drbd_nl.c revision f2024e7ce29f4287395ce879364cd68c7ac226f2
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 enum drbd_state_rv 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 rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE); 1009 retcode = rv; /* FIXME: Type mismatch. */ 1010 drbd_resume_io(mdev); 1011 if (rv < SS_SUCCESS) 1012 goto fail; 1013 1014 if (!get_ldev_if_state(mdev, D_ATTACHING)) 1015 goto force_diskless; 1016 1017 drbd_md_set_sector_offsets(mdev, nbc); 1018 1019 /* allocate a second IO page if logical_block_size != 512 */ 1020 logical_block_size = bdev_logical_block_size(nbc->md_bdev); 1021 if (logical_block_size == 0) 1022 logical_block_size = MD_SECTOR_SIZE; 1023 1024 if (logical_block_size != MD_SECTOR_SIZE) { 1025 if (!mdev->md_io_tmpp) { 1026 struct page *page = alloc_page(GFP_NOIO); 1027 if (!page) 1028 goto force_diskless_dec; 1029 1030 dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n", 1031 logical_block_size, MD_SECTOR_SIZE); 1032 dev_warn(DEV, "Workaround engaged (has performance impact).\n"); 1033 1034 mdev->md_io_tmpp = page; 1035 } 1036 } 1037 1038 if (!mdev->bitmap) { 1039 if (drbd_bm_init(mdev)) { 1040 retcode = ERR_NOMEM; 1041 goto force_diskless_dec; 1042 } 1043 } 1044 1045 retcode = drbd_md_read(mdev, nbc); 1046 if (retcode != NO_ERROR) 1047 goto force_diskless_dec; 1048 1049 if (mdev->state.conn < C_CONNECTED && 1050 mdev->state.role == R_PRIMARY && 1051 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) { 1052 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n", 1053 (unsigned long long)mdev->ed_uuid); 1054 retcode = ERR_DATA_NOT_CURRENT; 1055 goto force_diskless_dec; 1056 } 1057 1058 /* Since we are diskless, fix the activity log first... */ 1059 if (drbd_check_al_size(mdev)) { 1060 retcode = ERR_NOMEM; 1061 goto force_diskless_dec; 1062 } 1063 1064 /* Prevent shrinking of consistent devices ! */ 1065 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && 1066 drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) { 1067 dev_warn(DEV, "refusing to truncate a consistent device\n"); 1068 retcode = ERR_DISK_TO_SMALL; 1069 goto force_diskless_dec; 1070 } 1071 1072 if (!drbd_al_read_log(mdev, nbc)) { 1073 retcode = ERR_IO_MD_DISK; 1074 goto force_diskless_dec; 1075 } 1076 1077 /* Reset the "barriers don't work" bits here, then force meta data to 1078 * be written, to ensure we determine if barriers are supported. */ 1079 if (nbc->dc.no_md_flush) 1080 set_bit(MD_NO_FUA, &mdev->flags); 1081 else 1082 clear_bit(MD_NO_FUA, &mdev->flags); 1083 1084 /* Point of no return reached. 1085 * Devices and memory are no longer released by error cleanup below. 1086 * now mdev takes over responsibility, and the state engine should 1087 * clean it up somewhere. */ 1088 D_ASSERT(mdev->ldev == NULL); 1089 mdev->ldev = nbc; 1090 mdev->resync = resync_lru; 1091 nbc = NULL; 1092 resync_lru = NULL; 1093 1094 mdev->write_ordering = WO_bdev_flush; 1095 drbd_bump_write_ordering(mdev, WO_bdev_flush); 1096 1097 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY)) 1098 set_bit(CRASHED_PRIMARY, &mdev->flags); 1099 else 1100 clear_bit(CRASHED_PRIMARY, &mdev->flags); 1101 1102 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) && 1103 !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) { 1104 set_bit(CRASHED_PRIMARY, &mdev->flags); 1105 cp_discovered = 1; 1106 } 1107 1108 mdev->send_cnt = 0; 1109 mdev->recv_cnt = 0; 1110 mdev->read_cnt = 0; 1111 mdev->writ_cnt = 0; 1112 1113 max_bio_size = DRBD_MAX_BIO_SIZE; 1114 if (mdev->state.conn == C_CONNECTED) { 1115 /* We are Primary, Connected, and now attach a new local 1116 * backing store. We must not increase the user visible maximum 1117 * bio size on this device to something the peer may not be 1118 * able to handle. */ 1119 if (mdev->agreed_pro_version < 94) 1120 max_bio_size = queue_max_hw_sectors(mdev->rq_queue) << 9; 1121 else if (mdev->agreed_pro_version == 94) 1122 max_bio_size = DRBD_MAX_SIZE_H80_PACKET; 1123 /* else: drbd 8.3.9 and later, stay with default */ 1124 } 1125 1126 drbd_setup_queue_param(mdev, max_bio_size); 1127 1128 /* If I am currently not R_PRIMARY, 1129 * but meta data primary indicator is set, 1130 * I just now recover from a hard crash, 1131 * and have been R_PRIMARY before that crash. 1132 * 1133 * Now, if I had no connection before that crash 1134 * (have been degraded R_PRIMARY), chances are that 1135 * I won't find my peer now either. 1136 * 1137 * In that case, and _only_ in that case, 1138 * we use the degr-wfc-timeout instead of the default, 1139 * so we can automatically recover from a crash of a 1140 * degraded but active "cluster" after a certain timeout. 1141 */ 1142 clear_bit(USE_DEGR_WFC_T, &mdev->flags); 1143 if (mdev->state.role != R_PRIMARY && 1144 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) && 1145 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND)) 1146 set_bit(USE_DEGR_WFC_T, &mdev->flags); 1147 1148 dd = drbd_determin_dev_size(mdev, 0); 1149 if (dd == dev_size_error) { 1150 retcode = ERR_NOMEM_BITMAP; 1151 goto force_diskless_dec; 1152 } else if (dd == grew) 1153 set_bit(RESYNC_AFTER_NEG, &mdev->flags); 1154 1155 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) { 1156 dev_info(DEV, "Assuming that all blocks are out of sync " 1157 "(aka FullSync)\n"); 1158 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from attaching")) { 1159 retcode = ERR_IO_MD_DISK; 1160 goto force_diskless_dec; 1161 } 1162 } else { 1163 if (drbd_bitmap_io(mdev, &drbd_bm_read, "read from attaching") < 0) { 1164 retcode = ERR_IO_MD_DISK; 1165 goto force_diskless_dec; 1166 } 1167 } 1168 1169 if (cp_discovered) { 1170 drbd_al_apply_to_bm(mdev); 1171 drbd_al_to_on_disk_bm(mdev); 1172 } 1173 1174 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev)) 1175 drbd_suspend_al(mdev); /* IO is still suspended here... */ 1176 1177 spin_lock_irq(&mdev->req_lock); 1178 os = mdev->state; 1179 ns.i = os.i; 1180 /* If MDF_CONSISTENT is not set go into inconsistent state, 1181 otherwise investigate MDF_WasUpToDate... 1182 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state, 1183 otherwise into D_CONSISTENT state. 1184 */ 1185 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) { 1186 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE)) 1187 ns.disk = D_CONSISTENT; 1188 else 1189 ns.disk = D_OUTDATED; 1190 } else { 1191 ns.disk = D_INCONSISTENT; 1192 } 1193 1194 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED)) 1195 ns.pdsk = D_OUTDATED; 1196 1197 if ( ns.disk == D_CONSISTENT && 1198 (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE)) 1199 ns.disk = D_UP_TO_DATE; 1200 1201 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND, 1202 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before 1203 this point, because drbd_request_state() modifies these 1204 flags. */ 1205 1206 /* In case we are C_CONNECTED postpone any decision on the new disk 1207 state after the negotiation phase. */ 1208 if (mdev->state.conn == C_CONNECTED) { 1209 mdev->new_state_tmp.i = ns.i; 1210 ns.i = os.i; 1211 ns.disk = D_NEGOTIATING; 1212 1213 /* We expect to receive up-to-date UUIDs soon. 1214 To avoid a race in receive_state, free p_uuid while 1215 holding req_lock. I.e. atomic with the state change */ 1216 kfree(mdev->p_uuid); 1217 mdev->p_uuid = NULL; 1218 } 1219 1220 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); 1221 ns = mdev->state; 1222 spin_unlock_irq(&mdev->req_lock); 1223 1224 if (rv < SS_SUCCESS) 1225 goto force_diskless_dec; 1226 1227 if (mdev->state.role == R_PRIMARY) 1228 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1; 1229 else 1230 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1; 1231 1232 drbd_md_mark_dirty(mdev); 1233 drbd_md_sync(mdev); 1234 1235 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); 1236 put_ldev(mdev); 1237 reply->ret_code = retcode; 1238 drbd_reconfig_done(mdev); 1239 return 0; 1240 1241 force_diskless_dec: 1242 put_ldev(mdev); 1243 force_diskless: 1244 drbd_force_state(mdev, NS(disk, D_FAILED)); 1245 drbd_md_sync(mdev); 1246 fail: 1247 if (nbc) { 1248 if (nbc->backing_bdev) 1249 blkdev_put(nbc->backing_bdev, 1250 FMODE_READ | FMODE_WRITE | FMODE_EXCL); 1251 if (nbc->md_bdev) 1252 blkdev_put(nbc->md_bdev, 1253 FMODE_READ | FMODE_WRITE | FMODE_EXCL); 1254 kfree(nbc); 1255 } 1256 lc_destroy(resync_lru); 1257 1258 reply->ret_code = retcode; 1259 drbd_reconfig_done(mdev); 1260 return 0; 1261} 1262 1263/* Detaching the disk is a process in multiple stages. First we need to lock 1264 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io. 1265 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all 1266 * internal references as well. 1267 * Only then we have finally detached. */ 1268static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1269 struct drbd_nl_cfg_reply *reply) 1270{ 1271 drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */ 1272 reply->ret_code = drbd_request_state(mdev, NS(disk, D_DISKLESS)); 1273 if (mdev->state.disk == D_DISKLESS) 1274 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt)); 1275 drbd_resume_io(mdev); 1276 return 0; 1277} 1278 1279static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1280 struct drbd_nl_cfg_reply *reply) 1281{ 1282 int i, ns; 1283 enum drbd_ret_code retcode; 1284 struct net_conf *new_conf = NULL; 1285 struct crypto_hash *tfm = NULL; 1286 struct crypto_hash *integrity_w_tfm = NULL; 1287 struct crypto_hash *integrity_r_tfm = NULL; 1288 struct hlist_head *new_tl_hash = NULL; 1289 struct hlist_head *new_ee_hash = NULL; 1290 struct drbd_conf *odev; 1291 char hmac_name[CRYPTO_MAX_ALG_NAME]; 1292 void *int_dig_out = NULL; 1293 void *int_dig_in = NULL; 1294 void *int_dig_vv = NULL; 1295 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr; 1296 1297 drbd_reconfig_start(mdev); 1298 1299 if (mdev->state.conn > C_STANDALONE) { 1300 retcode = ERR_NET_CONFIGURED; 1301 goto fail; 1302 } 1303 1304 /* allocation not in the IO path, cqueue thread context */ 1305 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL); 1306 if (!new_conf) { 1307 retcode = ERR_NOMEM; 1308 goto fail; 1309 } 1310 1311 new_conf->timeout = DRBD_TIMEOUT_DEF; 1312 new_conf->try_connect_int = DRBD_CONNECT_INT_DEF; 1313 new_conf->ping_int = DRBD_PING_INT_DEF; 1314 new_conf->max_epoch_size = DRBD_MAX_EPOCH_SIZE_DEF; 1315 new_conf->max_buffers = DRBD_MAX_BUFFERS_DEF; 1316 new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF; 1317 new_conf->sndbuf_size = DRBD_SNDBUF_SIZE_DEF; 1318 new_conf->rcvbuf_size = DRBD_RCVBUF_SIZE_DEF; 1319 new_conf->ko_count = DRBD_KO_COUNT_DEF; 1320 new_conf->after_sb_0p = DRBD_AFTER_SB_0P_DEF; 1321 new_conf->after_sb_1p = DRBD_AFTER_SB_1P_DEF; 1322 new_conf->after_sb_2p = DRBD_AFTER_SB_2P_DEF; 1323 new_conf->want_lose = 0; 1324 new_conf->two_primaries = 0; 1325 new_conf->wire_protocol = DRBD_PROT_C; 1326 new_conf->ping_timeo = DRBD_PING_TIMEO_DEF; 1327 new_conf->rr_conflict = DRBD_RR_CONFLICT_DEF; 1328 new_conf->on_congestion = DRBD_ON_CONGESTION_DEF; 1329 new_conf->cong_extents = DRBD_CONG_EXTENTS_DEF; 1330 1331 if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) { 1332 retcode = ERR_MANDATORY_TAG; 1333 goto fail; 1334 } 1335 1336 if (new_conf->two_primaries 1337 && (new_conf->wire_protocol != DRBD_PROT_C)) { 1338 retcode = ERR_NOT_PROTO_C; 1339 goto fail; 1340 } 1341 1342 if (get_ldev(mdev)) { 1343 enum drbd_fencing_p fp = mdev->ldev->dc.fencing; 1344 put_ldev(mdev); 1345 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) { 1346 retcode = ERR_STONITH_AND_PROT_A; 1347 goto fail; 1348 } 1349 } 1350 1351 if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) { 1352 retcode = ERR_CONG_NOT_PROTO_A; 1353 goto fail; 1354 } 1355 1356 if (mdev->state.role == R_PRIMARY && new_conf->want_lose) { 1357 retcode = ERR_DISCARD; 1358 goto fail; 1359 } 1360 1361 retcode = NO_ERROR; 1362 1363 new_my_addr = (struct sockaddr *)&new_conf->my_addr; 1364 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr; 1365 for (i = 0; i < minor_count; i++) { 1366 odev = minor_to_mdev(i); 1367 if (!odev || odev == mdev) 1368 continue; 1369 if (get_net_conf(odev)) { 1370 taken_addr = (struct sockaddr *)&odev->net_conf->my_addr; 1371 if (new_conf->my_addr_len == odev->net_conf->my_addr_len && 1372 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len)) 1373 retcode = ERR_LOCAL_ADDR; 1374 1375 taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr; 1376 if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len && 1377 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len)) 1378 retcode = ERR_PEER_ADDR; 1379 1380 put_net_conf(odev); 1381 if (retcode != NO_ERROR) 1382 goto fail; 1383 } 1384 } 1385 1386 if (new_conf->cram_hmac_alg[0] != 0) { 1387 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)", 1388 new_conf->cram_hmac_alg); 1389 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC); 1390 if (IS_ERR(tfm)) { 1391 tfm = NULL; 1392 retcode = ERR_AUTH_ALG; 1393 goto fail; 1394 } 1395 1396 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { 1397 retcode = ERR_AUTH_ALG_ND; 1398 goto fail; 1399 } 1400 } 1401 1402 if (new_conf->integrity_alg[0]) { 1403 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC); 1404 if (IS_ERR(integrity_w_tfm)) { 1405 integrity_w_tfm = NULL; 1406 retcode=ERR_INTEGRITY_ALG; 1407 goto fail; 1408 } 1409 1410 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) { 1411 retcode=ERR_INTEGRITY_ALG_ND; 1412 goto fail; 1413 } 1414 1415 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC); 1416 if (IS_ERR(integrity_r_tfm)) { 1417 integrity_r_tfm = NULL; 1418 retcode=ERR_INTEGRITY_ALG; 1419 goto fail; 1420 } 1421 } 1422 1423 ns = new_conf->max_epoch_size/8; 1424 if (mdev->tl_hash_s != ns) { 1425 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL); 1426 if (!new_tl_hash) { 1427 retcode = ERR_NOMEM; 1428 goto fail; 1429 } 1430 } 1431 1432 ns = new_conf->max_buffers/8; 1433 if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) { 1434 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL); 1435 if (!new_ee_hash) { 1436 retcode = ERR_NOMEM; 1437 goto fail; 1438 } 1439 } 1440 1441 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0; 1442 1443 if (integrity_w_tfm) { 1444 i = crypto_hash_digestsize(integrity_w_tfm); 1445 int_dig_out = kmalloc(i, GFP_KERNEL); 1446 if (!int_dig_out) { 1447 retcode = ERR_NOMEM; 1448 goto fail; 1449 } 1450 int_dig_in = kmalloc(i, GFP_KERNEL); 1451 if (!int_dig_in) { 1452 retcode = ERR_NOMEM; 1453 goto fail; 1454 } 1455 int_dig_vv = kmalloc(i, GFP_KERNEL); 1456 if (!int_dig_vv) { 1457 retcode = ERR_NOMEM; 1458 goto fail; 1459 } 1460 } 1461 1462 if (!mdev->bitmap) { 1463 if(drbd_bm_init(mdev)) { 1464 retcode = ERR_NOMEM; 1465 goto fail; 1466 } 1467 } 1468 1469 drbd_flush_workqueue(mdev); 1470 spin_lock_irq(&mdev->req_lock); 1471 if (mdev->net_conf != NULL) { 1472 retcode = ERR_NET_CONFIGURED; 1473 spin_unlock_irq(&mdev->req_lock); 1474 goto fail; 1475 } 1476 mdev->net_conf = new_conf; 1477 1478 mdev->send_cnt = 0; 1479 mdev->recv_cnt = 0; 1480 1481 if (new_tl_hash) { 1482 kfree(mdev->tl_hash); 1483 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8; 1484 mdev->tl_hash = new_tl_hash; 1485 } 1486 1487 if (new_ee_hash) { 1488 kfree(mdev->ee_hash); 1489 mdev->ee_hash_s = mdev->net_conf->max_buffers/8; 1490 mdev->ee_hash = new_ee_hash; 1491 } 1492 1493 crypto_free_hash(mdev->cram_hmac_tfm); 1494 mdev->cram_hmac_tfm = tfm; 1495 1496 crypto_free_hash(mdev->integrity_w_tfm); 1497 mdev->integrity_w_tfm = integrity_w_tfm; 1498 1499 crypto_free_hash(mdev->integrity_r_tfm); 1500 mdev->integrity_r_tfm = integrity_r_tfm; 1501 1502 kfree(mdev->int_dig_out); 1503 kfree(mdev->int_dig_in); 1504 kfree(mdev->int_dig_vv); 1505 mdev->int_dig_out=int_dig_out; 1506 mdev->int_dig_in=int_dig_in; 1507 mdev->int_dig_vv=int_dig_vv; 1508 retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL); 1509 spin_unlock_irq(&mdev->req_lock); 1510 1511 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); 1512 reply->ret_code = retcode; 1513 drbd_reconfig_done(mdev); 1514 return 0; 1515 1516fail: 1517 kfree(int_dig_out); 1518 kfree(int_dig_in); 1519 kfree(int_dig_vv); 1520 crypto_free_hash(tfm); 1521 crypto_free_hash(integrity_w_tfm); 1522 crypto_free_hash(integrity_r_tfm); 1523 kfree(new_tl_hash); 1524 kfree(new_ee_hash); 1525 kfree(new_conf); 1526 1527 reply->ret_code = retcode; 1528 drbd_reconfig_done(mdev); 1529 return 0; 1530} 1531 1532static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1533 struct drbd_nl_cfg_reply *reply) 1534{ 1535 int retcode; 1536 struct disconnect dc; 1537 1538 memset(&dc, 0, sizeof(struct disconnect)); 1539 if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) { 1540 retcode = ERR_MANDATORY_TAG; 1541 goto fail; 1542 } 1543 1544 if (dc.force) { 1545 spin_lock_irq(&mdev->req_lock); 1546 if (mdev->state.conn >= C_WF_CONNECTION) 1547 _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL); 1548 spin_unlock_irq(&mdev->req_lock); 1549 goto done; 1550 } 1551 1552 retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED); 1553 1554 if (retcode == SS_NOTHING_TO_DO) 1555 goto done; 1556 else if (retcode == SS_ALREADY_STANDALONE) 1557 goto done; 1558 else if (retcode == SS_PRIMARY_NOP) { 1559 /* Our statche checking code wants to see the peer outdated. */ 1560 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING, 1561 pdsk, D_OUTDATED)); 1562 } else if (retcode == SS_CW_FAILED_BY_PEER) { 1563 /* The peer probably wants to see us outdated. */ 1564 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING, 1565 disk, D_OUTDATED), 1566 CS_ORDERED); 1567 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) { 1568 drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); 1569 retcode = SS_SUCCESS; 1570 } 1571 } 1572 1573 if (retcode < SS_SUCCESS) 1574 goto fail; 1575 1576 if (wait_event_interruptible(mdev->state_wait, 1577 mdev->state.conn != C_DISCONNECTING)) { 1578 /* Do not test for mdev->state.conn == C_STANDALONE, since 1579 someone else might connect us in the mean time! */ 1580 retcode = ERR_INTR; 1581 goto fail; 1582 } 1583 1584 done: 1585 retcode = NO_ERROR; 1586 fail: 1587 drbd_md_sync(mdev); 1588 reply->ret_code = retcode; 1589 return 0; 1590} 1591 1592void resync_after_online_grow(struct drbd_conf *mdev) 1593{ 1594 int iass; /* I am sync source */ 1595 1596 dev_info(DEV, "Resync of new storage after online grow\n"); 1597 if (mdev->state.role != mdev->state.peer) 1598 iass = (mdev->state.role == R_PRIMARY); 1599 else 1600 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags); 1601 1602 if (iass) 1603 drbd_start_resync(mdev, C_SYNC_SOURCE); 1604 else 1605 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE); 1606} 1607 1608static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1609 struct drbd_nl_cfg_reply *reply) 1610{ 1611 struct resize rs; 1612 int retcode = NO_ERROR; 1613 enum determine_dev_size dd; 1614 enum dds_flags ddsf; 1615 1616 memset(&rs, 0, sizeof(struct resize)); 1617 if (!resize_from_tags(mdev, nlp->tag_list, &rs)) { 1618 retcode = ERR_MANDATORY_TAG; 1619 goto fail; 1620 } 1621 1622 if (mdev->state.conn > C_CONNECTED) { 1623 retcode = ERR_RESIZE_RESYNC; 1624 goto fail; 1625 } 1626 1627 if (mdev->state.role == R_SECONDARY && 1628 mdev->state.peer == R_SECONDARY) { 1629 retcode = ERR_NO_PRIMARY; 1630 goto fail; 1631 } 1632 1633 if (!get_ldev(mdev)) { 1634 retcode = ERR_NO_DISK; 1635 goto fail; 1636 } 1637 1638 if (rs.no_resync && mdev->agreed_pro_version < 93) { 1639 retcode = ERR_NEED_APV_93; 1640 goto fail; 1641 } 1642 1643 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) 1644 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); 1645 1646 mdev->ldev->dc.disk_size = (sector_t)rs.resize_size; 1647 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0); 1648 dd = drbd_determin_dev_size(mdev, ddsf); 1649 drbd_md_sync(mdev); 1650 put_ldev(mdev); 1651 if (dd == dev_size_error) { 1652 retcode = ERR_NOMEM_BITMAP; 1653 goto fail; 1654 } 1655 1656 if (mdev->state.conn == C_CONNECTED) { 1657 if (dd == grew) 1658 set_bit(RESIZE_PENDING, &mdev->flags); 1659 1660 drbd_send_uuids(mdev); 1661 drbd_send_sizes(mdev, 1, ddsf); 1662 } 1663 1664 fail: 1665 reply->ret_code = retcode; 1666 return 0; 1667} 1668 1669static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1670 struct drbd_nl_cfg_reply *reply) 1671{ 1672 int retcode = NO_ERROR; 1673 int err; 1674 int ovr; /* online verify running */ 1675 int rsr; /* re-sync running */ 1676 struct crypto_hash *verify_tfm = NULL; 1677 struct crypto_hash *csums_tfm = NULL; 1678 struct syncer_conf sc; 1679 cpumask_var_t new_cpu_mask; 1680 int *rs_plan_s = NULL; 1681 int fifo_size; 1682 1683 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) { 1684 retcode = ERR_NOMEM; 1685 goto fail; 1686 } 1687 1688 if (nlp->flags & DRBD_NL_SET_DEFAULTS) { 1689 memset(&sc, 0, sizeof(struct syncer_conf)); 1690 sc.rate = DRBD_RATE_DEF; 1691 sc.after = DRBD_AFTER_DEF; 1692 sc.al_extents = DRBD_AL_EXTENTS_DEF; 1693 sc.on_no_data = DRBD_ON_NO_DATA_DEF; 1694 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF; 1695 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF; 1696 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF; 1697 sc.c_max_rate = DRBD_C_MAX_RATE_DEF; 1698 sc.c_min_rate = DRBD_C_MIN_RATE_DEF; 1699 } else 1700 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf)); 1701 1702 if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) { 1703 retcode = ERR_MANDATORY_TAG; 1704 goto fail; 1705 } 1706 1707 /* re-sync running */ 1708 rsr = ( mdev->state.conn == C_SYNC_SOURCE || 1709 mdev->state.conn == C_SYNC_TARGET || 1710 mdev->state.conn == C_PAUSED_SYNC_S || 1711 mdev->state.conn == C_PAUSED_SYNC_T ); 1712 1713 if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) { 1714 retcode = ERR_CSUMS_RESYNC_RUNNING; 1715 goto fail; 1716 } 1717 1718 if (!rsr && sc.csums_alg[0]) { 1719 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC); 1720 if (IS_ERR(csums_tfm)) { 1721 csums_tfm = NULL; 1722 retcode = ERR_CSUMS_ALG; 1723 goto fail; 1724 } 1725 1726 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) { 1727 retcode = ERR_CSUMS_ALG_ND; 1728 goto fail; 1729 } 1730 } 1731 1732 /* online verify running */ 1733 ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T); 1734 1735 if (ovr) { 1736 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) { 1737 retcode = ERR_VERIFY_RUNNING; 1738 goto fail; 1739 } 1740 } 1741 1742 if (!ovr && sc.verify_alg[0]) { 1743 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC); 1744 if (IS_ERR(verify_tfm)) { 1745 verify_tfm = NULL; 1746 retcode = ERR_VERIFY_ALG; 1747 goto fail; 1748 } 1749 1750 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) { 1751 retcode = ERR_VERIFY_ALG_ND; 1752 goto fail; 1753 } 1754 } 1755 1756 /* silently ignore cpu mask on UP kernel */ 1757 if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) { 1758 err = __bitmap_parse(sc.cpu_mask, 32, 0, 1759 cpumask_bits(new_cpu_mask), nr_cpu_ids); 1760 if (err) { 1761 dev_warn(DEV, "__bitmap_parse() failed with %d\n", err); 1762 retcode = ERR_CPU_MASK_PARSE; 1763 goto fail; 1764 } 1765 } 1766 1767 ERR_IF (sc.rate < 1) sc.rate = 1; 1768 ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */ 1769#define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT) 1770 if (sc.al_extents > AL_MAX) { 1771 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX); 1772 sc.al_extents = AL_MAX; 1773 } 1774#undef AL_MAX 1775 1776 /* to avoid spurious errors when configuring minors before configuring 1777 * the minors they depend on: if necessary, first create the minor we 1778 * depend on */ 1779 if (sc.after >= 0) 1780 ensure_mdev(sc.after, 1); 1781 1782 /* most sanity checks done, try to assign the new sync-after 1783 * dependency. need to hold the global lock in there, 1784 * to avoid a race in the dependency loop check. */ 1785 retcode = drbd_alter_sa(mdev, sc.after); 1786 if (retcode != NO_ERROR) 1787 goto fail; 1788 1789 fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ; 1790 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { 1791 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); 1792 if (!rs_plan_s) { 1793 dev_err(DEV, "kmalloc of fifo_buffer failed"); 1794 retcode = ERR_NOMEM; 1795 goto fail; 1796 } 1797 } 1798 1799 /* ok, assign the rest of it as well. 1800 * lock against receive_SyncParam() */ 1801 spin_lock(&mdev->peer_seq_lock); 1802 mdev->sync_conf = sc; 1803 1804 if (!rsr) { 1805 crypto_free_hash(mdev->csums_tfm); 1806 mdev->csums_tfm = csums_tfm; 1807 csums_tfm = NULL; 1808 } 1809 1810 if (!ovr) { 1811 crypto_free_hash(mdev->verify_tfm); 1812 mdev->verify_tfm = verify_tfm; 1813 verify_tfm = NULL; 1814 } 1815 1816 if (fifo_size != mdev->rs_plan_s.size) { 1817 kfree(mdev->rs_plan_s.values); 1818 mdev->rs_plan_s.values = rs_plan_s; 1819 mdev->rs_plan_s.size = fifo_size; 1820 mdev->rs_planed = 0; 1821 rs_plan_s = NULL; 1822 } 1823 1824 spin_unlock(&mdev->peer_seq_lock); 1825 1826 if (get_ldev(mdev)) { 1827 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log)); 1828 drbd_al_shrink(mdev); 1829 err = drbd_check_al_size(mdev); 1830 lc_unlock(mdev->act_log); 1831 wake_up(&mdev->al_wait); 1832 1833 put_ldev(mdev); 1834 drbd_md_sync(mdev); 1835 1836 if (err) { 1837 retcode = ERR_NOMEM; 1838 goto fail; 1839 } 1840 } 1841 1842 if (mdev->state.conn >= C_CONNECTED) 1843 drbd_send_sync_param(mdev, &sc); 1844 1845 if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) { 1846 cpumask_copy(mdev->cpu_mask, new_cpu_mask); 1847 drbd_calc_cpu_mask(mdev); 1848 mdev->receiver.reset_cpu_mask = 1; 1849 mdev->asender.reset_cpu_mask = 1; 1850 mdev->worker.reset_cpu_mask = 1; 1851 } 1852 1853 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE); 1854fail: 1855 kfree(rs_plan_s); 1856 free_cpumask_var(new_cpu_mask); 1857 crypto_free_hash(csums_tfm); 1858 crypto_free_hash(verify_tfm); 1859 reply->ret_code = retcode; 1860 return 0; 1861} 1862 1863static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1864 struct drbd_nl_cfg_reply *reply) 1865{ 1866 int retcode; 1867 1868 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED); 1869 1870 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION) 1871 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T)); 1872 1873 while (retcode == SS_NEED_CONNECTION) { 1874 spin_lock_irq(&mdev->req_lock); 1875 if (mdev->state.conn < C_CONNECTED) 1876 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL); 1877 spin_unlock_irq(&mdev->req_lock); 1878 1879 if (retcode != SS_NEED_CONNECTION) 1880 break; 1881 1882 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T)); 1883 } 1884 1885 reply->ret_code = retcode; 1886 return 0; 1887} 1888 1889static int drbd_bmio_set_susp_al(struct drbd_conf *mdev) 1890{ 1891 int rv; 1892 1893 rv = drbd_bmio_set_n_write(mdev); 1894 drbd_suspend_al(mdev); 1895 return rv; 1896} 1897 1898static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1899 struct drbd_nl_cfg_reply *reply) 1900{ 1901 int retcode; 1902 1903 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED); 1904 1905 if (retcode < SS_SUCCESS) { 1906 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) { 1907 /* The peer will get a resync upon connect anyways. Just make that 1908 into a full resync. */ 1909 retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT)); 1910 if (retcode >= SS_SUCCESS) { 1911 /* open coded drbd_bitmap_io() */ 1912 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al, 1913 "set_n_write from invalidate_peer")) 1914 retcode = ERR_IO_MD_DISK; 1915 } 1916 } else 1917 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S)); 1918 } 1919 1920 reply->ret_code = retcode; 1921 return 0; 1922} 1923 1924static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1925 struct drbd_nl_cfg_reply *reply) 1926{ 1927 int retcode = NO_ERROR; 1928 1929 if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO) 1930 retcode = ERR_PAUSE_IS_SET; 1931 1932 reply->ret_code = retcode; 1933 return 0; 1934} 1935 1936static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1937 struct drbd_nl_cfg_reply *reply) 1938{ 1939 int retcode = NO_ERROR; 1940 1941 if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) 1942 retcode = ERR_PAUSE_IS_CLEAR; 1943 1944 reply->ret_code = retcode; 1945 return 0; 1946} 1947 1948static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1949 struct drbd_nl_cfg_reply *reply) 1950{ 1951 reply->ret_code = drbd_request_state(mdev, NS(susp, 1)); 1952 1953 return 0; 1954} 1955 1956static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1957 struct drbd_nl_cfg_reply *reply) 1958{ 1959 if (test_bit(NEW_CUR_UUID, &mdev->flags)) { 1960 drbd_uuid_new_current(mdev); 1961 clear_bit(NEW_CUR_UUID, &mdev->flags); 1962 } 1963 drbd_suspend_io(mdev); 1964 reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0)); 1965 if (reply->ret_code == SS_SUCCESS) { 1966 if (mdev->state.conn < C_CONNECTED) 1967 tl_clear(mdev); 1968 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED) 1969 tl_restart(mdev, fail_frozen_disk_io); 1970 } 1971 drbd_resume_io(mdev); 1972 1973 return 0; 1974} 1975 1976static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1977 struct drbd_nl_cfg_reply *reply) 1978{ 1979 reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED)); 1980 return 0; 1981} 1982 1983static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 1984 struct drbd_nl_cfg_reply *reply) 1985{ 1986 unsigned short *tl; 1987 1988 tl = reply->tag_list; 1989 1990 if (get_ldev(mdev)) { 1991 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl); 1992 put_ldev(mdev); 1993 } 1994 1995 if (get_net_conf(mdev)) { 1996 tl = net_conf_to_tags(mdev, mdev->net_conf, tl); 1997 put_net_conf(mdev); 1998 } 1999 tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl); 2000 2001 put_unaligned(TT_END, tl++); /* Close the tag list */ 2002 2003 return (int)((char *)tl - (char *)reply->tag_list); 2004} 2005 2006static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2007 struct drbd_nl_cfg_reply *reply) 2008{ 2009 unsigned short *tl = reply->tag_list; 2010 union drbd_state s = mdev->state; 2011 unsigned long rs_left; 2012 unsigned int res; 2013 2014 tl = get_state_to_tags(mdev, (struct get_state *)&s, tl); 2015 2016 /* no local ref, no bitmap, no syncer progress. */ 2017 if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) { 2018 if (get_ldev(mdev)) { 2019 drbd_get_syncer_progress(mdev, &rs_left, &res); 2020 tl = tl_add_int(tl, T_sync_progress, &res); 2021 put_ldev(mdev); 2022 } 2023 } 2024 put_unaligned(TT_END, tl++); /* Close the tag list */ 2025 2026 return (int)((char *)tl - (char *)reply->tag_list); 2027} 2028 2029static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2030 struct drbd_nl_cfg_reply *reply) 2031{ 2032 unsigned short *tl; 2033 2034 tl = reply->tag_list; 2035 2036 if (get_ldev(mdev)) { 2037 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64)); 2038 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags); 2039 put_ldev(mdev); 2040 } 2041 put_unaligned(TT_END, tl++); /* Close the tag list */ 2042 2043 return (int)((char *)tl - (char *)reply->tag_list); 2044} 2045 2046/** 2047 * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use 2048 * @mdev: DRBD device. 2049 * @nlp: Netlink/connector packet from drbdsetup 2050 * @reply: Reply packet for drbdsetup 2051 */ 2052static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2053 struct drbd_nl_cfg_reply *reply) 2054{ 2055 unsigned short *tl; 2056 char rv; 2057 2058 tl = reply->tag_list; 2059 2060 rv = mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED : 2061 test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT; 2062 2063 tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv)); 2064 put_unaligned(TT_END, tl++); /* Close the tag list */ 2065 2066 return (int)((char *)tl - (char *)reply->tag_list); 2067} 2068 2069static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2070 struct drbd_nl_cfg_reply *reply) 2071{ 2072 /* default to resume from last known position, if possible */ 2073 struct start_ov args = 2074 { .start_sector = mdev->ov_start_sector }; 2075 2076 if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) { 2077 reply->ret_code = ERR_MANDATORY_TAG; 2078 return 0; 2079 } 2080 /* w_make_ov_request expects position to be aligned */ 2081 mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT; 2082 reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S)); 2083 return 0; 2084} 2085 2086 2087static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp, 2088 struct drbd_nl_cfg_reply *reply) 2089{ 2090 int retcode = NO_ERROR; 2091 int skip_initial_sync = 0; 2092 int err; 2093 2094 struct new_c_uuid args; 2095 2096 memset(&args, 0, sizeof(struct new_c_uuid)); 2097 if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) { 2098 reply->ret_code = ERR_MANDATORY_TAG; 2099 return 0; 2100 } 2101 2102 mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */ 2103 2104 if (!get_ldev(mdev)) { 2105 retcode = ERR_NO_DISK; 2106 goto out; 2107 } 2108 2109 /* this is "skip initial sync", assume to be clean */ 2110 if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 && 2111 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) { 2112 dev_info(DEV, "Preparing to skip initial sync\n"); 2113 skip_initial_sync = 1; 2114 } else if (mdev->state.conn != C_STANDALONE) { 2115 retcode = ERR_CONNECTED; 2116 goto out_dec; 2117 } 2118 2119 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */ 2120 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */ 2121 2122 if (args.clear_bm) { 2123 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, "clear_n_write from new_c_uuid"); 2124 if (err) { 2125 dev_err(DEV, "Writing bitmap failed with %d\n",err); 2126 retcode = ERR_IO_MD_DISK; 2127 } 2128 if (skip_initial_sync) { 2129 drbd_send_uuids_skip_initial_sync(mdev); 2130 _drbd_uuid_set(mdev, UI_BITMAP, 0); 2131 spin_lock_irq(&mdev->req_lock); 2132 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), 2133 CS_VERBOSE, NULL); 2134 spin_unlock_irq(&mdev->req_lock); 2135 } 2136 } 2137 2138 drbd_md_sync(mdev); 2139out_dec: 2140 put_ldev(mdev); 2141out: 2142 mutex_unlock(&mdev->state_mutex); 2143 2144 reply->ret_code = retcode; 2145 return 0; 2146} 2147 2148struct cn_handler_struct { 2149 int (*function)(struct drbd_conf *, 2150 struct drbd_nl_cfg_req *, 2151 struct drbd_nl_cfg_reply *); 2152 int reply_body_size; 2153}; 2154 2155static struct cn_handler_struct cnd_table[] = { 2156 [ P_primary ] = { &drbd_nl_primary, 0 }, 2157 [ P_secondary ] = { &drbd_nl_secondary, 0 }, 2158 [ P_disk_conf ] = { &drbd_nl_disk_conf, 0 }, 2159 [ P_detach ] = { &drbd_nl_detach, 0 }, 2160 [ P_net_conf ] = { &drbd_nl_net_conf, 0 }, 2161 [ P_disconnect ] = { &drbd_nl_disconnect, 0 }, 2162 [ P_resize ] = { &drbd_nl_resize, 0 }, 2163 [ P_syncer_conf ] = { &drbd_nl_syncer_conf, 0 }, 2164 [ P_invalidate ] = { &drbd_nl_invalidate, 0 }, 2165 [ P_invalidate_peer ] = { &drbd_nl_invalidate_peer, 0 }, 2166 [ P_pause_sync ] = { &drbd_nl_pause_sync, 0 }, 2167 [ P_resume_sync ] = { &drbd_nl_resume_sync, 0 }, 2168 [ P_suspend_io ] = { &drbd_nl_suspend_io, 0 }, 2169 [ P_resume_io ] = { &drbd_nl_resume_io, 0 }, 2170 [ P_outdate ] = { &drbd_nl_outdate, 0 }, 2171 [ P_get_config ] = { &drbd_nl_get_config, 2172 sizeof(struct syncer_conf_tag_len_struct) + 2173 sizeof(struct disk_conf_tag_len_struct) + 2174 sizeof(struct net_conf_tag_len_struct) }, 2175 [ P_get_state ] = { &drbd_nl_get_state, 2176 sizeof(struct get_state_tag_len_struct) + 2177 sizeof(struct sync_progress_tag_len_struct) }, 2178 [ P_get_uuids ] = { &drbd_nl_get_uuids, 2179 sizeof(struct get_uuids_tag_len_struct) }, 2180 [ P_get_timeout_flag ] = { &drbd_nl_get_timeout_flag, 2181 sizeof(struct get_timeout_flag_tag_len_struct)}, 2182 [ P_start_ov ] = { &drbd_nl_start_ov, 0 }, 2183 [ P_new_c_uuid ] = { &drbd_nl_new_c_uuid, 0 }, 2184}; 2185 2186static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp) 2187{ 2188 struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data; 2189 struct cn_handler_struct *cm; 2190 struct cn_msg *cn_reply; 2191 struct drbd_nl_cfg_reply *reply; 2192 struct drbd_conf *mdev; 2193 int retcode, rr; 2194 int reply_size = sizeof(struct cn_msg) 2195 + sizeof(struct drbd_nl_cfg_reply) 2196 + sizeof(short int); 2197 2198 if (!try_module_get(THIS_MODULE)) { 2199 printk(KERN_ERR "drbd: try_module_get() failed!\n"); 2200 return; 2201 } 2202 2203 if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) { 2204 retcode = ERR_PERM; 2205 goto fail; 2206 } 2207 2208 mdev = ensure_mdev(nlp->drbd_minor, 2209 (nlp->flags & DRBD_NL_CREATE_DEVICE)); 2210 if (!mdev) { 2211 retcode = ERR_MINOR_INVALID; 2212 goto fail; 2213 } 2214 2215 if (nlp->packet_type >= P_nl_after_last_packet || 2216 nlp->packet_type == P_return_code_only) { 2217 retcode = ERR_PACKET_NR; 2218 goto fail; 2219 } 2220 2221 cm = cnd_table + nlp->packet_type; 2222 2223 /* This may happen if packet number is 0: */ 2224 if (cm->function == NULL) { 2225 retcode = ERR_PACKET_NR; 2226 goto fail; 2227 } 2228 2229 reply_size += cm->reply_body_size; 2230 2231 /* allocation not in the IO path, cqueue thread context */ 2232 cn_reply = kzalloc(reply_size, GFP_KERNEL); 2233 if (!cn_reply) { 2234 retcode = ERR_NOMEM; 2235 goto fail; 2236 } 2237 reply = (struct drbd_nl_cfg_reply *) cn_reply->data; 2238 2239 reply->packet_type = 2240 cm->reply_body_size ? nlp->packet_type : P_return_code_only; 2241 reply->minor = nlp->drbd_minor; 2242 reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */ 2243 /* reply->tag_list; might be modified by cm->function. */ 2244 2245 rr = cm->function(mdev, nlp, reply); 2246 2247 cn_reply->id = req->id; 2248 cn_reply->seq = req->seq; 2249 cn_reply->ack = req->ack + 1; 2250 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr; 2251 cn_reply->flags = 0; 2252 2253 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL); 2254 if (rr && rr != -ESRCH) 2255 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr); 2256 2257 kfree(cn_reply); 2258 module_put(THIS_MODULE); 2259 return; 2260 fail: 2261 drbd_nl_send_reply(req, retcode); 2262 module_put(THIS_MODULE); 2263} 2264 2265static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */ 2266 2267static unsigned short * 2268__tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, 2269 unsigned short len, int nul_terminated) 2270{ 2271 unsigned short l = tag_descriptions[tag_number(tag)].max_len; 2272 len = (len < l) ? len : l; 2273 put_unaligned(tag, tl++); 2274 put_unaligned(len, tl++); 2275 memcpy(tl, data, len); 2276 tl = (unsigned short*)((char*)tl + len); 2277 if (nul_terminated) 2278 *((char*)tl - 1) = 0; 2279 return tl; 2280} 2281 2282static unsigned short * 2283tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len) 2284{ 2285 return __tl_add_blob(tl, tag, data, len, 0); 2286} 2287 2288static unsigned short * 2289tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str) 2290{ 2291 return __tl_add_blob(tl, tag, str, strlen(str)+1, 0); 2292} 2293 2294static unsigned short * 2295tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val) 2296{ 2297 put_unaligned(tag, tl++); 2298 switch(tag_type(tag)) { 2299 case TT_INTEGER: 2300 put_unaligned(sizeof(int), tl++); 2301 put_unaligned(*(int *)val, (int *)tl); 2302 tl = (unsigned short*)((char*)tl+sizeof(int)); 2303 break; 2304 case TT_INT64: 2305 put_unaligned(sizeof(u64), tl++); 2306 put_unaligned(*(u64 *)val, (u64 *)tl); 2307 tl = (unsigned short*)((char*)tl+sizeof(u64)); 2308 break; 2309 default: 2310 /* someone did something stupid. */ 2311 ; 2312 } 2313 return tl; 2314} 2315 2316void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state) 2317{ 2318 char buffer[sizeof(struct cn_msg)+ 2319 sizeof(struct drbd_nl_cfg_reply)+ 2320 sizeof(struct get_state_tag_len_struct)+ 2321 sizeof(short int)]; 2322 struct cn_msg *cn_reply = (struct cn_msg *) buffer; 2323 struct drbd_nl_cfg_reply *reply = 2324 (struct drbd_nl_cfg_reply *)cn_reply->data; 2325 unsigned short *tl = reply->tag_list; 2326 2327 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */ 2328 2329 tl = get_state_to_tags(mdev, (struct get_state *)&state, tl); 2330 2331 put_unaligned(TT_END, tl++); /* Close the tag list */ 2332 2333 cn_reply->id.idx = CN_IDX_DRBD; 2334 cn_reply->id.val = CN_VAL_DRBD; 2335 2336 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq); 2337 cn_reply->ack = 0; /* not used here. */ 2338 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + 2339 (int)((char *)tl - (char *)reply->tag_list); 2340 cn_reply->flags = 0; 2341 2342 reply->packet_type = P_get_state; 2343 reply->minor = mdev_to_minor(mdev); 2344 reply->ret_code = NO_ERROR; 2345 2346 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2347} 2348 2349void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name) 2350{ 2351 char buffer[sizeof(struct cn_msg)+ 2352 sizeof(struct drbd_nl_cfg_reply)+ 2353 sizeof(struct call_helper_tag_len_struct)+ 2354 sizeof(short int)]; 2355 struct cn_msg *cn_reply = (struct cn_msg *) buffer; 2356 struct drbd_nl_cfg_reply *reply = 2357 (struct drbd_nl_cfg_reply *)cn_reply->data; 2358 unsigned short *tl = reply->tag_list; 2359 2360 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */ 2361 2362 tl = tl_add_str(tl, T_helper, helper_name); 2363 put_unaligned(TT_END, tl++); /* Close the tag list */ 2364 2365 cn_reply->id.idx = CN_IDX_DRBD; 2366 cn_reply->id.val = CN_VAL_DRBD; 2367 2368 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq); 2369 cn_reply->ack = 0; /* not used here. */ 2370 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + 2371 (int)((char *)tl - (char *)reply->tag_list); 2372 cn_reply->flags = 0; 2373 2374 reply->packet_type = P_call_helper; 2375 reply->minor = mdev_to_minor(mdev); 2376 reply->ret_code = NO_ERROR; 2377 2378 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2379} 2380 2381void drbd_bcast_ee(struct drbd_conf *mdev, 2382 const char *reason, const int dgs, 2383 const char* seen_hash, const char* calc_hash, 2384 const struct drbd_epoch_entry* e) 2385{ 2386 struct cn_msg *cn_reply; 2387 struct drbd_nl_cfg_reply *reply; 2388 unsigned short *tl; 2389 struct page *page; 2390 unsigned len; 2391 2392 if (!e) 2393 return; 2394 if (!reason || !reason[0]) 2395 return; 2396 2397 /* apparently we have to memcpy twice, first to prepare the data for the 2398 * struct cn_msg, then within cn_netlink_send from the cn_msg to the 2399 * netlink skb. */ 2400 /* receiver thread context, which is not in the writeout path (of this node), 2401 * but may be in the writeout path of the _other_ node. 2402 * GFP_NOIO to avoid potential "distributed deadlock". */ 2403 cn_reply = kzalloc( 2404 sizeof(struct cn_msg)+ 2405 sizeof(struct drbd_nl_cfg_reply)+ 2406 sizeof(struct dump_ee_tag_len_struct)+ 2407 sizeof(short int), 2408 GFP_NOIO); 2409 2410 if (!cn_reply) { 2411 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n", 2412 (unsigned long long)e->sector, e->size); 2413 return; 2414 } 2415 2416 reply = (struct drbd_nl_cfg_reply*)cn_reply->data; 2417 tl = reply->tag_list; 2418 2419 tl = tl_add_str(tl, T_dump_ee_reason, reason); 2420 tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs); 2421 tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs); 2422 tl = tl_add_int(tl, T_ee_sector, &e->sector); 2423 tl = tl_add_int(tl, T_ee_block_id, &e->block_id); 2424 2425 /* dump the first 32k */ 2426 len = min_t(unsigned, e->size, 32 << 10); 2427 put_unaligned(T_ee_data, tl++); 2428 put_unaligned(len, tl++); 2429 2430 page = e->pages; 2431 page_chain_for_each(page) { 2432 void *d = kmap_atomic(page, KM_USER0); 2433 unsigned l = min_t(unsigned, len, PAGE_SIZE); 2434 memcpy(tl, d, l); 2435 kunmap_atomic(d, KM_USER0); 2436 tl = (unsigned short*)((char*)tl + l); 2437 len -= l; 2438 if (len == 0) 2439 break; 2440 } 2441 put_unaligned(TT_END, tl++); /* Close the tag list */ 2442 2443 cn_reply->id.idx = CN_IDX_DRBD; 2444 cn_reply->id.val = CN_VAL_DRBD; 2445 2446 cn_reply->seq = atomic_add_return(1,&drbd_nl_seq); 2447 cn_reply->ack = 0; // not used here. 2448 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + 2449 (int)((char*)tl - (char*)reply->tag_list); 2450 cn_reply->flags = 0; 2451 2452 reply->packet_type = P_dump_ee; 2453 reply->minor = mdev_to_minor(mdev); 2454 reply->ret_code = NO_ERROR; 2455 2456 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2457 kfree(cn_reply); 2458} 2459 2460void drbd_bcast_sync_progress(struct drbd_conf *mdev) 2461{ 2462 char buffer[sizeof(struct cn_msg)+ 2463 sizeof(struct drbd_nl_cfg_reply)+ 2464 sizeof(struct sync_progress_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 unsigned long rs_left; 2471 unsigned int res; 2472 2473 /* no local ref, no bitmap, no syncer progress, no broadcast. */ 2474 if (!get_ldev(mdev)) 2475 return; 2476 drbd_get_syncer_progress(mdev, &rs_left, &res); 2477 put_ldev(mdev); 2478 2479 tl = tl_add_int(tl, T_sync_progress, &res); 2480 put_unaligned(TT_END, tl++); /* Close the tag list */ 2481 2482 cn_reply->id.idx = CN_IDX_DRBD; 2483 cn_reply->id.val = CN_VAL_DRBD; 2484 2485 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq); 2486 cn_reply->ack = 0; /* not used here. */ 2487 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + 2488 (int)((char *)tl - (char *)reply->tag_list); 2489 cn_reply->flags = 0; 2490 2491 reply->packet_type = P_sync_progress; 2492 reply->minor = mdev_to_minor(mdev); 2493 reply->ret_code = NO_ERROR; 2494 2495 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2496} 2497 2498int __init drbd_nl_init(void) 2499{ 2500 static struct cb_id cn_id_drbd; 2501 int err, try=10; 2502 2503 cn_id_drbd.val = CN_VAL_DRBD; 2504 do { 2505 cn_id_drbd.idx = cn_idx; 2506 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback); 2507 if (!err) 2508 break; 2509 cn_idx = (cn_idx + CN_IDX_STEP); 2510 } while (try--); 2511 2512 if (err) { 2513 printk(KERN_ERR "drbd: cn_drbd failed to register\n"); 2514 return err; 2515 } 2516 2517 return 0; 2518} 2519 2520void drbd_nl_cleanup(void) 2521{ 2522 static struct cb_id cn_id_drbd; 2523 2524 cn_id_drbd.idx = cn_idx; 2525 cn_id_drbd.val = CN_VAL_DRBD; 2526 2527 cn_del_callback(&cn_id_drbd); 2528} 2529 2530void drbd_nl_send_reply(struct cn_msg *req, int ret_code) 2531{ 2532 char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)]; 2533 struct cn_msg *cn_reply = (struct cn_msg *) buffer; 2534 struct drbd_nl_cfg_reply *reply = 2535 (struct drbd_nl_cfg_reply *)cn_reply->data; 2536 int rr; 2537 2538 memset(buffer, 0, sizeof(buffer)); 2539 cn_reply->id = req->id; 2540 2541 cn_reply->seq = req->seq; 2542 cn_reply->ack = req->ack + 1; 2543 cn_reply->len = sizeof(struct drbd_nl_cfg_reply); 2544 cn_reply->flags = 0; 2545 2546 reply->packet_type = P_return_code_only; 2547 reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor; 2548 reply->ret_code = ret_code; 2549 2550 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO); 2551 if (rr && rr != -ESRCH) 2552 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr); 2553} 2554 2555