rv < SS_SUCCESS) { 370 rv = _drbd_request_state(mdev, mask, val, 371 CS_VERBOSE + CS_WAIT_COMPLETE); 372 if (rv < SS_SUCCESS) 373 goto fail; 374 } 375 break; 376 } 377 378 if (rv < SS_SUCCESS) 379 goto fail; 380 381 if (forced) 382 dev_warn(DEV, "Forced to consider local data as UpToDate!\n"); 383 384 /* Wait until nothing is on the fly :) */ 385 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0); 386 387 if (new_role == R_SECONDARY) { 388 set_disk_ro(mdev->vdisk, TRUE); 389 if (get_ldev(mdev)) { 390 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1; 391 put_ldev(mdev); 392 } 393 } else { 394 if (get_net_conf(mdev)) { 395 mdev->net_conf->want_lose = 0; 396 put_net_conf(mdev); 397 } 398 set_disk_ro(mdev->vdisk, FALSE); 399 if (get_ldev(mdev)) { 400 if (((mdev->state.conn < C_CONNECTED || 401 mdev->state.pdsk <= D_FAILED) 402 && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced) 403 drbd_uuid_new_current(mdev); 404 405 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1; 406 put_ldev(mdev); 407 } 408 } 409 410 if ((new_role == R_SECONDARY) && get_ldev(mdev)) { 411 drbd_al_to_on_disk_bm(mdev); 412 put_ldev(mdev); 413 } 414 415 if (mdev->state.conn >= C_WF_REPORT_PARAMS) { 416 /* if this was forced, we should consider sync */ 417 if (forced) 418 drbd_send_uuids(mdev); 419 drbd_send_state(mdev); 420 } 421 422 drbd_md_sync(mdev); 423 424 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); 425 fail: 426 mutex_unlock(&mdev->state_mutex); 427 return rv; 428} 429 430static struct drbd_conf *ensure_mdev(int minor, int create) 431{ 432 struct drbd_conf *mdev; 433 434 if (minor >= minor_count) 435 return NULL; 436 437 mdev = minor_to_mdev(minor); 438 439 if (!mdev && create) { 440 struct gendisk *disk = NULL; 441 mdev = drbd_new_device(minor); 442 443 spin_lock_irq(&drbd_pp_lock); 444 if (minor_table[minor] == NULL) { 445 minor_table[minor] = mdev; 446 disk = mdev->vdisk; 447 mdev = NULL; 448 } /* else: we lost the race */ 449 spin_unlock_irq(&drbd_pp_lock); 450 451 if (disk) /* we won the race above */ 452 /* in case we ever add a drbd_delete_device(), 453 * don't forget the del_gendisk! */ 454 add_disk(disk); 455 else /* we lost the race above */ 456 drbd_free_mdev(mdev); 457 458 mdev = minor_to_mdev(minor); 459 } 460 461 return mdev; 462} 463 464static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 465 struct drbd_nl_cfg_reply *reply) 466{ 467 struct primary primary_args; 468 469 memset(&primary_args, 0, sizeof(struct primary)); 470 if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) { 471 reply->ret_code = ERR_MANDATORY_TAG; 472 return 0; 473 } 474 475 reply->ret_code = 476 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force); 477 478 return 0; 479} 480 481static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 482 struct drbd_nl_cfg_reply *reply) 483{ 484 reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0); 485 486 return 0; 487} 488 489/* initializes the md.*_offset members, so we are able to find 490 * the on disk meta data */ 491static void drbd_md_set_sector_offsets(struct drbd_conf *mdev, 492 struct drbd_backing_dev *bdev) 493{ 494 sector_t md_size_sect = 0; 495 switch (bdev->dc.meta_dev_idx) { 496 default: 497 /* v07 style fixed size indexed meta data */ 498 bdev->md.md_size_sect = MD_RESERVED_SECT; 499 bdev->md.md_offset = drbd_md_ss__(mdev, bdev); 500 bdev->md.al_offset = MD_AL_OFFSET; 501 bdev->md.bm_offset = MD_BM_OFFSET; 502 break; 503 case DRBD_MD_INDEX_FLEX_EXT: 504 /* just occupy the full device; unit: sectors */ 505 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev); 506 bdev->md.md_offset = 0; 507 bdev->md.al_offset = MD_AL_OFFSET; 508 bdev->md.bm_offset = MD_BM_OFFSET; 509 break; 510 case DRBD_MD_INDEX_INTERNAL: 511 case DRBD_MD_INDEX_FLEX_INT: 512 bdev->md.md_offset = drbd_md_ss__(mdev, bdev); 513 /* al size is still fixed */ 514 bdev->md.al_offset = -MD_AL_MAX_SIZE; 515 /* we need (slightly less than) ~ this much bitmap sectors: */ 516 md_size_sect = drbd_get_capacity(bdev->backing_bdev); 517 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT); 518 md_size_sect = BM_SECT_TO_EXT(md_size_sect); 519 md_size_sect = ALIGN(md_size_sect, 8); 520 521 /* plus the "drbd meta data super block", 522 * and the activity log; */ 523 md_size_sect += MD_BM_OFFSET; 524 525 bdev->md.md_size_sect = md_size_sect; 526 /* bitmap offset is adjusted by 'super' block size */ 527 bdev->md.bm_offset = -md_size_sect + MD_AL_OFFSET; 528 break; 529 } 530} 531 532char *ppsize(char *buf, unsigned long long size) 533{ 534 /* Needs 9 bytes at max. */ 535 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' }; 536 int base = 0; 537 while (size >= 10000) { 538 /* shift + round */ 539 size = (size >> 10) + !!(size & (1<<9)); 540 base++; 541 } 542 sprintf(buf, "%lu %cB", (long)size, units[base]); 543 544 return buf; 545} 546 547/* there is still a theoretical deadlock when called from receiver 548 * on an D_INCONSISTENT R_PRIMARY: 549 * remote READ does inc_ap_bio, receiver would need to receive answer 550 * packet from remote to dec_ap_bio again. 551 * receiver receive_sizes(), comes here, 552 * waits for ap_bio_cnt == 0. -> deadlock. 553 * but this cannot happen, actually, because: 554 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable 555 * (not connected, or bad/no disk on peer): 556 * see drbd_fail_request_early, ap_bio_cnt is zero. 557 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET: 558 * peer may not initiate a resize. 559 */ 560void drbd_suspend_io(struct drbd_conf *mdev) 561{ 562 set_bit(SUSPEND_IO, &mdev->flags); 563 if (is_susp(mdev->state)) 564 return; 565 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt)); 566} 567 568void drbd_resume_io(struct drbd_conf *mdev) 569{ 570 clear_bit(SUSPEND_IO, &mdev->flags); 571 wake_up(&mdev->misc_wait); 572} 573 574/** 575 * drbd_determine_dev_size() - Sets the right device size obeying all constraints 576 * @mdev: DRBD device. 577 * 578 * Returns 0 on success, negative return values indicate errors. 579 * You should call drbd_md_sync() after calling this function. 580 */ 581enum determine_dev_size drbd_determin_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local) 582{ 583 sector_t prev_first_sect, prev_size; /* previous meta location */ 584 sector_t la_size; 585 sector_t size; 586 char ppb[10]; 587 588 int md_moved, la_size_changed; 589 enum determine_dev_size rv = unchanged; 590 591 /* race: 592 * application request passes inc_ap_bio, 593 * but then cannot get an AL-reference. 594 * this function later may wait on ap_bio_cnt == 0. -> deadlock. 595 * 596 * to avoid that: 597 * Suspend IO right here. 598 * still lock the act_log to not trigger ASSERTs there. 599 */ 600 drbd_suspend_io(mdev); 601 602 /* no wait necessary anymore, actually we could assert that */ 603 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log)); 604 605 prev_first_sect = drbd_md_first_sector(mdev->ldev); 606 prev_size = mdev->ldev->md.md_size_sect; 607 la_size = mdev->ldev->md.la_size_sect; 608 609 /* TODO: should only be some assert here, not (re)init... */ 610 drbd_md_set_sector_offsets(mdev, mdev->ldev); 611 612 size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED); 613 614 if (drbd_get_capacity(mdev->this_bdev) != size || 615 drbd_bm_capacity(mdev) != size) { 616 int err; 617 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC)); 618 if (unlikely(err)) { 619 /* currently there is only one error: ENOMEM! */ 620 size = drbd_bm_capacity(mdev)>>1; 621 if (size == 0) { 622 dev_err(DEV, "OUT OF MEMORY! " 623 "Could not allocate bitmap!\n"); 624 } else { 625 dev_err(DEV, "BM resizing failed. " 626 "Leaving size unchanged at size = %lu KB\n", 627 (unsigned long)size); 628 } 629 rv = dev_size_error; 630 } 631 /* racy, see comments above. */ 632 drbd_set_my_capacity(mdev, size); 633 mdev->ldev->md.la_size_sect = size; 634 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1), 635 (unsigned long long)size>>1); 636 } 637 if (rv == dev_size_error) 638 goto out; 639 640 la_size_changed = (la_size != mdev->ldev->md.la_size_sect); 641 642 md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev) 643 || prev_size != mdev->ldev->md.md_size_sect; 644 645 if (la_size_changed || md_moved) { 646 drbd_al_shrink(mdev); /* All extents inactive. */ 647 dev_info(DEV, "Writing the whole bitmap, %s\n", 648 la_size_changed && md_moved ? "size changed and md moved" : 649 la_size_changed ? "size changed" : "md moved"); 650 rv = drbd_bitmap_io(mdev, &drbd_bm_write, "size changed"); /* does drbd_resume_io() ! */ 651 drbd_md_mark_dirty(mdev); 652 } 653 654 if (size > la_size) 655 rv = grew; 656 if (size < la_size) 657 rv = shrunk; 658out: 659 lc_unlock(mdev->act_log); 660 wake_up(&mdev->al_wait); 661 drbd_resume_io(mdev); 662 663 return rv; 664} 665 666sector_t 667drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space) 668{ 669 sector_t p_size = mdev->p_size; /* partner's disk size. */ 670 sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */ 671 sector_t m_size; /* my size */ 672 sector_t u_size = bdev->dc.disk_size; /* size requested by user. */ 673 sector_t size = 0; 674 675 m_size = drbd_get_max_capacity(bdev); 676 677 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) { 678 dev_warn(DEV, "Resize while not connected was forced by the user!\n"); 679 p_size = m_size; 680 } 681 682 if (p_size && m_size) { 683 size = min_t(sector_t, p_size, m_size); 684 } else { 685 if (la_size) { 686 size = la_size; 687 if (m_size && m_size < size) 688 size = m_size; 689 if (p_size && p_size < size) 690 size = p_size; 691 } else { 692 if (m_size) 693 size = m_size; 694 if (p_size) 695 size = p_size; 696 } 697 } 698 699 if (size == 0) 700 dev_err(DEV, "Both nodes diskless!\n"); 701 702 if (u_size) { 703 if (u_size > size) 704 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n", 705 (unsigned long)u_size>>1, (unsigned long)size>>1); 706 else 707 size = u_size; 708 } 709 710 return size; 711} 712 713/** 714 * drbd_check_al_size() - Ensures that the AL is of the right size 715 * @mdev: DRBD device. 716 * 717 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation 718 * failed, and 0 on success. You should call drbd_md_sync() after you called 719 * this function. 720 */ 721static int drbd_check_al_size(struct drbd_conf *mdev) 722{ 723 struct lru_cache *n, *t; 724 struct lc_element *e; 725 unsigned int in_use; 726 int i; 727 728 ERR_IF(mdev->sync_conf.al_extents < 7) 729 mdev->sync_conf.al_extents = 127; 730 731 if (mdev->act_log && 732 mdev->act_log->nr_elements == mdev->sync_conf.al_extents) 733 return 0; 734 735 in_use = 0; 736 t = mdev->act_log; 737 n = lc_create("act_log", drbd_al_ext_cache, 738 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0); 739 740 if (n == NULL) { 741 dev_err(DEV, "Cannot allocate act_log lru!\n"); 742 return -ENOMEM; 743 } 744 spin_lock_irq(&mdev->al_lock); 745 if (t) { 746 for (i = 0; i < t->nr_elements; i++) { 747 e = lc_element_by_index(t, i); 748 if (e->refcnt) 749 dev_err(DEV, "refcnt(%d)==%d\n", 750 e->lc_number, e->refcnt); 751 in_use += e->refcnt; 752 } 753 } 754 if (!in_use) 755 mdev->act_log = n; 756 spin_unlock_irq(&mdev->al_lock); 757 if (in_use) { 758 dev_err(DEV, "Activity log still in use!\n"); 759 lc_destroy(n); 760 return -EBUSY; 761 } else { 762 if (t) 763 lc_destroy(t); 764 } 765 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */ 766 return 0; 767} 768 769void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size) __must_hold(local) 770{ 771 struct request_queue * const q = mdev->rq_queue; 772 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue; 773 int max_segments = mdev->ldev->dc.max_bio_bvecs; 774 int max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9); 775 776 blk_queue_logical_block_size(q, 512); 777 blk_queue_max_hw_sectors(q, max_hw_sectors); 778 /* This is the workaround for "bio would need to, but cannot, be split" */ 779 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS); 780 blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1); 781 blk_queue_stack_limits(q, b); 782 783 dev_info(DEV, "max BIO size = %u\n", queue_max_hw_sectors(q) << 9); 784 785 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) { 786 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n", 787 q->backing_dev_info.ra_pages, 788 b->backing_dev_info.ra_pages); 789 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages; 790 } 791} 792 793/* serialize deconfig (worker exiting, doing cleanup) 794 * and reconfig (drbdsetup disk, drbdsetup net) 795 * 796 * Wait for a potentially exiting worker, then restart it, 797 * or start a new one. Flush any pending work, there may still be an 798 * after_state_change queued. 799 */ 800static void drbd_reconfig_start(struct drbd_conf *mdev) 801{ 802 wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags)); 803 wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags)); 804 drbd_thread_start(&mdev->worker); 805 drbd_flush_workqueue(mdev); 806} 807 808/* if still unconfigured, stops worker again. 809 * if configured now, clears CONFIG_PENDING. 810 * wakes potential waiters */ 811static void drbd_reconfig_done(struct drbd_conf *mdev) 812{ 813 spin_lock_irq(&mdev->req_lock); 814 if (mdev->state.disk == D_DISKLESS && 815 mdev->state.conn == C_STANDALONE && 816 mdev->state.role == R_SECONDARY) { 817 set_bit(DEVICE_DYING, &mdev->flags); 818 drbd_thread_stop_nowait(&mdev->worker); 819 } else 820 clear_bit(CONFIG_PENDING, &mdev->flags); 821 spin_unlock_irq(&mdev->req_lock); 822 wake_up(&mdev->state_wait); 823} 824 825/* Make sure IO is suspended before calling this function(). */ 826static void drbd_suspend_al(struct drbd_conf *mdev) 827{ 828 int s = 0; 829 830 if (lc_try_lock(mdev->act_log)) { 831 drbd_al_shrink(mdev); 832 lc_unlock(mdev->act_log); 833 } else { 834 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n"); 835 return; 836 } 837 838 spin_lock_irq(&mdev->req_lock); 839 if (mdev->state.conn < C_CONNECTED) 840 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags); 841 842 spin_unlock_irq(&mdev->req_lock); 843 844 if (s) 845 dev_info(DEV, "Suspended AL updates\n"); 846} 847 848/* does always return 0; 849 * interesting return code is in reply->ret_code */ 850static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 851 struct drbd_nl_cfg_reply *reply) 852{ 853 enum drbd_ret_code retcode; 854 enum determine_dev_size dd; 855 sector_t max_possible_sectors; 856 sector_t min_md_device_sectors; 857 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */ 858 struct block_device *bdev; 859 struct lru_cache *resync_lru = NULL; 860 union drbd_state ns, os; 861 unsigned int max_bio_size; 862 int rv; 863 int cp_discovered = 0; 864 int logical_block_size; 865 866 drbd_reconfig_start(mdev); 867 868 /* if you want to reconfigure, please tear down first */ 869 if (mdev->state.disk > D_DISKLESS) { 870 retcode = ERR_DISK_CONFIGURED; 871 goto fail; 872 } 873 /* It may just now have detached because of IO error. Make sure 874 * drbd_ldev_destroy is done already, we may end up here very fast, 875 * e.g. if someone calls attach from the on-io-error handler, 876 * to realize a "hot spare" feature (not that I'd recommend that) */ 877 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt)); 878 879 /* allocation not in the IO path, cqueue thread context */ 880 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL); 881 if (!nbc) { 882 retcode = ERR_NOMEM; 883 goto fail; 884 } 885 886 nbc->dc.disk_size = DRBD_DISK_SIZE_SECT_DEF; 887 nbc->dc.on_io_error = DRBD_ON_IO_ERROR_DEF; 888 nbc->dc.fencing = DRBD_FENCING_DEF; 889 nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF; 890 891 if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) { 892 retcode = ERR_MANDATORY_TAG; 893 goto fail; 894 } 895 896 if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) { 897 retcode = ERR_MD_IDX_INVALID; 898 goto fail; 899 } 900 901 if (get_net_conf(mdev)) { 902 int prot = mdev->net_conf->wire_protocol; 903 put_net_conf(mdev); 904 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) { 905 retcode = ERR_STONITH_AND_PROT_A; 906 goto fail; 907 } 908 } 909 910 bdev = blkdev_get_by_path(nbc->dc.backing_dev, 911 FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev); 912 if (IS_ERR(bdev)) { 913 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev, 914 PTR_ERR(bdev)); 915 retcode = ERR_OPEN_DISK; 916 goto fail; 917 } 918 nbc->backing_bdev = bdev; 919 920 /* 921 * meta_dev_idx >= 0: external fixed size, possibly multiple 922 * drbd sharing one meta device. TODO in that case, paranoia 923 * check that [md_bdev, meta_dev_idx] is not yet used by some 924 * other drbd minor! (if you use drbd.conf + drbdadm, that 925 * should check it for you already; but if you don't, or 926 * someone fooled it, we need to double check here) 927 */ 928 bdev = blkdev_get_by_path(nbc->dc.meta_dev, 929 FMODE_READ | FMODE_WRITE | FMODE_EXCL, 930 (nbc->dc.meta_dev_idx < 0) ? 931 (void *)mdev : (void *)drbd_m_holder); 932 if (IS_ERR(bdev)) { 933 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev, 934 PTR_ERR(bdev)); 935 retcode = ERR_OPEN_MD_DISK; 936 goto fail; 937 } 938 nbc->md_bdev = bdev; 939 940 if ((nbc->backing_bdev == nbc->md_bdev) != 941 (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL || 942 nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) { 943 retcode = ERR_MD_IDX_INVALID; 944 goto fail; 945 } 946 947 resync_lru = lc_create("resync", drbd_bm_ext_cache, 948 61, sizeof(struct bm_extent), 949 offsetof(struct bm_extent, lce)); 950 if (!resync_lru) { 951 retcode = ERR_NOMEM; 952 goto fail; 953 } 954 955 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */ 956 drbd_md_set_sector_offsets(mdev, nbc); 957 958 if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) { 959 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n", 960 (unsigned long long) drbd_get_max_capacity(nbc), 961 (unsigned long long) nbc->dc.disk_size); 962 retcode = ERR_DISK_TO_SMALL; 963 goto fail; 964 } 965 966 if (nbc->dc.meta_dev_idx < 0) { 967 max_possible_sectors = DRBD_MAX_SECTORS_FLEX; 968 /* at least one MB, otherwise it does not make sense */ 969 min_md_device_sectors = (2<<10); 970 } else { 971 max_possible_sectors = DRBD_MAX_SECTORS; 972 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1); 973 } 974 975 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) { 976 retcode = ERR_MD_DISK_TO_SMALL; 977 dev_warn(DEV, "refusing attach: md-device too small, " 978 "at least %llu sectors needed for this meta-disk type\n", 979 (unsigned long long) min_md_device_sectors); 980 goto fail; 981 } 982 983 /* Make sure the new disk is big enough 984 * (we may currently be R_PRIMARY with no local disk...) */ 985 if (drbd_get_max_capacity(nbc) < 986 drbd_get_capacity(mdev->this_bdev)) { 987 retcode = ERR_DISK_TO_SMALL; 988 goto fail; 989 } 990 991 nbc->known_size = drbd_get_capacity(nbc->backing_bdev); 992 993 if (nbc->known_size > max_possible_sectors) { 994 dev_warn(DEV, "==> truncating very big lower level device " 995 "to currently maximum possible %llu sectors <==\n", 996 (unsigned long long) max_possible_sectors); 997 if (nbc->dc.meta_dev_idx >= 0) 998 dev_warn(DEV, "==>> using internal or flexible " 999 "meta data may help <<==\n"); 1000 } 1001 1002 drbd_suspend_io(mdev); 1003 /* also wait for the last barrier ack. */ 1004 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state)); 1005 /* and for any other previously queued work */ 1006 drbd_flush_workqueue(mdev); 1007 1008 retcode = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE); 1009 drbd_resume_io(mdev); 1010 if (retcode < SS_SUCCESS) 1011 goto fail; 1012 1013 if (!get_ldev_if_state(mdev, D_ATTACHING)) 1014 goto force_diskless; 1015 1016 drbd_md_set_sector_offsets(mdev, nbc); 1017 1018 /* allocate a second IO page if logical_block_size != 512 */ 1019 logical_block_size = bdev_logical_block_size(nbc->md_bdev); 1020 if (logical_block_size == 0) 1021 logical_block_size = MD_SECTOR_SIZE; 1022 1023 if (logical_block_size != MD_SECTOR_SIZE) { 1024 if (!mdev->md_io_tmpp) { 1025 struct page *page = alloc_page(GFP_NOIO); 1026 if (!page) 1027 goto force_diskless_dec; 1028 1029 dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n", 1030 logical_block_size, MD_SECTOR_SIZE); 1031 dev_warn(DEV, "Workaround engaged (has performance impact).\n"); 1032 1033 mdev->md_io_tmpp = page; 1034 } 1035 } 1036 1037 if (!mdev->bitmap) { 1038 if (drbd_bm_init(mdev)) { 1039 retcode = ERR_NOMEM; 1040 goto force_diskless_dec; 1041 } 1042 } 1043 1044 retcode = drbd_md_read(mdev, nbc); 1045 if (retcode != NO_ERROR) 1046 goto force_diskless_dec; 1047 1048 if (mdev->state.conn < C_CONNECTED && 1049 mdev->state.role == R_PRIMARY && 1050 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) { 1051 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n", 1052 (unsigned long long)mdev->ed_uuid); 1053 retcode = ERR_DATA_NOT_CURRENT; 1054 goto force_diskless_dec; 1055 } 1056 1057 /* Since we are diskless, fix the activity log first... */ 1058 if (drbd_check_al_size(mdev)) { 1059 retcode = ERR_NOMEM; 1060 goto force_diskless_dec; 1061 } 1062 1063 /* Prevent shrinking of consistent devices ! */ 1064 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && 1065 drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) { 1066 dev_warn(DEV, "refusing to truncate a consistent device\n"); 1067 retcode = ERR_DISK_TO_SMALL; 1068 goto force_diskless_dec; 1069 } 1070 1071 if (!drbd_al_read_log(mdev, nbc)) { 1072 retcode = ERR_IO_MD_DISK; 1073 goto force_diskless_dec; 1074 } 1075 1076 /* Reset the "barriers don't work" bits here, then force meta data to 1077 * be written, to ensure we determine if barriers are supported. */ 1078 if (nbc->dc.no_md_flush) 1079 set_bit(MD_NO_FUA, &mdev->flags); 1080 else 1081 clear_bit(MD_NO_FUA, &mdev->flags); 1082 1083 /* Point of no return reached. 1084 * Devices and memory are no longer released by error cleanup below. 1085 * now mdev takes over responsibility, and the state engine should 1086 * clean it up somewhere. */ 1087 D_ASSERT(mdev->ldev == NULL); 1088 mdev->ldev = nbc; 1089 mdev->resync = resync_lru; 1090 nbc = NULL; 1091 resync_lru = NULL; 1092 1093 mdev->write_ordering = WO_bdev_flush; 1094 drbd_bump_write_ordering(mdev, WO_bdev_flush); 1095 1096 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY)) 1097 set_bit(CRASHED_PRIMARY, &mdev->flags); 1098 else 1099 clear_bit(CRASHED_PRIMARY, &mdev->flags); 1100 1101 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) && 1102 !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) { 1103 set_bit(CRASHED_PRIMARY, &mdev->flags); 1104 cp_discovered = 1; 1105 } 1106 1107 mdev->send_cnt = 0; 1108 mdev->recv_cnt = 0; 1109 mdev->read_cnt = 0; 1110 mdev->writ_cnt = 0; 1111 1112 max_bio_size = DRBD_MAX_BIO_SIZE; 1113 if (mdev->state.conn == C_CONNECTED) { 1114 /* We are Primary, Connected, and now attach a new local 1115 * backing store. We must not increase the user visible maximum 1116 * bio size on this device to something the peer may not be 1117 * able to handle. */ 1118 if (mdev->agreed_pro_version < 94) 1119 max_bio_size = queue_max_hw_sectors(mdev->rq_queue) << 9; 1120 else if (mdev->agreed_pro_version == 94) 1121 max_bio_size = DRBD_MAX_SIZE_H80_PACKET; 1122 /* else: drbd 8.3.9 and later, stay with default */ 1123 } 1124 1125 drbd_setup_queue_param(mdev, max_bio_size); 1126 1127 /* If I am currently not R_PRIMARY, 1128 * but meta data primary indicator is set, 1129 * I just now recover from a hard crash, 1130 * and have been R_PRIMARY before that crash. 1131 * 1132 * Now, if I had no connection before that crash 1133 * (have been degraded R_PRIMARY), chances are that 1134 * I won't find my peer now either. 1135 * 1136 * In that case, and _only_ in that case, 1137 * we use the degr-wfc-timeout instead of the default, 1138 * so we can automatically recover from a crash of a 1139 * degraded but active "cluster" after a certain timeout. 1140 */ 1141 clear_bit(USE_DEGR_WFC_T, &mdev->flags); 1142 if (mdev->state.role != R_PRIMARY && 1143 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) && 1144 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND)) 1145 set_bit(USE_DEGR_WFC_T, &mdev->flags); 1146 1147 dd = drbd_determin_dev_size(mdev, 0); 1148 if (dd == dev_size_error) { 1149 retcode = ERR_NOMEM_BITMAP; 1150 goto force_diskless_dec; 1151 } else if (dd == grew) 1152 set_bit(RESYNC_AFTER_NEG, &mdev->flags); 1153 1154 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) { 1155 dev_info(DEV, "Assuming that all blocks are out of sync " 1156 "(aka FullSync)\n"); 1157 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) { 1158 retcode = ERR_IO_MD_DISK; 1159 goto force_diskless_dec; 1160 } 1161 } else { 1162 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) { 1163 retcode = ERR_IO_MD_DISK; 1164 goto force_diskless_dec; 1165 } 1166 } 1167 1168 if (cp_discovered) { 1169 drbd_al_apply_to_bm(mdev); 1170 drbd_al_to_on_disk_bm(mdev); 1171 } 1172 1173 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev)) 1174 drbd_suspend_al(mdev); /* IO is still suspended here... */ 1175 1176 spin_lock_irq(&mdev->req_lock); 1177 os = mdev->state; 1178 ns.i = os.i; 1179 /* If MDF_CONSISTENT is not set go into inconsistent state, 1180 otherwise investigate MDF_WasUpToDate... 1181 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state, 1182 otherwise into D_CONSISTENT state. 1183 */ 1184 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) { 1185 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE)) 1186 ns.disk = D_CONSISTENT; 1187 else 1188 ns.disk = D_OUTDATED; 1189 } else { 1190 ns.disk = D_INCONSISTENT; 1191 } 1192 1193 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED)) 1194 ns.pdsk = D_OUTDATED; 1195 1196 if ( ns.disk == D_CONSISTENT && 1197 (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE)) 1198 ns.disk = D_UP_TO_DATE; 1199 1200 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND, 1201 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before 1202 this point, because drbd_request_state() modifies these 1203 flags. */ 1204 1205 /* In case we are C_CONNECTED postpone any decision on the new disk 1206 state after the negotiation phase. */ 1207 if (mdev->state.conn == C_CONNECTED) { 1208 mdev->new_state_tmp.i = ns.i; 1209 ns.i = os.i; 1210 ns.disk = D_NEGOTIATING; 1211 1212 /* We expect to receive up-to-date UUIDs soon. 1213 To avoid a race in receive_state, free p_uuid while 1214 holding req_lock. I.e. atomic with the state change */ 1215 kfree(mdev->p_uuid); 1216 mdev->p_uuid = NULL; 1217 } 1218 1219 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); 1220 ns = mdev->state; 1221 spin_unlock_irq(&mdev->req_lock); 1222 1223 if (rv < SS_SUCCESS) 1224 goto force_diskless_dec; 1225 1226 if (mdev->state.role == R_PRIMARY) 1227 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1; 1228 else 1229 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1; 1230 1231 drbd_md_mark_dirty(mdev); 1232 drbd_md_sync(mdev); 1233 1234 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); 1235 put_ldev(mdev); 1236 reply->ret_code = retcode; 1237 drbd_reconfig_done(mdev); 1238 return 0; 1239 1240 force_diskless_dec: 1241 put_ldev(mdev); 1242 force_diskless: 1243 drbd_force_state(mdev, NS(disk, D_FAILED)); 1244 drbd_md_sync(mdev); 1245 fail: 1246 if (nbc) { 1247 if (nbc->backing_bdev) 1248 blkdev_put(nbc->backing_bdev, 1249 FMODE_READ | FMODE_WRITE | FMODE_EXCL); 1250 if (nbc->md_bdev) 1251 blkdev_put(nbc->md_bdev, 1252 FMODE_READ | FMODE_WRITE | FMODE_EXCL); 1253 kfree(nbc); 1254 } 1255 lc_destroy(resync_lru); 1256 1257 reply->ret_code = retcode; 1258 drbd_reconfig_done(mdev); 1259 return 0; 1260} 1261 1262/* Detaching the disk is a process in multiple stages. First we need to lock 1263 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io. 1264 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all 1265 * internal references as well. 1266 * Only then we have finally detached. */ 1267static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1268 struct drbd_nl_cfg_reply *reply) 1269{ 1270 drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */ 1271 reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS)); 1272 if (mdev->state.disk == D_DISKLESS) 1273 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt)); 1274 drbd_resume_io(mdev); 1275 return 0; 1276} 1277 1278static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1279 struct drbd_nl_cfg_reply *reply) 1280{ 1281 int i, ns; 1282 enum drbd_ret_code retcode; 1283 struct net_conf *new_conf = NULL; 1284 struct crypto_hash *tfm = NULL; 1285 struct crypto_hash *integrity_w_tfm = NULL; 1286 struct crypto_hash *integrity_r_tfm = NULL; 1287 struct hlist_head *new_tl_hash = NULL; 1288 struct hlist_head *new_ee_hash = NULL; 1289 struct drbd_conf *odev; 1290 char hmac_name[CRYPTO_MAX_ALG_NAME]; 1291 void *int_dig_out = NULL; 1292 void *int_dig_in = NULL; 1293 void *int_dig_vv = NULL; 1294 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr; 1295 1296 drbd_reconfig_start(mdev); 1297 1298 if (mdev->state.conn > C_STANDALONE) { 1299 retcode = ERR_NET_CONFIGURED; 1300 goto fail; 1301 } 1302 1303 /* allocation not in the IO path, cqueue thread context */ 1304 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); 1305 if (!new_conf) { 1306 retcode = ERR_NOMEM; 1307 goto fail; 1308 } 1309 1310 new_conf->timeout = DRBD_TIMEOUT_DEF; 1311 new_conf->try_connect_int = DRBD_CONNECT_INT_DEF; 1312 new_conf->ping_int = DRBD_PING_INT_DEF; 1313 new_conf->max_epoch_size = DRBD_MAX_EPOCH_SIZE_DEF; 1314 new_conf->max_buffers = DRBD_MAX_BUFFERS_DEF; 1315 new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF; 1316 new_conf->sndbuf_size = DRBD_SNDBUF_SIZE_DEF; 1317 new_conf->rcvbuf_size = DRBD_RCVBUF_SIZE_DEF; 1318 new_conf->ko_count = DRBD_KO_COUNT_DEF; 1319 new_conf->after_sb_0p = DRBD_AFTER_SB_0P_DEF; 1320 new_conf->after_sb_1p = DRBD_AFTER_SB_1P_DEF; 1321 new_conf->after_sb_2p = DRBD_AFTER_SB_2P_DEF; 1322 new_conf->want_lose = 0; 1323 new_conf->two_primaries = 0; 1324 new_conf->wire_protocol = DRBD_PROT_C; 1325 new_conf->ping_timeo = DRBD_PING_TIMEO_DEF; 1326 new_conf->rr_conflict = DRBD_RR_CONFLICT_DEF; 1327 new_conf->on_congestion = DRBD_ON_CONGESTION_DEF; 1328 new_conf->cong_extents = DRBD_CONG_EXTENTS_DEF; 1329 1330 if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) { 1331 retcode = ERR_MANDATORY_TAG; 1332 goto fail; 1333 } 1334 1335 if (new_conf->two_primaries 1336 && (new_conf->wire_protocol != DRBD_PROT_C)) { 1337 retcode = ERR_NOT_PROTO_C; 1338 goto fail; 1339 } 1340 1341 if (get_ldev(mdev)) { 1342 enum drbd_fencing_p fp = mdev->ldev->dc.fencing; 1343 put_ldev(mdev); 1344 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) { 1345 retcode = ERR_STONITH_AND_PROT_A; 1346 goto fail; 1347 } 1348 } 1349 1350 if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) { 1351 retcode = ERR_CONG_NOT_PROTO_A; 1352 goto fail; 1353 } 1354 1355 if (mdev->state.role == R_PRIMARY && new_conf->want_lose) { 1356 retcode = ERR_DISCARD; 1357 goto fail; 1358 } 1359 1360 retcode = NO_ERROR; 1361 1362 new_my_addr = (struct sockaddr *)&new_conf->my_addr; 1363 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr; 1364 for (i = 0; i < minor_count; i++) { 1365 odev = minor_to_mdev(i); 1366 if (!odev || odev == mdev) 1367 continue; 1368 if (get_net_conf(odev)) { 1369 taken_addr = (struct sockaddr *)&odev->net_conf->my_addr; 1370 if (new_conf->my_addr_len == odev->net_conf->my_addr_len && 1371 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len)) 1372 retcode = ERR_LOCAL_ADDR; 1373 1374 taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr; 1375 if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len && 1376 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len)) 1377 retcode = ERR_PEER_ADDR; 1378 1379 put_net_conf(odev); 1380 if (retcode != NO_ERROR) 1381 goto fail; 1382 } 1383 } 1384 1385 if (new_conf->cram_hmac_alg[0] != 0) { 1386 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)", 1387 new_conf->cram_hmac_alg); 1388 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC); 1389 if (IS_ERR(tfm)) { 1390 tfm = NULL; 1391 retcode = ERR_AUTH_ALG; 1392 goto fail; 1393 } 1394 1395 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { 1396 retcode = ERR_AUTH_ALG_ND; 1397 goto fail; 1398 } 1399 } 1400 1401 if (new_conf->integrity_alg[0]) { 1402 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC); 1403 if (IS_ERR(integrity_w_tfm)) { 1404 integrity_w_tfm = NULL; 1405 retcode=ERR_INTEGRITY_ALG; 1406 goto fail; 1407 } 1408 1409 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) { 1410 retcode=ERR_INTEGRITY_ALG_ND; 1411 goto fail; 1412 } 1413 1414 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC); 1415 if (IS_ERR(integrity_r_tfm)) { 1416 integrity_r_tfm = NULL; 1417 retcode=ERR_INTEGRITY_ALG; 1418 goto fail; 1419 } 1420 } 1421 1422 ns = new_conf->max_epoch_size/8; 1423 if (mdev->tl_hash_s != ns) { 1424 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL); 1425 if (!new_tl_hash) { 1426 retcode = ERR_NOMEM; 1427 goto fail; 1428 } 1429 } 1430 1431 ns = new_conf->max_buffers/8; 1432 if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) { 1433 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL); 1434 if (!new_ee_hash) { 1435 retcode = ERR_NOMEM; 1436 goto fail; 1437 } 1438 } 1439 1440 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0; 1441 1442 if (integrity_w_tfm) { 1443 i = crypto_hash_digestsize(integrity_w_tfm); 1444 int_dig_out = kmalloc(i, GFP_KERNEL); 1445 if (!int_dig_out) { 1446 retcode = ERR_NOMEM; 1447 goto fail; 1448 } 1449 int_dig_in = kmalloc(i, GFP_KERNEL); 1450 if (!int_dig_in) { 1451 retcode = ERR_NOMEM; 1452 goto fail; 1453 } 1454 int_dig_vv = kmalloc(i, GFP_KERNEL); 1455 if (!int_dig_vv) { 1456 retcode = ERR_NOMEM; 1457 goto fail; 1458 } 1459 } 1460 1461 if (!mdev->bitmap) { 1462 if(drbd_bm_init(mdev)) { 1463 retcode = ERR_NOMEM; 1464 goto fail; 1465 } 1466 } 1467 1468 drbd_flush_workqueue(mdev); 1469 spin_lock_irq(&mdev->req_lock); 1470 if (mdev->net_conf != NULL) { 1471 retcode = ERR_NET_CONFIGURED; 1472 spin_unlock_irq(&mdev->req_lock); 1473 goto fail; 1474 } 1475 mdev->net_conf = new_conf; 1476 1477 mdev->send_cnt = 0; 1478 mdev->recv_cnt = 0; 1479 1480 if (new_tl_hash) { 1481 kfree(mdev->tl_hash); 1482 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8; 1483 mdev->tl_hash = new_tl_hash; 1484 } 1485 1486 if (new_ee_hash) { 1487 kfree(mdev->ee_hash); 1488 mdev->ee_hash_s = mdev->net_conf->max_buffers/8; 1489 mdev->ee_hash = new_ee_hash; 1490 } 1491 1492 crypto_free_hash(mdev->cram_hmac_tfm); 1493 mdev->cram_hmac_tfm = tfm; 1494 1495 crypto_free_hash(mdev->integrity_w_tfm); 1496 mdev->integrity_w_tfm = integrity_w_tfm; 1497 1498 crypto_free_hash(mdev->integrity_r_tfm); 1499 mdev->integrity_r_tfm = integrity_r_tfm; 1500 1501 kfree(mdev->int_dig_out); 1502 kfree(mdev->int_dig_in); 1503 kfree(mdev->int_dig_vv); 1504 mdev->int_dig_out=int_dig_out; 1505 mdev->int_dig_in=int_dig_in; 1506 mdev->int_dig_vv=int_dig_vv; 1507 retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL); 1508 spin_unlock_irq(&mdev->req_lock); 1509 1510 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); 1511 reply->ret_code = retcode; 1512 drbd_reconfig_done(mdev); 1513 return 0; 1514 1515fail: 1516 kfree(int_dig_out); 1517 kfree(int_dig_in); 1518 kfree(int_dig_vv); 1519 crypto_free_hash(tfm); 1520 crypto_free_hash(integrity_w_tfm); 1521 crypto_free_hash(integrity_r_tfm); 1522 kfree(new_tl_hash); 1523 kfree(new_ee_hash); 1524 kfree(new_conf); 1525 1526 reply->ret_code = retcode; 1527 drbd_reconfig_done(mdev); 1528 return 0; 1529} 1530 1531static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1532 struct drbd_nl_cfg_reply *reply) 1533{ 1534 int retcode; 1535 struct disconnect dc; 1536 1537 memset(&dc, 0, sizeof(struct disconnect)); 1538 if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) { 1539 retcode = ERR_MANDATORY_TAG; 1540 goto fail; 1541 } 1542 1543 if (dc.force) { 1544 spin_lock_irq(&mdev->req_lock); 1545 if (mdev->state.conn >= C_WF_CONNECTION) 1546 _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL); 1547 spin_unlock_irq(&mdev->req_lock); 1548 goto done; 1549 } 1550 1551 retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED); 1552 1553 if (retcode == SS_NOTHING_TO_DO) 1554 goto done; 1555 else if (retcode == SS_ALREADY_STANDALONE) 1556 goto done; 1557 else if (retcode == SS_PRIMARY_NOP) { 1558 /* Our statche checking code wants to see the peer outdated. */ 1559 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING, 1560 pdsk, D_OUTDATED)); 1561 } else if (retcode == SS_CW_FAILED_BY_PEER) { 1562 /* The peer probably wants to see us outdated. */ 1563 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING, 1564 disk, D_OUTDATED), 1565 CS_ORDERED); 1566 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) { 1567 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 1568 retcode = SS_SUCCESS; 1569 } 1570 } 1571 1572 if (retcode < SS_SUCCESS) 1573 goto fail; 1574 1575 if (wait_event_interruptible(mdev->state_wait, 1576 mdev->state.conn != C_DISCONNECTING)) { 1577 /* Do not test for mdev->state.conn == C_STANDALONE, since 1578 someone else might connect us in the mean time! */ 1579 retcode = ERR_INTR; 1580 goto fail; 1581 } 1582 1583 done: 1584 retcode = NO_ERROR; 1585 fail: 1586 drbd_md_sync(mdev); 1587 reply->ret_code = retcode; 1588 return 0; 1589} 1590 1591void resync_after_online_grow(struct drbd_conf *mdev) 1592{ 1593 int iass; /* I am sync source */ 1594 1595 dev_info(DEV, "Resync of new storage after online grow\n"); 1596 if (mdev->state.role != mdev->state.peer) 1597 iass = (mdev->state.role == R_PRIMARY); 1598 else 1599 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags); 1600 1601 if (iass) 1602 drbd_start_resync(mdev, C_SYNC_SOURCE); 1603 else 1604 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE); 1605} 1606 1607static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1608 struct drbd_nl_cfg_reply *reply) 1609{ 1610 struct resize rs; 1611 int retcode = NO_ERROR; 1612 enum determine_dev_size dd; 1613 enum dds_flags ddsf; 1614 1615 memset(&rs, 0, sizeof(struct resize)); 1616 if (!resize_from_tags(mdev, nlp->tag_list, &rs)) { 1617 retcode = ERR_MANDATORY_TAG; 1618 goto fail; 1619 } 1620 1621 if (mdev->state.conn > C_CONNECTED) { 1622 retcode = ERR_RESIZE_RESYNC; 1623 goto fail; 1624 } 1625 1626 if (mdev->state.role == R_SECONDARY && 1627 mdev->state.peer == R_SECONDARY) { 1628 retcode = ERR_NO_PRIMARY; 1629 goto fail; 1630 } 1631 1632 if (!get_ldev(mdev)) { 1633 retcode = ERR_NO_DISK; 1634 goto fail; 1635 } 1636 1637 if (rs.no_resync && mdev->agreed_pro_version < 93) { 1638 retcode = ERR_NEED_APV_93; 1639 goto fail; 1640 } 1641 1642 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) 1643 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); 1644 1645 mdev->ldev->dc.disk_size = (sector_t)rs.resize_size; 1646 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0); 1647 dd = drbd_determin_dev_size(mdev, ddsf); 1648 drbd_md_sync(mdev); 1649 put_ldev(mdev); 1650 if (dd == dev_size_error) { 1651 retcode = ERR_NOMEM_BITMAP; 1652 goto fail; 1653 } 1654 1655 if (mdev->state.conn == C_CONNECTED) { 1656 if (dd == grew) 1657 set_bit(RESIZE_PENDING, &mdev->flags); 1658 1659 drbd_send_uuids(mdev); 1660 drbd_send_sizes(mdev, 1, ddsf); 1661 } 1662 1663 fail: 1664 reply->ret_code = retcode; 1665 return 0; 1666} 1667 1668static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1669 struct drbd_nl_cfg_reply *reply) 1670{ 1671 int retcode = NO_ERROR; 1672 int err; 1673 int ovr; /* online verify running */ 1674 int rsr; /* re-sync running */ 1675 struct crypto_hash *verify_tfm = NULL; 1676 struct crypto_hash *csums_tfm = NULL; 1677 struct syncer_conf sc; 1678 cpumask_var_t new_cpu_mask; 1679 int *rs_plan_s = NULL; 1680 int fifo_size; 1681 1682 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) { 1683 retcode = ERR_NOMEM; 1684 goto fail; 1685 } 1686 1687 if (nlp->flags & DRBD_NL_SET_DEFAULTS) { 1688 memset(&sc, 0, sizeof(struct syncer_conf)); 1689 sc.rate = DRBD_RATE_DEF; 1690 sc.after = DRBD_AFTER_DEF; 1691 sc.al_extents = DRBD_AL_EXTENTS_DEF; 1692 sc.on_no_data = DRBD_ON_NO_DATA_DEF; 1693 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF; 1694 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF; 1695 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF; 1696 sc.c_max_rate = DRBD_C_MAX_RATE_DEF; 1697 sc.c_min_rate = DRBD_C_MIN_RATE_DEF; 1698 } else 1699 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf)); 1700 1701 if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) { 1702 retcode = ERR_MANDATORY_TAG; 1703 goto fail; 1704 } 1705 1706 /* re-sync running */ 1707 rsr = ( mdev->state.conn == C_SYNC_SOURCE || 1708 mdev->state.conn == C_SYNC_TARGET || 1709 mdev->state.conn == C_PAUSED_SYNC_S || 1710 mdev->state.conn == C_PAUSED_SYNC_T ); 1711 1712 if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) { 1713 retcode = ERR_CSUMS_RESYNC_RUNNING; 1714 goto fail; 1715 } 1716 1717 if (!rsr && sc.csums_alg[0]) { 1718 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC); 1719 if (IS_ERR(csums_tfm)) { 1720 csums_tfm = NULL; 1721 retcode = ERR_CSUMS_ALG; 1722 goto fail; 1723 } 1724 1725 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) { 1726 retcode = ERR_CSUMS_ALG_ND; 1727 goto fail; 1728 } 1729 } 1730 1731 /* online verify running */ 1732 ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T); 1733 1734 if (ovr) { 1735 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) { 1736 retcode = ERR_VERIFY_RUNNING; 1737 goto fail; 1738 } 1739 } 1740 1741 if (!ovr && sc.verify_alg[0]) { 1742 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC); 1743 if (IS_ERR(verify_tfm)) { 1744 verify_tfm = NULL; 1745 retcode = ERR_VERIFY_ALG; 1746 goto fail; 1747 } 1748 1749 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) { 1750 retcode = ERR_VERIFY_ALG_ND; 1751 goto fail; 1752 } 1753 } 1754 1755 /* silently ignore cpu mask on UP kernel */ 1756 if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) { 1757 err = __bitmap_parse(sc.cpu_mask, 32, 0, 1758 cpumask_bits(new_cpu_mask), nr_cpu_ids); 1759 if (err) { 1760 dev_warn(DEV, "__bitmap_parse() failed with %d\n", err); 1761 retcode = ERR_CPU_MASK_PARSE; 1762 goto fail; 1763 } 1764 } 1765 1766 ERR_IF (sc.rate < 1) sc.rate = 1; 1767 ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */ 1768#define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT) 1769 if (sc.al_extents > AL_MAX) { 1770 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX); 1771 sc.al_extents = AL_MAX; 1772 } 1773#undef AL_MAX 1774 1775 /* to avoid spurious errors when configuring minors before configuring 1776 * the minors they depend on: if necessary, first create the minor we 1777 * depend on */ 1778 if (sc.after >= 0) 1779 ensure_mdev(sc.after, 1); 1780 1781 /* most sanity checks done, try to assign the new sync-after 1782 * dependency. need to hold the global lock in there, 1783 * to avoid a race in the dependency loop check. */ 1784 retcode = drbd_alter_sa(mdev, sc.after); 1785 if (retcode != NO_ERROR) 1786 goto fail; 1787 1788 fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ; 1789 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { 1790 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); 1791 if (!rs_plan_s) { 1792 dev_err(DEV, "kmalloc of fifo_buffer failed"); 1793 retcode = ERR_NOMEM; 1794 goto fail; 1795 } 1796 } 1797 1798 /* ok, assign the rest of it as well. 1799 * lock against receive_SyncParam() */ 1800 spin_lock(&mdev->peer_seq_lock); 1801 mdev->sync_conf = sc; 1802 1803 if (!rsr) { 1804 crypto_free_hash(mdev->csums_tfm); 1805 mdev->csums_tfm = csums_tfm; 1806 csums_tfm = NULL; 1807 } 1808 1809 if (!ovr) { 1810 crypto_free_hash(mdev->verify_tfm); 1811 mdev->verify_tfm = verify_tfm; 1812 verify_tfm = NULL; 1813 } 1814 1815 if (fifo_size != mdev->rs_plan_s.size) { 1816 kfree(mdev->rs_plan_s.values); 1817 mdev->rs_plan_s.values = rs_plan_s; 1818 mdev->rs_plan_s.size = fifo_size; 1819 mdev->rs_planed = 0; 1820 rs_plan_s = NULL; 1821 } 1822 1823 spin_unlock(&mdev->peer_seq_lock); 1824 1825 if (get_ldev(mdev)) { 1826 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log)); 1827 drbd_al_shrink(mdev); 1828 err = drbd_check_al_size(mdev); 1829 lc_unlock(mdev->act_log); 1830 wake_up(&mdev->al_wait); 1831 1832 put_ldev(mdev); 1833 drbd_md_sync(mdev); 1834 1835 if (err) { 1836 retcode = ERR_NOMEM; 1837 goto fail; 1838 } 1839 } 1840 1841 if (mdev->state.conn >= C_CONNECTED) 1842 drbd_send_sync_param(mdev, &sc); 1843 1844 if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) { 1845 cpumask_copy(mdev->cpu_mask, new_cpu_mask); 1846 drbd_calc_cpu_mask(mdev); 1847 mdev->receiver.reset_cpu_mask = 1; 1848 mdev->asender.reset_cpu_mask = 1; 1849 mdev->worker.reset_cpu_mask = 1; 1850 } 1851 1852 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); 1853fail: 1854 kfree(rs_plan_s); 1855 free_cpumask_var(new_cpu_mask); 1856 crypto_free_hash(csums_tfm); 1857 crypto_free_hash(verify_tfm); 1858 reply->ret_code = retcode; 1859 return 0; 1860} 1861 1862static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1863 struct drbd_nl_cfg_reply *reply) 1864{ 1865 int retcode; 1866 1867 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED); 1868 1869 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION) 1870 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T)); 1871 1872 while (retcode == SS_NEED_CONNECTION) { 1873 spin_lock_irq(&mdev->req_lock); 1874 if (mdev->state.conn < C_CONNECTED) 1875 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL); 1876 spin_unlock_irq(&mdev->req_lock); 1877 1878 if (retcode != SS_NEED_CONNECTION) 1879 break; 1880 1881 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T)); 1882 } 1883 1884 reply->ret_code = retcode; 1885 return 0; 1886} 1887 1888static int drbd_bmio_set_susp_al(struct drbd_conf *mdev) 1889{ 1890 int rv; 1891 1892 rv = drbd_bmio_set_n_write(mdev); 1893 drbd_suspend_al(mdev); 1894 return rv; 1895} 1896 1897static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1898 struct drbd_nl_cfg_reply *reply) 1899{ 1900 int retcode; 1901 1902 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED); 1903 1904 if (retcode < SS_SUCCESS) { 1905 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) { 1906 /* The peer will get a resync upon connect anyways. Just make that 1907 into a full resync. */ 1908 retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT)); 1909 if (retcode >= SS_SUCCESS) { 1910 /* open coded drbd_bitmap_io() */ 1911 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al, 1912 "set_n_write from invalidate_peer")) 1913 retcode = ERR_IO_MD_DISK; 1914 } 1915 } else 1916 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S)); 1917 } 1918 1919 reply->ret_code = retcode; 1920 return 0; 1921} 1922 1923static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1924 struct drbd_nl_cfg_reply *reply) 1925{ 1926 int retcode = NO_ERROR; 1927 1928 if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO) 1929 retcode = ERR_PAUSE_IS_SET; 1930 1931 reply->ret_code = retcode; 1932 return 0; 1933} 1934 1935static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1936 struct drbd_nl_cfg_reply *reply) 1937{ 1938 int retcode = NO_ERROR; 1939 1940 if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) 1941 retcode = ERR_PAUSE_IS_CLEAR; 1942 1943 reply->ret_code = retcode; 1944 return 0; 1945} 1946 1947static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1948 struct drbd_nl_cfg_reply *reply) 1949{ 1950 reply->ret_code = drbd_request_state(mdev, NS(susp, 1)); 1951 1952 return 0; 1953} 1954 1955static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1956 struct drbd_nl_cfg_reply *reply) 1957{ 1958 if (test_bit(NEW_CUR_UUID, &mdev->flags)) { 1959 drbd_uuid_new_current(mdev); 1960 clear_bit(NEW_CUR_UUID, &mdev->flags); 1961 } 1962 drbd_suspend_io(mdev); 1963 reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0)); 1964 if (reply->ret_code == SS_SUCCESS) { 1965 if (mdev->state.conn < C_CONNECTED) 1966 tl_clear(mdev); 1967 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED) 1968 tl_restart(mdev, fail_frozen_disk_io); 1969 } 1970 drbd_resume_io(mdev); 1971 1972 return 0; 1973} 1974 1975static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1976 struct drbd_nl_cfg_reply *reply) 1977{ 1978 reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED)); 1979 return 0; 1980} 1981 1982static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1983 struct drbd_nl_cfg_reply *reply) 1984{ 1985 unsigned short *tl; 1986 1987 tl = reply->tag_list; 1988 1989 if (get_ldev(mdev)) { 1990 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl); 1991 put_ldev(mdev); 1992 } 1993 1994 if (get_net_conf(mdev)) { 1995 tl = net_conf_to_tags(mdev, mdev->net_conf, tl); 1996 put_net_conf(mdev); 1997 } 1998 tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl); 1999 2000 put_unaligned(TT_END, tl++); /* Close the tag list */ 2001 2002 return (int)((char *)tl - (char *)reply->tag_list); 2003} 2004 2005static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2006 struct drbd_nl_cfg_reply *reply) 2007{ 2008 unsigned short *tl = reply->tag_list; 2009 union drbd_state s = mdev->state; 2010 unsigned long rs_left; 2011 unsigned int res; 2012 2013 tl = get_state_to_tags(mdev, (struct get_state *)&s, tl); 2014 2015 /* no local ref, no bitmap, no syncer progress. */ 2016 if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) { 2017 if (get_ldev(mdev)) { 2018 drbd_get_syncer_progress(mdev, &rs_left, &res); 2019 tl = tl_add_int(tl, T_sync_progress, &res); 2020 put_ldev(mdev); 2021 } 2022 } 2023 put_unaligned(TT_END, tl++); /* Close the tag list */ 2024 2025 return (int)((char *)tl - (char *)reply->tag_list); 2026} 2027 2028static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2029 struct drbd_nl_cfg_reply *reply) 2030{ 2031 unsigned short *tl; 2032 2033 tl = reply->tag_list; 2034 2035 if (get_ldev(mdev)) { 2036 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64)); 2037 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags); 2038 put_ldev(mdev); 2039 } 2040 put_unaligned(TT_END, tl++); /* Close the tag list */ 2041 2042 return (int)((char *)tl - (char *)reply->tag_list); 2043} 2044 2045/** 2046 * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use 2047 * @mdev: DRBD device. 2048 * @nlp: Netlink/connector packet from drbdsetup 2049 * @reply: Reply packet for drbdsetup 2050 */ 2051static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2052 struct drbd_nl_cfg_reply *reply) 2053{ 2054 unsigned short *tl; 2055 char rv; 2056 2057 tl = reply->tag_list; 2058 2059 rv = mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED : 2060 test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT; 2061 2062 tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv)); 2063 put_unaligned(TT_END, tl++); /* Close the tag list */ 2064 2065 return (int)((char *)tl - (char *)reply->tag_list); 2066} 2067 2068static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2069 struct drbd_nl_cfg_reply *reply) 2070{ 2071 /* default to resume from last known position, if possible */ 2072 struct start_ov args = 2073 { .start_sector = mdev->ov_start_sector }; 2074 2075 if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) { 2076 reply->ret_code = ERR_MANDATORY_TAG; 2077 return 0; 2078 } 2079 /* w_make_ov_request expects position to be aligned */ 2080 mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT; 2081 reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S)); 2082 return 0; 2083} 2084 2085 2086static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2087 struct drbd_nl_cfg_reply *reply) 2088{ 2089 int retcode = NO_ERROR; 2090 int skip_initial_sync = 0; 2091 int err; 2092 2093 struct new_c_uuid args; 2094 2095 memset(&args, 0, sizeof(struct new_c_uuid)); 2096 if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) { 2097 reply->ret_code = ERR_MANDATORY_TAG; 2098 return 0; 2099 } 2100 2101 mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */ 2102 2103 if (!get_ldev(mdev)) { 2104 retcode = ERR_NO_DISK; 2105 goto out; 2106 } 2107 2108 /* this is "skip initial sync", assume to be clean */ 2109 if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 && 2110 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) { 2111 dev_info(DEV, "Preparing to skip initial sync\n"); 2112 skip_initial_sync = 1; 2113 } else if (mdev->state.conn != C_STANDALONE) { 2114 retcode = ERR_CONNECTED; 2115 goto out_dec; 2116 } 2117 2118 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */ 2119 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */ 2120 2121 if (args.clear_bm) { 2122 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid"); 2123 if (err) { 2124 dev_err(DEV, "Writing bitmap failed with %d\n",err); 2125 retcode = ERR_IO_MD_DISK; 2126 } 2127 if (skip_initial_sync) { 2128 drbd_send_uuids_skip_initial_sync(mdev); 2129 _drbd_uuid_set(mdev, UI_BITMAP, 0); 2130 spin_lock_irq(&mdev->req_lock); 2131 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), 2132 CS_VERBOSE, NULL); 2133 spin_unlock_irq(&mdev->req_lock); 2134 } 2135 } 2136 2137 drbd_md_sync(mdev); 2138out_dec: 2139 put_ldev(mdev); 2140out: 2141 mutex_unlock(&mdev->state_mutex); 2142 2143 reply->ret_code = retcode; 2144 return 0; 2145} 2146 2147struct cn_handler_struct { 2148 int (*function)(struct drbd_conf *, 2149 struct drbd_nl_cfg_req *, 2150 struct drbd_nl_cfg_reply *); 2151 int reply_body_size; 2152}; 2153 2154static struct cn_handler_struct cnd_table[] = { 2155 [ P_primary ] = { &drbd_nl_primary, 0 }, 2156 [ P_secondary ] = { &drbd_nl_secondary, 0 }, 2157 [ P_disk_conf ] = { &drbd_nl_disk_conf, 0 }, 2158 [ P_detach ] = { &drbd_nl_detach, 0 }, 2159 [ P_net_conf ] = { &drbd_nl_net_conf, 0 }, 2160 [ P_disconnect ] = { &drbd_nl_disconnect, 0 }, 2161 [ P_resize ] = { &drbd_nl_resize, 0 }, 2162 [ P_syncer_conf ] = { &drbd_nl_syncer_conf, 0 }, 2163 [ P_invalidate ] = { &drbd_nl_invalidate, 0 }, 2164 [ P_invalidate_peer ] = { &drbd_nl_invalidate_peer, 0 }, 2165 [ P_pause_sync ] = { &drbd_nl_pause_sync, 0 }, 2166 [ P_resume_sync ] = { &drbd_nl_resume_sync, 0 }, 2167 [ P_suspend_io ] = { &drbd_nl_suspend_io, 0 }, 2168 [ P_resume_io ] = { &drbd_nl_resume_io, 0 }, 2169 [ P_outdate ] = { &drbd_nl_outdate, 0 }, 2170 [ P_get_config ] = { &drbd_nl_get_config, 2171 sizeof(struct syncer_conf_tag_len_struct) + 2172 sizeof(struct disk_conf_tag_len_struct) + 2173 sizeof(struct net_conf_tag_len_struct) }, 2174 [ P_get_state ] = { &drbd_nl_get_state, 2175 sizeof(struct get_state_tag_len_struct) + 2176 sizeof(struct sync_progress_tag_len_struct) }, 2177 [ P_get_uuids ] = { &drbd_nl_get_uuids, 2178 sizeof(struct get_uuids_tag_len_struct) }, 2179 [ P_get_timeout_flag ] = { &drbd_nl_get_timeout_flag, 2180 sizeof(struct get_timeout_flag_tag_len_struct)}, 2181 [ P_start_ov ] = { &drbd_nl_start_ov, 0 }, 2182 [ P_new_c_uuid ] = { &drbd_nl_new_c_uuid, 0 }, 2183}; 2184 2185static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp) 2186{ 2187 struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data; 2188 struct cn_handler_struct *cm; 2189 struct cn_msg *cn_reply; 2190 struct drbd_nl_cfg_reply *reply; 2191 struct drbd_conf *mdev; 2192 int retcode, rr; 2193 int reply_size = sizeof(struct cn_msg) 2194 + sizeof(struct drbd_nl_cfg_reply) 2195 + sizeof(short int); 2196 2197 if (!try_module_get(THIS_MODULE)) { 2198 printk(KERN_ERR "drbd: try_module_get() failed!\n"); 2199 return; 2200 } 2201 2202 if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) { 2203 retcode = ERR_PERM; 2204 goto fail; 2205 } 2206 2207 mdev = ensure_mdev(nlp->drbd_minor, 2208 (nlp->flags & DRBD_NL_CREATE_DEVICE)); 2209 if (!mdev) { 2210 retcode = ERR_MINOR_INVALID; 2211 goto fail; 2212 } 2213 2214 if (nlp->packet_type >= P_nl_after_last_packet || 2215 nlp->packet_type == P_return_code_only) { 2216 retcode = ERR_PACKET_NR; 2217 goto fail; 2218 } 2219 2220 cm = cnd_table + nlp->packet_type; 2221 2222 /* This may happen if packet number is 0: */ 2223 if (cm->function == NULL) { 2224 retcode = ERR_PACKET_NR; 2225 goto fail; 2226 } 2227 2228 reply_size += cm->reply_body_size; 2229 2230 /* allocation not in the IO path, cqueue thread context */ 2231 cn_reply = kzalloc(reply_size, GFP_KERNEL); 2232 if (!cn_reply) { 2233 retcode = ERR_NOMEM; 2234 goto fail; 2235 } 2236 reply = (struct drbd_nl_cfg_reply *) cn_reply->data; 2237 2238 reply->packet_type = 2239 cm->reply_body_size ? nlp->packet_type : P_return_code_only; 2240 reply->minor = nlp->drbd_minor; 2241 reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */ 2242 /* reply->tag_list; might be modified by cm->function. */ 2243 2244 rr = cm->function(mdev, nlp, reply); 2245 2246 cn_reply->id = req->id; 2247 cn_reply->seq = req->seq; 2248 cn_reply->ack = req->ack + 1; 2249 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr; 2250 cn_reply->flags = 0; 2251 2252 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL); 2253 if (rr && rr != -ESRCH) 2254 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr); 2255 2256 kfree(cn_reply); 2257 module_put(THIS_MODULE); 2258 return; 2259 fail: 2260 drbd_nl_send_reply(req, retcode); 2261 module_put(THIS_MODULE); 2262} 2263 2264static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */ 2265 2266static unsigned short * 2267__tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, 2268 unsigned short len, int nul_terminated) 2269{ 2270 unsigned short l = tag_descriptions[tag_number(tag)].max_len; 2271 len = (len < l) ? len : l; 2272 put_unaligned(tag, tl++); 2273 put_unaligned(len, tl++); 2274 memcpy(tl, data, len); 2275 tl = (unsigned short*)((char*)tl + len); 2276 if (nul_terminated) 2277 *((char*)tl - 1) = 0; 2278 return tl; 2279} 2280 2281static unsigned short * 2282tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len) 2283{ 2284 return __tl_add_blob(tl, tag, data, len, 0); 2285} 2286 2287static unsigned short * 2288tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str) 2289{ 2290 return __tl_add_blob(tl, tag, str, strlen(str)+1, 0); 2291} 2292 2293static unsigned short * 2294tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val) 2295{ 2296 put_unaligned(tag, tl++); 2297 switch(tag_type(tag)) { 2298 case TT_INTEGER: 2299 put_unaligned(sizeof(int), tl++); 2300 put_unaligned(*(int *)val, (int *)tl); 2301 tl = (unsigned short*)((char*)tl+sizeof(int)); 2302 break; 2303 case TT_INT64: 2304 put_unaligned(sizeof(u64), tl++); 2305 put_unaligned(*(u64 *)val, (u64 *)tl); 2306 tl = (unsigned short*)((char*)tl+sizeof(u64)); 2307 break; 2308 default: 2309 /* someone did something stupid. */ 2310 ; 2311 } 2312 return tl; 2313} 2314 2315void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state) 2316{ 2317 char buffer[sizeof(struct cn_msg)+ 2318 sizeof(struct drbd_nl_cfg_reply)+ 2319 sizeof(struct get_state_tag_len_struct)+ 2320 sizeof(short int)]; 2321 struct cn_msg *cn_reply = (struct cn_msg *) buffer; 2322 struct drbd_nl_cfg_reply *reply = 2323 (struct drbd_nl_cfg_reply *)cn_reply->data; 2324 unsigned short *tl = reply->tag_list; 2325 2326 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */ 2327 2328 tl = get_state_to_tags(mdev, (struct get_state *)&state, tl); 2329 2330 put_unaligned(TT_END, tl++); /* Close the tag list */ 2331 2332 cn_reply->id.idx = CN_IDX_DRBD; 2333 cn_reply->id.val = CN_VAL_DRBD; 2334 2335 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq); 2336 cn_reply->ack = 0; /* not used here. */ 2337 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + 2338 (int)((char *)tl - (char *)reply->tag_list); 2339 cn_reply->flags = 0; 2340 2341 reply->packet_type = P_get_state; 2342 reply->minor = mdev_to_minor(mdev); 2343 reply->ret_code = NO_ERROR; 2344 2345 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2346} 2347 2348void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name) 2349{ 2350 char buffer[sizeof(struct cn_msg)+ 2351 sizeof(struct drbd_nl_cfg_reply)+ 2352 sizeof(struct call_helper_tag_len_struct)+ 2353 sizeof(short int)]; 2354 struct cn_msg *cn_reply = (struct cn_msg *) buffer; 2355 struct drbd_nl_cfg_reply *reply = 2356 (struct drbd_nl_cfg_reply *)cn_reply->data; 2357 unsigned short *tl = reply->tag_list; 2358 2359 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */ 2360 2361 tl = tl_add_str(tl, T_helper, helper_name); 2362 put_unaligned(TT_END, tl++); /* Close the tag list */ 2363 2364 cn_reply->id.idx = CN_IDX_DRBD; 2365 cn_reply->id.val = CN_VAL_DRBD; 2366 2367 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq); 2368 cn_reply->ack = 0; /* not used here. */ 2369 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + 2370 (int)((char *)tl - (char *)reply->tag_list); 2371 cn_reply->flags = 0; 2372 2373 reply->packet_type = P_call_helper; 2374 reply->minor = mdev_to_minor(mdev); 2375 reply->ret_code = NO_ERROR; 2376 2377 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2378} 2379 2380void drbd_bcast_ee(struct drbd_conf *mdev, 2381 const char *reason, const int dgs, 2382 const char* seen_hash, const char* calc_hash, 2383 const struct drbd_epoch_entry* e) 2384{ 2385 struct cn_msg *cn_reply; 2386 struct drbd_nl_cfg_reply *reply; 2387 unsigned short *tl; 2388 struct page *page; 2389 unsigned len; 2390 2391 if (!e) 2392 return; 2393 if (!reason || !reason[0]) 2394 return; 2395 2396 /* apparently we have to memcpy twice, first to prepare the data for the 2397 * struct cn_msg, then within cn_netlink_send from the cn_msg to the 2398 * netlink skb. */ 2399 /* receiver thread context, which is not in the writeout path (of this node), 2400 * but may be in the writeout path of the _other_ node. 2401 * GFP_NOIO to avoid potential "distributed deadlock". */ 2402 cn_reply = kzalloc( 2403 sizeof(struct cn_msg)+ 2404 sizeof(struct drbd_nl_cfg_reply)+ 2405 sizeof(struct dump_ee_tag_len_struct)+ 2406 sizeof(short int), 2407 GFP_NOIO); 2408 2409 if (!cn_reply) { 2410 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n", 2411 (unsigned long long)e->sector, e->size); 2412 return; 2413 } 2414 2415 reply = (struct drbd_nl_cfg_reply*)cn_reply->data; 2416 tl = reply->tag_list; 2417 2418 tl = tl_add_str(tl, T_dump_ee_reason, reason); 2419 tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs); 2420 tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs); 2421 tl = tl_add_int(tl, T_ee_sector, &e->sector); 2422 tl = tl_add_int(tl, T_ee_block_id, &e->block_id); 2423 2424 /* dump the first 32k */ 2425 len = min_t(unsigned, e->size, 32 << 10); 2426 put_unaligned(T_ee_data, tl++); 2427 put_unaligned(len, tl++); 2428 2429 page = e->pages; 2430 page_chain_for_each(page) { 2431 void *d = kmap_atomic(page, KM_USER0); 2432 unsigned l = min_t(unsigned, len, PAGE_SIZE); 2433 memcpy(tl, d, l); 2434 kunmap_atomic(d, KM_USER0); 2435 tl = (unsigned short*)((char*)tl + l); 2436 len -= l; 2437 if (len == 0) 2438 break; 2439 } 2440 put_unaligned(TT_END, tl++); /* Close the tag list */ 2441 2442 cn_reply->id.idx = CN_IDX_DRBD; 2443 cn_reply->id.val = CN_VAL_DRBD; 2444 2445 cn_reply->seq = atomic_add_return(1,&drbd_nl_seq); 2446 cn_reply->ack = 0; // not used here. 2447 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + 2448 (int)((char*)tl - (char*)reply->tag_list); 2449 cn_reply->flags = 0; 2450 2451 reply->packet_type = P_dump_ee; 2452 reply->minor = mdev_to_minor(mdev); 2453 reply->ret_code = NO_ERROR; 2454 2455 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2456 kfree(cn_reply); 2457} 2458 2459void drbd_bcast_sync_progress(struct drbd_conf *mdev) 2460{ 2461 char buffer[sizeof(struct cn_msg)+ 2462 sizeof(struct drbd_nl_cfg_reply)+ 2463 sizeof(struct sync_progress_tag_len_struct)+ 2464 sizeof(short int)]; 2465 struct cn_msg *cn_reply = (struct cn_msg *) buffer; 2466 struct drbd_nl_cfg_reply *reply = 2467 (struct drbd_nl_cfg_reply *)cn_reply->data; 2468 unsigned short *tl = reply->tag_list; 2469 unsigned long rs_left; 2470 unsigned int res; 2471 2472 /* no local ref, no bitmap, no syncer progress, no broadcast. */ 2473 if (!get_ldev(mdev)) 2474 return; 2475 drbd_get_syncer_progress(mdev, &rs_left, &res); 2476 put_ldev(mdev); 2477 2478 tl = tl_add_int(tl, T_sync_progress, &res); 2479 put_unaligned(TT_END, tl++); /* Close the tag list */ 2480 2481 cn_reply->id.idx = CN_IDX_DRBD; 2482 cn_reply->id.val = CN_VAL_DRBD; 2483 2484 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq); 2485 cn_reply->ack = 0; /* not used here. */ 2486 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + 2487 (int)((char *)tl - (char *)reply->tag_list); 2488 cn_reply->flags = 0; 2489 2490 reply->packet_type = P_sync_progress; 2491 reply->minor = mdev_to_minor(mdev); 2492 reply->ret_code = NO_ERROR; 2493 2494 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2495} 2496 2497int __init drbd_nl_init(void) 2498{ 2499 static struct cb_id cn_id_drbd; 2500 int err, try=10; 2501 2502 cn_id_drbd.val = CN_VAL_DRBD; 2503 do { 2504 cn_id_drbd.idx = cn_idx; 2505 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback); 2506 if (!err) 2507 break; 2508 cn_idx = (cn_idx + CN_IDX_STEP); 2509 } while (try--); 2510 2511 if (err) { 2512 printk(KERN_ERR "drbd: cn_drbd failed to register\n"); 2513 return err; 2514 } 2515 2516 return 0; 2517} 2518 2519void drbd_nl_cleanup(void) 2520{ 2521 static struct cb_id cn_id_drbd; 2522 2523 cn_id_drbd.idx = cn_idx; 2524 cn_id_drbd.val = CN_VAL_DRBD; 2525 2526 cn_del_callback(&cn_id_drbd); 2527} 2528 2529void drbd_nl_send_reply(struct cn_msg *req, int ret_code) 2530{ 2531 char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)]; 2532 struct cn_msg *cn_reply = (struct cn_msg *) buffer; 2533 struct drbd_nl_cfg_reply *reply = 2534 (struct drbd_nl_cfg_reply *)cn_reply->data; 2535 int rr; 2536 2537 memset(buffer, 0, sizeof(buffer)); 2538 cn_reply->id = req->id; 2539 2540 cn_reply->seq = req->seq; 2541 cn_reply->ack = req->ack + 1; 2542 cn_reply->len = sizeof(struct drbd_nl_cfg_reply); 2543 cn_reply->flags = 0; 2544 2545 reply->packet_type = P_return_code_only; 2546 reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor; 2547 reply->ret_code = ret_code; 2548 2549 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2550 if (rr && rr != -ESRCH) 2551 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr); 2552} 2553 2554