4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
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>.
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)
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.
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.
26 #include <linux/module.h>
27 #include <linux/drbd.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>
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>
44 static unsigned short *tl_add_blob(unsigned short *, enum drbd_tags, const void *, int);
45 static unsigned short *tl_add_str(unsigned short *, enum drbd_tags, const char *);
46 static unsigned short *tl_add_int(unsigned short *, enum drbd_tags, const void *);
48 /* see get_sb_bdev and bd_claim */
49 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
51 /* Generate the tag_list to struct functions */
52 #define NL_PACKET(name, number, fields) \
53 static int name ## _from_tags(struct drbd_conf *mdev, \
54 unsigned short *tags, struct name *arg) __attribute__ ((unused)); \
55 static int name ## _from_tags(struct drbd_conf *mdev, \
56 unsigned short *tags, struct name *arg) \
61 while ((tag = get_unaligned(tags++)) != TT_END) { \
62 dlen = get_unaligned(tags++); \
63 switch (tag_number(tag)) { \
66 if (tag & T_MANDATORY) { \
67 dev_err(DEV, "Unknown tag: %d\n", tag_number(tag)); \
71 tags = (unsigned short *)((char *)tags + dlen); \
75 #define NL_INTEGER(pn, pr, member) \
76 case pn: /* D_ASSERT( tag_type(tag) == TT_INTEGER ); */ \
77 arg->member = get_unaligned((int *)(tags)); \
79 #define NL_INT64(pn, pr, member) \
80 case pn: /* D_ASSERT( tag_type(tag) == TT_INT64 ); */ \
81 arg->member = get_unaligned((u64 *)(tags)); \
83 #define NL_BIT(pn, pr, member) \
84 case pn: /* D_ASSERT( tag_type(tag) == TT_BIT ); */ \
85 arg->member = *(char *)(tags) ? 1 : 0; \
87 #define NL_STRING(pn, pr, member, len) \
88 case pn: /* D_ASSERT( tag_type(tag) == TT_STRING ); */ \
90 dev_err(DEV, "arg too long: %s (%u wanted, max len: %u bytes)\n", \
91 #member, dlen, (unsigned int)len); \
94 arg->member ## _len = dlen; \
95 memcpy(arg->member, tags, min_t(size_t, dlen, len)); \
97 #include <linux/drbd_nl.h>
99 /* Generate the struct to tag_list functions */
100 #define NL_PACKET(name, number, fields) \
101 static unsigned short* \
102 name ## _to_tags(struct drbd_conf *mdev, \
103 struct name *arg, unsigned short *tags) __attribute__ ((unused)); \
104 static unsigned short* \
105 name ## _to_tags(struct drbd_conf *mdev, \
106 struct name *arg, unsigned short *tags) \
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>
134 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name);
135 void drbd_nl_send_reply(struct cn_msg *, int);
137 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
139 char *envp[] = { "HOME=/",
141 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
142 NULL, /* Will be set to address family */
143 NULL, /* Will be set to address */
146 char mb[12], af[20], ad[60], *afs;
147 char *argv[] = {usermode_helper, cmd, mb, NULL };
150 if (current == mdev->worker.task)
151 set_bit(CALLBACK_PENDING, &mdev->flags);
153 snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
155 if (get_net_conf(mdev)) {
156 switch (((struct sockaddr *)mdev->net_conf->peer_addr)->sa_family) {
159 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
160 &((struct sockaddr_in6 *)mdev->net_conf->peer_addr)->sin6_addr);
164 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
165 &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
169 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
170 &((struct sockaddr_in *)mdev->net_conf->peer_addr)->sin_addr);
172 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
178 /* The helper may take some time.
179 * write out any unsynced meta data changes now */
182 dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
184 drbd_bcast_ev_helper(mdev, cmd);
185 ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
187 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
188 usermode_helper, cmd, mb,
189 (ret >> 8) & 0xff, ret);
191 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
192 usermode_helper, cmd, mb,
193 (ret >> 8) & 0xff, ret);
195 if (current == mdev->worker.task)
196 clear_bit(CALLBACK_PENDING, &mdev->flags);
198 if (ret < 0) /* Ignore any ERRNOs we got. */
204 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
208 enum drbd_disk_state nps;
209 enum drbd_fencing_p fp;
211 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
213 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
214 fp = mdev->ldev->dc.fencing;
217 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
218 nps = mdev->state.pdsk;
222 r = drbd_khelper(mdev, "fence-peer");
224 switch ((r>>8) & 0xff) {
225 case 3: /* peer is inconsistent */
226 ex_to_string = "peer is inconsistent or worse";
227 nps = D_INCONSISTENT;
229 case 4: /* peer got outdated, or was already outdated */
230 ex_to_string = "peer was fenced";
233 case 5: /* peer was down */
234 if (mdev->state.disk == D_UP_TO_DATE) {
235 /* we will(have) create(d) a new UUID anyways... */
236 ex_to_string = "peer is unreachable, assumed to be dead";
239 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
240 nps = mdev->state.pdsk;
243 case 6: /* Peer is primary, voluntarily outdate myself.
244 * This is useful when an unconnected R_SECONDARY is asked to
245 * become R_PRIMARY, but finds the other peer being active. */
246 ex_to_string = "peer is active";
247 dev_warn(DEV, "Peer is primary, outdating myself.\n");
249 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
252 if (fp != FP_STONITH)
253 dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
254 ex_to_string = "peer was stonithed";
258 /* The script is broken ... */
260 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
264 dev_info(DEV, "fence-peer helper returned %d (%s)\n",
265 (r>>8) & 0xff, ex_to_string);
268 if (mdev->state.susp_fen && nps >= D_UNKNOWN) {
269 /* The handler was not successful... unfreeze here, the
270 state engine can not unfreeze... */
271 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE);
277 static int _try_outdate_peer_async(void *data)
279 struct drbd_conf *mdev = (struct drbd_conf *)data;
280 enum drbd_disk_state nps;
283 nps = drbd_try_outdate_peer(mdev);
286 drbd_request_state(mdev, NS(pdsk, nps));
287 here, because we might were able to re-establish the connection
288 in the meantime. This can only partially be solved in the state's
289 engine is_valid_state() and is_valid_state_transition()
292 nps can be D_INCONSISTENT, D_OUTDATED or D_UNKNOWN.
293 pdsk == D_INCONSISTENT while conn >= C_CONNECTED is valid,
294 therefore we have to have the pre state change check here.
296 spin_lock_irq(&mdev->req_lock);
298 if (ns.conn < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &mdev->flags)) {
300 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
302 spin_unlock_irq(&mdev->req_lock);
307 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
309 struct task_struct *opa;
311 opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
313 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
317 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
319 const int max_tries = 4;
320 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
323 union drbd_state mask, val;
324 enum drbd_disk_state nps;
326 if (new_role == R_PRIMARY)
327 request_ping(mdev); /* Detect a dead peer ASAP */
329 mutex_lock(&mdev->state_mutex);
331 mask.i = 0; mask.role = R_MASK;
332 val.i = 0; val.role = new_role;
334 while (try++ < max_tries) {
335 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
337 /* in case we first succeeded to outdate,
338 * but now suddenly could establish a connection */
339 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
345 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
346 (mdev->state.disk < D_UP_TO_DATE &&
347 mdev->state.disk >= D_INCONSISTENT)) {
349 val.disk = D_UP_TO_DATE;
354 if (rv == SS_NO_UP_TO_DATE_DISK &&
355 mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
356 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
357 nps = drbd_try_outdate_peer(mdev);
359 if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
360 val.disk = D_UP_TO_DATE;
370 if (rv == SS_NOTHING_TO_DO)
372 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
373 nps = drbd_try_outdate_peer(mdev);
375 if (force && nps > D_OUTDATED) {
376 dev_warn(DEV, "Forced into split brain situation!\n");
385 if (rv == SS_TWO_PRIMARIES) {
386 /* Maybe the peer is detected as dead very soon...
387 retry at most once more in this case. */
388 schedule_timeout_interruptible((mdev->net_conf->ping_timeo+1)*HZ/10);
393 if (rv < SS_SUCCESS) {
394 rv = _drbd_request_state(mdev, mask, val,
395 CS_VERBOSE + CS_WAIT_COMPLETE);
406 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
408 /* Wait until nothing is on the fly :) */
409 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
411 if (new_role == R_SECONDARY) {
412 set_disk_ro(mdev->vdisk, true);
413 if (get_ldev(mdev)) {
414 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
418 if (get_net_conf(mdev)) {
419 mdev->net_conf->want_lose = 0;
422 set_disk_ro(mdev->vdisk, false);
423 if (get_ldev(mdev)) {
424 if (((mdev->state.conn < C_CONNECTED ||
425 mdev->state.pdsk <= D_FAILED)
426 && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
427 drbd_uuid_new_current(mdev);
429 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
434 /* writeout of activity log covered areas of the bitmap
435 * to stable storage done in after state change already */
437 if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
438 /* if this was forced, we should consider sync */
440 drbd_send_uuids(mdev);
441 drbd_send_current_state(mdev);
446 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
448 mutex_unlock(&mdev->state_mutex);
452 static struct drbd_conf *ensure_mdev(int minor, int create)
454 struct drbd_conf *mdev;
456 if (minor >= minor_count)
459 mdev = minor_to_mdev(minor);
461 if (!mdev && create) {
462 struct gendisk *disk = NULL;
463 mdev = drbd_new_device(minor);
465 spin_lock_irq(&drbd_pp_lock);
466 if (minor_table[minor] == NULL) {
467 minor_table[minor] = mdev;
470 } /* else: we lost the race */
471 spin_unlock_irq(&drbd_pp_lock);
473 if (disk) /* we won the race above */
474 /* in case we ever add a drbd_delete_device(),
475 * don't forget the del_gendisk! */
477 else /* we lost the race above */
478 drbd_free_mdev(mdev);
480 mdev = minor_to_mdev(minor);
486 static int drbd_nl_primary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
487 struct drbd_nl_cfg_reply *reply)
489 struct primary primary_args;
491 memset(&primary_args, 0, sizeof(struct primary));
492 if (!primary_from_tags(mdev, nlp->tag_list, &primary_args)) {
493 reply->ret_code = ERR_MANDATORY_TAG;
498 drbd_set_role(mdev, R_PRIMARY, primary_args.primary_force);
503 static int drbd_nl_secondary(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
504 struct drbd_nl_cfg_reply *reply)
506 reply->ret_code = drbd_set_role(mdev, R_SECONDARY, 0);
511 /* initializes the md.*_offset members, so we are able to find
512 * the on disk meta data */
513 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
514 struct drbd_backing_dev *bdev)
516 sector_t md_size_sect = 0;
517 switch (bdev->dc.meta_dev_idx) {
519 /* v07 style fixed size indexed meta data */
520 bdev->md.md_size_sect = MD_RESERVED_SECT;
521 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
522 bdev->md.al_offset = MD_AL_OFFSET;
523 bdev->md.bm_offset = MD_BM_OFFSET;
525 case DRBD_MD_INDEX_FLEX_EXT:
526 /* just occupy the full device; unit: sectors */
527 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
528 bdev->md.md_offset = 0;
529 bdev->md.al_offset = MD_AL_OFFSET;
530 bdev->md.bm_offset = MD_BM_OFFSET;
532 case DRBD_MD_INDEX_INTERNAL:
533 case DRBD_MD_INDEX_FLEX_INT:
534 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
535 /* al size is still fixed */
536 bdev->md.al_offset = -MD_AL_MAX_SIZE;
537 /* we need (slightly less than) ~ this much bitmap sectors: */
538 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
539 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
540 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
541 md_size_sect = ALIGN(md_size_sect, 8);
543 /* plus the "drbd meta data super block",
544 * and the activity log; */
545 md_size_sect += MD_BM_OFFSET;
547 bdev->md.md_size_sect = md_size_sect;
548 /* bitmap offset is adjusted by 'super' block size */
549 bdev->md.bm_offset = -md_size_sect + MD_AL_OFFSET;
554 /* input size is expected to be in KB */
555 char *ppsize(char *buf, unsigned long long size)
557 /* Needs 9 bytes at max including trailing NUL:
558 * -1ULL ==> "16384 EB" */
559 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
561 while (size >= 10000 && base < sizeof(units)-1) {
563 size = (size >> 10) + !!(size & (1<<9));
566 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
571 /* there is still a theoretical deadlock when called from receiver
572 * on an D_INCONSISTENT R_PRIMARY:
573 * remote READ does inc_ap_bio, receiver would need to receive answer
574 * packet from remote to dec_ap_bio again.
575 * receiver receive_sizes(), comes here,
576 * waits for ap_bio_cnt == 0. -> deadlock.
577 * but this cannot happen, actually, because:
578 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
579 * (not connected, or bad/no disk on peer):
580 * see drbd_fail_request_early, ap_bio_cnt is zero.
581 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
582 * peer may not initiate a resize.
584 void drbd_suspend_io(struct drbd_conf *mdev)
586 set_bit(SUSPEND_IO, &mdev->flags);
587 if (is_susp(mdev->state))
589 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
592 void drbd_resume_io(struct drbd_conf *mdev)
594 clear_bit(SUSPEND_IO, &mdev->flags);
595 wake_up(&mdev->misc_wait);
599 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
600 * @mdev: DRBD device.
602 * Returns 0 on success, negative return values indicate errors.
603 * You should call drbd_md_sync() after calling this function.
605 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
607 sector_t prev_first_sect, prev_size; /* previous meta location */
612 int md_moved, la_size_changed;
613 enum determine_dev_size rv = unchanged;
616 * application request passes inc_ap_bio,
617 * but then cannot get an AL-reference.
618 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
621 * Suspend IO right here.
622 * still lock the act_log to not trigger ASSERTs there.
624 drbd_suspend_io(mdev);
626 /* no wait necessary anymore, actually we could assert that */
627 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
629 prev_first_sect = drbd_md_first_sector(mdev->ldev);
630 prev_size = mdev->ldev->md.md_size_sect;
631 la_size = mdev->ldev->md.la_size_sect;
633 /* TODO: should only be some assert here, not (re)init... */
634 drbd_md_set_sector_offsets(mdev, mdev->ldev);
636 size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
638 if (drbd_get_capacity(mdev->this_bdev) != size ||
639 drbd_bm_capacity(mdev) != size) {
641 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
643 /* currently there is only one error: ENOMEM! */
644 size = drbd_bm_capacity(mdev)>>1;
646 dev_err(DEV, "OUT OF MEMORY! "
647 "Could not allocate bitmap!\n");
649 dev_err(DEV, "BM resizing failed. "
650 "Leaving size unchanged at size = %lu KB\n",
651 (unsigned long)size);
655 /* racy, see comments above. */
656 drbd_set_my_capacity(mdev, size);
657 mdev->ldev->md.la_size_sect = size;
658 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
659 (unsigned long long)size>>1);
661 if (rv == dev_size_error)
664 la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
666 md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
667 || prev_size != mdev->ldev->md.md_size_sect;
669 if (la_size_changed || md_moved) {
672 drbd_al_shrink(mdev); /* All extents inactive. */
673 dev_info(DEV, "Writing the whole bitmap, %s\n",
674 la_size_changed && md_moved ? "size changed and md moved" :
675 la_size_changed ? "size changed" : "md moved");
676 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
677 err = drbd_bitmap_io(mdev, &drbd_bm_write,
678 "size changed", BM_LOCKED_MASK);
683 drbd_md_mark_dirty(mdev);
691 lc_unlock(mdev->act_log);
692 wake_up(&mdev->al_wait);
693 drbd_resume_io(mdev);
699 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
701 sector_t p_size = mdev->p_size; /* partner's disk size. */
702 sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
703 sector_t m_size; /* my size */
704 sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
707 m_size = drbd_get_max_capacity(bdev);
709 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
710 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
714 if (p_size && m_size) {
715 size = min_t(sector_t, p_size, m_size);
719 if (m_size && m_size < size)
721 if (p_size && p_size < size)
732 dev_err(DEV, "Both nodes diskless!\n");
736 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
737 (unsigned long)u_size>>1, (unsigned long)size>>1);
746 * drbd_check_al_size() - Ensures that the AL is of the right size
747 * @mdev: DRBD device.
749 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
750 * failed, and 0 on success. You should call drbd_md_sync() after you called
753 static int drbd_check_al_size(struct drbd_conf *mdev)
755 struct lru_cache *n, *t;
756 struct lc_element *e;
760 ERR_IF(mdev->sync_conf.al_extents < 7)
761 mdev->sync_conf.al_extents = 127;
764 mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
769 n = lc_create("act_log", drbd_al_ext_cache,
770 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
773 dev_err(DEV, "Cannot allocate act_log lru!\n");
776 spin_lock_irq(&mdev->al_lock);
778 for (i = 0; i < t->nr_elements; i++) {
779 e = lc_element_by_index(t, i);
781 dev_err(DEV, "refcnt(%d)==%d\n",
782 e->lc_number, e->refcnt);
788 spin_unlock_irq(&mdev->al_lock);
790 dev_err(DEV, "Activity log still in use!\n");
797 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
801 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
803 struct request_queue * const q = mdev->rq_queue;
804 unsigned int max_hw_sectors = max_bio_size >> 9;
805 unsigned int max_segments = 0;
807 if (get_ldev_if_state(mdev, D_ATTACHING)) {
808 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
810 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
811 max_segments = mdev->ldev->dc.max_bio_bvecs;
815 blk_queue_logical_block_size(q, 512);
816 blk_queue_max_hw_sectors(q, max_hw_sectors);
817 /* This is the workaround for "bio would need to, but cannot, be split" */
818 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
819 blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
821 if (get_ldev_if_state(mdev, D_ATTACHING)) {
822 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
824 blk_queue_stack_limits(q, b);
826 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
827 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
828 q->backing_dev_info.ra_pages,
829 b->backing_dev_info.ra_pages);
830 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
836 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
838 unsigned int now, new, local, peer;
840 now = queue_max_hw_sectors(mdev->rq_queue) << 9;
841 local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
842 peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
844 if (get_ldev_if_state(mdev, D_ATTACHING)) {
845 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
846 mdev->local_max_bio_size = local;
849 local = min(local, DRBD_MAX_BIO_SIZE);
851 /* We may ignore peer limits if the peer is modern enough.
852 Because new from 8.3.8 onwards the peer can use multiple
853 BIOs for a single peer_request */
854 if (mdev->state.conn >= C_CONNECTED) {
855 if (mdev->agreed_pro_version < 94) {
856 peer = min(mdev->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
857 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
858 } else if (mdev->agreed_pro_version == 94)
859 peer = DRBD_MAX_SIZE_H80_PACKET;
860 else /* drbd 8.3.8 onwards */
861 peer = DRBD_MAX_BIO_SIZE;
864 new = min(local, peer);
866 if (mdev->state.role == R_PRIMARY && new < now)
867 dev_err(DEV, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
870 dev_info(DEV, "max BIO size = %u\n", new);
872 drbd_setup_queue_param(mdev, new);
875 /* serialize deconfig (worker exiting, doing cleanup)
876 * and reconfig (drbdsetup disk, drbdsetup net)
878 * Wait for a potentially exiting worker, then restart it,
879 * or start a new one. Flush any pending work, there may still be an
880 * after_state_change queued.
882 static void drbd_reconfig_start(struct drbd_conf *mdev)
884 wait_event(mdev->state_wait, !test_and_set_bit(CONFIG_PENDING, &mdev->flags));
885 wait_event(mdev->state_wait, !test_bit(DEVICE_DYING, &mdev->flags));
886 drbd_thread_start(&mdev->worker);
887 drbd_flush_workqueue(mdev);
890 /* if still unconfigured, stops worker again.
891 * if configured now, clears CONFIG_PENDING.
892 * wakes potential waiters */
893 static void drbd_reconfig_done(struct drbd_conf *mdev)
895 spin_lock_irq(&mdev->req_lock);
896 if (mdev->state.disk == D_DISKLESS &&
897 mdev->state.conn == C_STANDALONE &&
898 mdev->state.role == R_SECONDARY) {
899 set_bit(DEVICE_DYING, &mdev->flags);
900 drbd_thread_stop_nowait(&mdev->worker);
902 clear_bit(CONFIG_PENDING, &mdev->flags);
903 spin_unlock_irq(&mdev->req_lock);
904 wake_up(&mdev->state_wait);
907 /* Make sure IO is suspended before calling this function(). */
908 static void drbd_suspend_al(struct drbd_conf *mdev)
912 if (lc_try_lock(mdev->act_log)) {
913 drbd_al_shrink(mdev);
914 lc_unlock(mdev->act_log);
916 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
920 spin_lock_irq(&mdev->req_lock);
921 if (mdev->state.conn < C_CONNECTED)
922 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
924 spin_unlock_irq(&mdev->req_lock);
927 dev_info(DEV, "Suspended AL updates\n");
930 /* does always return 0;
931 * interesting return code is in reply->ret_code */
932 static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
933 struct drbd_nl_cfg_reply *reply)
935 enum drbd_ret_code retcode;
936 enum determine_dev_size dd;
937 sector_t max_possible_sectors;
938 sector_t min_md_device_sectors;
939 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
940 struct block_device *bdev;
941 struct lru_cache *resync_lru = NULL;
942 union drbd_state ns, os;
943 enum drbd_state_rv rv;
944 int cp_discovered = 0;
945 int logical_block_size;
947 drbd_reconfig_start(mdev);
949 /* if you want to reconfigure, please tear down first */
950 if (mdev->state.disk > D_DISKLESS) {
951 retcode = ERR_DISK_CONFIGURED;
954 /* It may just now have detached because of IO error. Make sure
955 * drbd_ldev_destroy is done already, we may end up here very fast,
956 * e.g. if someone calls attach from the on-io-error handler,
957 * to realize a "hot spare" feature (not that I'd recommend that) */
958 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
960 /* make sure there is no leftover from previous force-detach attempts */
961 clear_bit(FORCE_DETACH, &mdev->flags);
963 /* and no leftover from previously aborted resync or verify, either */
966 atomic_set(&mdev->rs_pending_cnt, 0);
968 /* allocation not in the IO path, cqueue thread context */
969 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
975 nbc->dc.disk_size = DRBD_DISK_SIZE_SECT_DEF;
976 nbc->dc.on_io_error = DRBD_ON_IO_ERROR_DEF;
977 nbc->dc.fencing = DRBD_FENCING_DEF;
978 nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
980 if (!disk_conf_from_tags(mdev, nlp->tag_list, &nbc->dc)) {
981 retcode = ERR_MANDATORY_TAG;
985 if (nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
986 retcode = ERR_MD_IDX_INVALID;
990 if (get_net_conf(mdev)) {
991 int prot = mdev->net_conf->wire_protocol;
993 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
994 retcode = ERR_STONITH_AND_PROT_A;
999 bdev = blkdev_get_by_path(nbc->dc.backing_dev,
1000 FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1002 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
1004 retcode = ERR_OPEN_DISK;
1007 nbc->backing_bdev = bdev;
1010 * meta_dev_idx >= 0: external fixed size, possibly multiple
1011 * drbd sharing one meta device. TODO in that case, paranoia
1012 * check that [md_bdev, meta_dev_idx] is not yet used by some
1013 * other drbd minor! (if you use drbd.conf + drbdadm, that
1014 * should check it for you already; but if you don't, or
1015 * someone fooled it, we need to double check here)
1017 bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1018 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1019 (nbc->dc.meta_dev_idx < 0) ?
1020 (void *)mdev : (void *)drbd_m_holder);
1022 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1024 retcode = ERR_OPEN_MD_DISK;
1027 nbc->md_bdev = bdev;
1029 if ((nbc->backing_bdev == nbc->md_bdev) !=
1030 (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1031 nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1032 retcode = ERR_MD_IDX_INVALID;
1036 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1037 61, sizeof(struct bm_extent),
1038 offsetof(struct bm_extent, lce));
1040 retcode = ERR_NOMEM;
1044 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1045 drbd_md_set_sector_offsets(mdev, nbc);
1047 if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1048 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1049 (unsigned long long) drbd_get_max_capacity(nbc),
1050 (unsigned long long) nbc->dc.disk_size);
1051 retcode = ERR_DISK_TOO_SMALL;
1055 if (nbc->dc.meta_dev_idx < 0) {
1056 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1057 /* at least one MB, otherwise it does not make sense */
1058 min_md_device_sectors = (2<<10);
1060 max_possible_sectors = DRBD_MAX_SECTORS;
1061 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1064 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1065 retcode = ERR_MD_DISK_TOO_SMALL;
1066 dev_warn(DEV, "refusing attach: md-device too small, "
1067 "at least %llu sectors needed for this meta-disk type\n",
1068 (unsigned long long) min_md_device_sectors);
1072 /* Make sure the new disk is big enough
1073 * (we may currently be R_PRIMARY with no local disk...) */
1074 if (drbd_get_max_capacity(nbc) <
1075 drbd_get_capacity(mdev->this_bdev)) {
1076 retcode = ERR_DISK_TOO_SMALL;
1080 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1082 if (nbc->known_size > max_possible_sectors) {
1083 dev_warn(DEV, "==> truncating very big lower level device "
1084 "to currently maximum possible %llu sectors <==\n",
1085 (unsigned long long) max_possible_sectors);
1086 if (nbc->dc.meta_dev_idx >= 0)
1087 dev_warn(DEV, "==>> using internal or flexible "
1088 "meta data may help <<==\n");
1091 drbd_suspend_io(mdev);
1092 /* also wait for the last barrier ack. */
1093 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1094 /* and for any other previously queued work */
1095 drbd_flush_workqueue(mdev);
1097 rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1098 retcode = rv; /* FIXME: Type mismatch. */
1099 drbd_resume_io(mdev);
1100 if (rv < SS_SUCCESS)
1103 if (!get_ldev_if_state(mdev, D_ATTACHING))
1104 goto force_diskless;
1106 drbd_md_set_sector_offsets(mdev, nbc);
1108 /* allocate a second IO page if logical_block_size != 512 */
1109 logical_block_size = bdev_logical_block_size(nbc->md_bdev);
1110 if (logical_block_size == 0)
1111 logical_block_size = MD_SECTOR_SIZE;
1113 if (logical_block_size != MD_SECTOR_SIZE) {
1114 if (!mdev->md_io_tmpp) {
1115 struct page *page = alloc_page(GFP_NOIO);
1117 goto force_diskless_dec;
1119 dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
1120 logical_block_size, MD_SECTOR_SIZE);
1121 dev_warn(DEV, "Workaround engaged (has performance impact).\n");
1123 mdev->md_io_tmpp = page;
1127 if (!mdev->bitmap) {
1128 if (drbd_bm_init(mdev)) {
1129 retcode = ERR_NOMEM;
1130 goto force_diskless_dec;
1134 retcode = drbd_md_read(mdev, nbc);
1135 if (retcode != NO_ERROR)
1136 goto force_diskless_dec;
1138 if (mdev->state.conn < C_CONNECTED &&
1139 mdev->state.role == R_PRIMARY &&
1140 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1141 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1142 (unsigned long long)mdev->ed_uuid);
1143 retcode = ERR_DATA_NOT_CURRENT;
1144 goto force_diskless_dec;
1147 /* Since we are diskless, fix the activity log first... */
1148 if (drbd_check_al_size(mdev)) {
1149 retcode = ERR_NOMEM;
1150 goto force_diskless_dec;
1153 /* Prevent shrinking of consistent devices ! */
1154 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1155 drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1156 dev_warn(DEV, "refusing to truncate a consistent device\n");
1157 retcode = ERR_DISK_TOO_SMALL;
1158 goto force_diskless_dec;
1161 if (!drbd_al_read_log(mdev, nbc)) {
1162 retcode = ERR_IO_MD_DISK;
1163 goto force_diskless_dec;
1166 /* Reset the "barriers don't work" bits here, then force meta data to
1167 * be written, to ensure we determine if barriers are supported. */
1168 if (nbc->dc.no_md_flush)
1169 set_bit(MD_NO_FUA, &mdev->flags);
1171 clear_bit(MD_NO_FUA, &mdev->flags);
1173 /* Point of no return reached.
1174 * Devices and memory are no longer released by error cleanup below.
1175 * now mdev takes over responsibility, and the state engine should
1176 * clean it up somewhere. */
1177 D_ASSERT(mdev->ldev == NULL);
1179 mdev->resync = resync_lru;
1183 mdev->write_ordering = WO_bdev_flush;
1184 drbd_bump_write_ordering(mdev, WO_bdev_flush);
1186 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1187 set_bit(CRASHED_PRIMARY, &mdev->flags);
1189 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1191 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1192 !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1193 set_bit(CRASHED_PRIMARY, &mdev->flags);
1202 drbd_reconsider_max_bio_size(mdev);
1204 /* If I am currently not R_PRIMARY,
1205 * but meta data primary indicator is set,
1206 * I just now recover from a hard crash,
1207 * and have been R_PRIMARY before that crash.
1209 * Now, if I had no connection before that crash
1210 * (have been degraded R_PRIMARY), chances are that
1211 * I won't find my peer now either.
1213 * In that case, and _only_ in that case,
1214 * we use the degr-wfc-timeout instead of the default,
1215 * so we can automatically recover from a crash of a
1216 * degraded but active "cluster" after a certain timeout.
1218 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1219 if (mdev->state.role != R_PRIMARY &&
1220 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1221 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1222 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1224 dd = drbd_determine_dev_size(mdev, 0);
1225 if (dd == dev_size_error) {
1226 retcode = ERR_NOMEM_BITMAP;
1227 goto force_diskless_dec;
1228 } else if (dd == grew)
1229 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1231 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1232 dev_info(DEV, "Assuming that all blocks are out of sync "
1233 "(aka FullSync)\n");
1234 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1235 "set_n_write from attaching", BM_LOCKED_MASK)) {
1236 retcode = ERR_IO_MD_DISK;
1237 goto force_diskless_dec;
1240 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1241 "read from attaching", BM_LOCKED_MASK) < 0) {
1242 retcode = ERR_IO_MD_DISK;
1243 goto force_diskless_dec;
1247 if (cp_discovered) {
1248 drbd_al_apply_to_bm(mdev);
1249 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1250 "crashed primary apply AL", BM_LOCKED_MASK)) {
1251 retcode = ERR_IO_MD_DISK;
1252 goto force_diskless_dec;
1256 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1257 drbd_suspend_al(mdev); /* IO is still suspended here... */
1259 spin_lock_irq(&mdev->req_lock);
1262 /* If MDF_CONSISTENT is not set go into inconsistent state,
1263 otherwise investigate MDF_WasUpToDate...
1264 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1265 otherwise into D_CONSISTENT state.
1267 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1268 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1269 ns.disk = D_CONSISTENT;
1271 ns.disk = D_OUTDATED;
1273 ns.disk = D_INCONSISTENT;
1276 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1277 ns.pdsk = D_OUTDATED;
1279 if ( ns.disk == D_CONSISTENT &&
1280 (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1281 ns.disk = D_UP_TO_DATE;
1283 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1284 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1285 this point, because drbd_request_state() modifies these
1288 /* In case we are C_CONNECTED postpone any decision on the new disk
1289 state after the negotiation phase. */
1290 if (mdev->state.conn == C_CONNECTED) {
1291 mdev->new_state_tmp.i = ns.i;
1293 ns.disk = D_NEGOTIATING;
1295 /* We expect to receive up-to-date UUIDs soon.
1296 To avoid a race in receive_state, free p_uuid while
1297 holding req_lock. I.e. atomic with the state change */
1298 kfree(mdev->p_uuid);
1299 mdev->p_uuid = NULL;
1302 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1304 spin_unlock_irq(&mdev->req_lock);
1306 if (rv < SS_SUCCESS)
1307 goto force_diskless_dec;
1309 if (mdev->state.role == R_PRIMARY)
1310 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1312 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1314 drbd_md_mark_dirty(mdev);
1317 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1319 reply->ret_code = retcode;
1320 drbd_reconfig_done(mdev);
1326 drbd_force_state(mdev, NS(disk, D_FAILED));
1330 if (nbc->backing_bdev)
1331 blkdev_put(nbc->backing_bdev,
1332 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1334 blkdev_put(nbc->md_bdev,
1335 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1338 lc_destroy(resync_lru);
1340 reply->ret_code = retcode;
1341 drbd_reconfig_done(mdev);
1345 /* Detaching the disk is a process in multiple stages. First we need to lock
1346 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1347 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1348 * internal references as well.
1349 * Only then we have finally detached. */
1350 static int drbd_nl_detach(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1351 struct drbd_nl_cfg_reply *reply)
1353 enum drbd_ret_code retcode;
1355 struct detach dt = {};
1357 if (!detach_from_tags(mdev, nlp->tag_list, &dt)) {
1358 reply->ret_code = ERR_MANDATORY_TAG;
1362 if (dt.detach_force) {
1363 set_bit(FORCE_DETACH, &mdev->flags);
1364 drbd_force_state(mdev, NS(disk, D_FAILED));
1365 reply->ret_code = SS_SUCCESS;
1369 drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1370 drbd_md_get_buffer(mdev); /* make sure there is no in-flight meta-data IO */
1371 retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1372 drbd_md_put_buffer(mdev);
1373 /* D_FAILED will transition to DISKLESS. */
1374 ret = wait_event_interruptible(mdev->misc_wait,
1375 mdev->state.disk != D_FAILED);
1376 drbd_resume_io(mdev);
1378 if ((int)retcode == (int)SS_IS_DISKLESS)
1379 retcode = SS_NOTHING_TO_DO;
1382 reply->ret_code = retcode;
1387 static int drbd_nl_net_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1388 struct drbd_nl_cfg_reply *reply)
1391 enum drbd_ret_code retcode;
1392 struct net_conf *new_conf = NULL;
1393 struct crypto_hash *tfm = NULL;
1394 struct crypto_hash *integrity_w_tfm = NULL;
1395 struct crypto_hash *integrity_r_tfm = NULL;
1396 struct hlist_head *new_tl_hash = NULL;
1397 struct hlist_head *new_ee_hash = NULL;
1398 struct drbd_conf *odev;
1399 char hmac_name[CRYPTO_MAX_ALG_NAME];
1400 void *int_dig_out = NULL;
1401 void *int_dig_in = NULL;
1402 void *int_dig_vv = NULL;
1403 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1405 drbd_reconfig_start(mdev);
1407 if (mdev->state.conn > C_STANDALONE) {
1408 retcode = ERR_NET_CONFIGURED;
1412 /* allocation not in the IO path, cqueue thread context */
1413 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1415 retcode = ERR_NOMEM;
1419 new_conf->timeout = DRBD_TIMEOUT_DEF;
1420 new_conf->try_connect_int = DRBD_CONNECT_INT_DEF;
1421 new_conf->ping_int = DRBD_PING_INT_DEF;
1422 new_conf->max_epoch_size = DRBD_MAX_EPOCH_SIZE_DEF;
1423 new_conf->max_buffers = DRBD_MAX_BUFFERS_DEF;
1424 new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1425 new_conf->sndbuf_size = DRBD_SNDBUF_SIZE_DEF;
1426 new_conf->rcvbuf_size = DRBD_RCVBUF_SIZE_DEF;
1427 new_conf->ko_count = DRBD_KO_COUNT_DEF;
1428 new_conf->after_sb_0p = DRBD_AFTER_SB_0P_DEF;
1429 new_conf->after_sb_1p = DRBD_AFTER_SB_1P_DEF;
1430 new_conf->after_sb_2p = DRBD_AFTER_SB_2P_DEF;
1431 new_conf->want_lose = 0;
1432 new_conf->two_primaries = 0;
1433 new_conf->wire_protocol = DRBD_PROT_C;
1434 new_conf->ping_timeo = DRBD_PING_TIMEO_DEF;
1435 new_conf->rr_conflict = DRBD_RR_CONFLICT_DEF;
1436 new_conf->on_congestion = DRBD_ON_CONGESTION_DEF;
1437 new_conf->cong_extents = DRBD_CONG_EXTENTS_DEF;
1439 if (!net_conf_from_tags(mdev, nlp->tag_list, new_conf)) {
1440 retcode = ERR_MANDATORY_TAG;
1444 if (new_conf->two_primaries
1445 && (new_conf->wire_protocol != DRBD_PROT_C)) {
1446 retcode = ERR_NOT_PROTO_C;
1450 if (get_ldev(mdev)) {
1451 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1453 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1454 retcode = ERR_STONITH_AND_PROT_A;
1459 if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1460 retcode = ERR_CONG_NOT_PROTO_A;
1464 if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1465 retcode = ERR_DISCARD;
1471 new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1472 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1473 for (i = 0; i < minor_count; i++) {
1474 odev = minor_to_mdev(i);
1475 if (!odev || odev == mdev)
1477 if (get_net_conf(odev)) {
1478 taken_addr = (struct sockaddr *)&odev->net_conf->my_addr;
1479 if (new_conf->my_addr_len == odev->net_conf->my_addr_len &&
1480 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1481 retcode = ERR_LOCAL_ADDR;
1483 taken_addr = (struct sockaddr *)&odev->net_conf->peer_addr;
1484 if (new_conf->peer_addr_len == odev->net_conf->peer_addr_len &&
1485 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1486 retcode = ERR_PEER_ADDR;
1489 if (retcode != NO_ERROR)
1494 if (new_conf->cram_hmac_alg[0] != 0) {
1495 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1496 new_conf->cram_hmac_alg);
1497 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1500 retcode = ERR_AUTH_ALG;
1504 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1505 retcode = ERR_AUTH_ALG_ND;
1510 if (new_conf->integrity_alg[0]) {
1511 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1512 if (IS_ERR(integrity_w_tfm)) {
1513 integrity_w_tfm = NULL;
1514 retcode=ERR_INTEGRITY_ALG;
1518 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1519 retcode=ERR_INTEGRITY_ALG_ND;
1523 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1524 if (IS_ERR(integrity_r_tfm)) {
1525 integrity_r_tfm = NULL;
1526 retcode=ERR_INTEGRITY_ALG;
1531 ns = new_conf->max_epoch_size/8;
1532 if (mdev->tl_hash_s != ns) {
1533 new_tl_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1535 retcode = ERR_NOMEM;
1540 ns = new_conf->max_buffers/8;
1541 if (new_conf->two_primaries && (mdev->ee_hash_s != ns)) {
1542 new_ee_hash = kzalloc(ns*sizeof(void *), GFP_KERNEL);
1544 retcode = ERR_NOMEM;
1549 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1551 if (integrity_w_tfm) {
1552 i = crypto_hash_digestsize(integrity_w_tfm);
1553 int_dig_out = kmalloc(i, GFP_KERNEL);
1555 retcode = ERR_NOMEM;
1558 int_dig_in = kmalloc(i, GFP_KERNEL);
1560 retcode = ERR_NOMEM;
1563 int_dig_vv = kmalloc(i, GFP_KERNEL);
1565 retcode = ERR_NOMEM;
1570 if (!mdev->bitmap) {
1571 if(drbd_bm_init(mdev)) {
1572 retcode = ERR_NOMEM;
1577 drbd_flush_workqueue(mdev);
1578 spin_lock_irq(&mdev->req_lock);
1579 if (mdev->net_conf != NULL) {
1580 retcode = ERR_NET_CONFIGURED;
1581 spin_unlock_irq(&mdev->req_lock);
1584 mdev->net_conf = new_conf;
1590 kfree(mdev->tl_hash);
1591 mdev->tl_hash_s = mdev->net_conf->max_epoch_size/8;
1592 mdev->tl_hash = new_tl_hash;
1596 kfree(mdev->ee_hash);
1597 mdev->ee_hash_s = mdev->net_conf->max_buffers/8;
1598 mdev->ee_hash = new_ee_hash;
1601 crypto_free_hash(mdev->cram_hmac_tfm);
1602 mdev->cram_hmac_tfm = tfm;
1604 crypto_free_hash(mdev->integrity_w_tfm);
1605 mdev->integrity_w_tfm = integrity_w_tfm;
1607 crypto_free_hash(mdev->integrity_r_tfm);
1608 mdev->integrity_r_tfm = integrity_r_tfm;
1610 kfree(mdev->int_dig_out);
1611 kfree(mdev->int_dig_in);
1612 kfree(mdev->int_dig_vv);
1613 mdev->int_dig_out=int_dig_out;
1614 mdev->int_dig_in=int_dig_in;
1615 mdev->int_dig_vv=int_dig_vv;
1616 retcode = _drbd_set_state(_NS(mdev, conn, C_UNCONNECTED), CS_VERBOSE, NULL);
1617 spin_unlock_irq(&mdev->req_lock);
1619 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1620 reply->ret_code = retcode;
1621 drbd_reconfig_done(mdev);
1628 crypto_free_hash(tfm);
1629 crypto_free_hash(integrity_w_tfm);
1630 crypto_free_hash(integrity_r_tfm);
1635 reply->ret_code = retcode;
1636 drbd_reconfig_done(mdev);
1640 static int drbd_nl_disconnect(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1641 struct drbd_nl_cfg_reply *reply)
1644 struct disconnect dc;
1646 memset(&dc, 0, sizeof(struct disconnect));
1647 if (!disconnect_from_tags(mdev, nlp->tag_list, &dc)) {
1648 retcode = ERR_MANDATORY_TAG;
1653 spin_lock_irq(&mdev->req_lock);
1654 if (mdev->state.conn >= C_WF_CONNECTION)
1655 _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), CS_HARD, NULL);
1656 spin_unlock_irq(&mdev->req_lock);
1660 retcode = _drbd_request_state(mdev, NS(conn, C_DISCONNECTING), CS_ORDERED);
1662 if (retcode == SS_NOTHING_TO_DO)
1664 else if (retcode == SS_ALREADY_STANDALONE)
1666 else if (retcode == SS_PRIMARY_NOP) {
1667 /* Our statche checking code wants to see the peer outdated. */
1668 retcode = drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1670 } else if (retcode == SS_CW_FAILED_BY_PEER) {
1671 /* The peer probably wants to see us outdated. */
1672 retcode = _drbd_request_state(mdev, NS2(conn, C_DISCONNECTING,
1675 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1676 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1677 retcode = SS_SUCCESS;
1681 if (retcode < SS_SUCCESS)
1684 if (wait_event_interruptible(mdev->state_wait,
1685 mdev->state.conn != C_DISCONNECTING)) {
1686 /* Do not test for mdev->state.conn == C_STANDALONE, since
1687 someone else might connect us in the mean time! */
1696 reply->ret_code = retcode;
1700 void resync_after_online_grow(struct drbd_conf *mdev)
1702 int iass; /* I am sync source */
1704 dev_info(DEV, "Resync of new storage after online grow\n");
1705 if (mdev->state.role != mdev->state.peer)
1706 iass = (mdev->state.role == R_PRIMARY);
1708 iass = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1711 drbd_start_resync(mdev, C_SYNC_SOURCE);
1713 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1716 static int drbd_nl_resize(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1717 struct drbd_nl_cfg_reply *reply)
1720 int retcode = NO_ERROR;
1721 enum determine_dev_size dd;
1722 enum dds_flags ddsf;
1724 memset(&rs, 0, sizeof(struct resize));
1725 if (!resize_from_tags(mdev, nlp->tag_list, &rs)) {
1726 retcode = ERR_MANDATORY_TAG;
1730 if (mdev->state.conn > C_CONNECTED) {
1731 retcode = ERR_RESIZE_RESYNC;
1735 if (mdev->state.role == R_SECONDARY &&
1736 mdev->state.peer == R_SECONDARY) {
1737 retcode = ERR_NO_PRIMARY;
1741 if (!get_ldev(mdev)) {
1742 retcode = ERR_NO_DISK;
1746 if (rs.no_resync && mdev->agreed_pro_version < 93) {
1747 retcode = ERR_NEED_APV_93;
1751 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1752 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1754 mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1755 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1756 dd = drbd_determine_dev_size(mdev, ddsf);
1759 if (dd == dev_size_error) {
1760 retcode = ERR_NOMEM_BITMAP;
1764 if (mdev->state.conn == C_CONNECTED) {
1766 set_bit(RESIZE_PENDING, &mdev->flags);
1768 drbd_send_uuids(mdev);
1769 drbd_send_sizes(mdev, 1, ddsf);
1773 reply->ret_code = retcode;
1781 static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1782 struct drbd_nl_cfg_reply *reply)
1784 int retcode = NO_ERROR;
1786 int ovr; /* online verify running */
1787 int rsr; /* re-sync running */
1788 struct crypto_hash *verify_tfm = NULL;
1789 struct crypto_hash *csums_tfm = NULL;
1790 struct syncer_conf sc;
1791 cpumask_var_t new_cpu_mask;
1792 int *rs_plan_s = NULL;
1795 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1796 retcode = ERR_NOMEM;
1800 if (nlp->flags & DRBD_NL_SET_DEFAULTS) {
1801 memset(&sc, 0, sizeof(struct syncer_conf));
1802 sc.rate = DRBD_RATE_DEF;
1803 sc.after = DRBD_AFTER_DEF;
1804 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1805 sc.on_no_data = DRBD_ON_NO_DATA_DEF;
1806 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1807 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1808 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1809 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1810 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1812 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1814 if (!syncer_conf_from_tags(mdev, nlp->tag_list, &sc)) {
1815 retcode = ERR_MANDATORY_TAG;
1819 /* re-sync running */
1820 rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1821 mdev->state.conn == C_SYNC_TARGET ||
1822 mdev->state.conn == C_PAUSED_SYNC_S ||
1823 mdev->state.conn == C_PAUSED_SYNC_T );
1825 if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1826 retcode = ERR_CSUMS_RESYNC_RUNNING;
1830 if (!rsr && sc.csums_alg[0]) {
1831 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1832 if (IS_ERR(csums_tfm)) {
1834 retcode = ERR_CSUMS_ALG;
1838 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1839 retcode = ERR_CSUMS_ALG_ND;
1844 /* online verify running */
1845 ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1848 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1849 retcode = ERR_VERIFY_RUNNING;
1854 if (!ovr && sc.verify_alg[0]) {
1855 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1856 if (IS_ERR(verify_tfm)) {
1858 retcode = ERR_VERIFY_ALG;
1862 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1863 retcode = ERR_VERIFY_ALG_ND;
1868 /* silently ignore cpu mask on UP kernel */
1869 if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1870 err = bitmap_parse(sc.cpu_mask, 32,
1871 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1873 dev_warn(DEV, "bitmap_parse() failed with %d\n", err);
1874 retcode = ERR_CPU_MASK_PARSE;
1879 ERR_IF (sc.rate < 1) sc.rate = 1;
1880 ERR_IF (sc.al_extents < 7) sc.al_extents = 127; /* arbitrary minimum */
1881 #define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
1882 if (sc.al_extents > AL_MAX) {
1883 dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
1884 sc.al_extents = AL_MAX;
1888 /* to avoid spurious errors when configuring minors before configuring
1889 * the minors they depend on: if necessary, first create the minor we
1892 ensure_mdev(sc.after, 1);
1894 /* most sanity checks done, try to assign the new sync-after
1895 * dependency. need to hold the global lock in there,
1896 * to avoid a race in the dependency loop check. */
1897 retcode = drbd_alter_sa(mdev, sc.after);
1898 if (retcode != NO_ERROR)
1901 fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1902 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1903 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1905 dev_err(DEV, "kmalloc of fifo_buffer failed");
1906 retcode = ERR_NOMEM;
1911 /* ok, assign the rest of it as well.
1912 * lock against receive_SyncParam() */
1913 spin_lock(&mdev->peer_seq_lock);
1914 mdev->sync_conf = sc;
1917 crypto_free_hash(mdev->csums_tfm);
1918 mdev->csums_tfm = csums_tfm;
1923 crypto_free_hash(mdev->verify_tfm);
1924 mdev->verify_tfm = verify_tfm;
1928 if (fifo_size != mdev->rs_plan_s.size) {
1929 kfree(mdev->rs_plan_s.values);
1930 mdev->rs_plan_s.values = rs_plan_s;
1931 mdev->rs_plan_s.size = fifo_size;
1932 mdev->rs_planed = 0;
1936 spin_unlock(&mdev->peer_seq_lock);
1938 if (get_ldev(mdev)) {
1939 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1940 drbd_al_shrink(mdev);
1941 err = drbd_check_al_size(mdev);
1942 lc_unlock(mdev->act_log);
1943 wake_up(&mdev->al_wait);
1949 retcode = ERR_NOMEM;
1954 if (mdev->state.conn >= C_CONNECTED)
1955 drbd_send_sync_param(mdev, &sc);
1957 if (!cpumask_equal(mdev->cpu_mask, new_cpu_mask)) {
1958 cpumask_copy(mdev->cpu_mask, new_cpu_mask);
1959 drbd_calc_cpu_mask(mdev);
1960 mdev->receiver.reset_cpu_mask = 1;
1961 mdev->asender.reset_cpu_mask = 1;
1962 mdev->worker.reset_cpu_mask = 1;
1965 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1968 free_cpumask_var(new_cpu_mask);
1969 crypto_free_hash(csums_tfm);
1970 crypto_free_hash(verify_tfm);
1971 reply->ret_code = retcode;
1975 static int drbd_nl_invalidate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
1976 struct drbd_nl_cfg_reply *reply)
1980 /* If there is still bitmap IO pending, probably because of a previous
1981 * resync just being finished, wait for it before requesting a new resync.
1982 * Also wait for it's after_state_ch(). */
1983 drbd_suspend_io(mdev);
1984 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
1985 drbd_flush_workqueue(mdev);
1987 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
1989 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
1990 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
1992 while (retcode == SS_NEED_CONNECTION) {
1993 spin_lock_irq(&mdev->req_lock);
1994 if (mdev->state.conn < C_CONNECTED)
1995 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
1996 spin_unlock_irq(&mdev->req_lock);
1998 if (retcode != SS_NEED_CONNECTION)
2001 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2003 drbd_resume_io(mdev);
2005 reply->ret_code = retcode;
2009 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2013 rv = drbd_bmio_set_n_write(mdev);
2014 drbd_suspend_al(mdev);
2018 static int drbd_nl_invalidate_peer(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2019 struct drbd_nl_cfg_reply *reply)
2023 /* If there is still bitmap IO pending, probably because of a previous
2024 * resync just being finished, wait for it before requesting a new resync.
2025 * Also wait for it's after_state_ch(). */
2026 drbd_suspend_io(mdev);
2027 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2028 drbd_flush_workqueue(mdev);
2030 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
2032 if (retcode < SS_SUCCESS) {
2033 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
2034 /* The peer will get a resync upon connect anyways. Just make that
2035 into a full resync. */
2036 retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
2037 if (retcode >= SS_SUCCESS) {
2038 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
2039 "set_n_write from invalidate_peer",
2040 BM_LOCKED_SET_ALLOWED))
2041 retcode = ERR_IO_MD_DISK;
2044 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
2046 drbd_resume_io(mdev);
2048 reply->ret_code = retcode;
2052 static int drbd_nl_pause_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2053 struct drbd_nl_cfg_reply *reply)
2055 int retcode = NO_ERROR;
2057 if (drbd_request_state(mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2058 retcode = ERR_PAUSE_IS_SET;
2060 reply->ret_code = retcode;
2064 static int drbd_nl_resume_sync(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2065 struct drbd_nl_cfg_reply *reply)
2067 int retcode = NO_ERROR;
2070 if (drbd_request_state(mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2072 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2073 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2074 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2076 retcode = ERR_PAUSE_IS_CLEAR;
2080 reply->ret_code = retcode;
2084 static int drbd_nl_suspend_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2085 struct drbd_nl_cfg_reply *reply)
2087 reply->ret_code = drbd_request_state(mdev, NS(susp, 1));
2092 static int drbd_nl_resume_io(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2093 struct drbd_nl_cfg_reply *reply)
2095 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2096 drbd_uuid_new_current(mdev);
2097 clear_bit(NEW_CUR_UUID, &mdev->flags);
2099 drbd_suspend_io(mdev);
2100 reply->ret_code = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2101 if (reply->ret_code == SS_SUCCESS) {
2102 if (mdev->state.conn < C_CONNECTED)
2104 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2105 tl_restart(mdev, fail_frozen_disk_io);
2107 drbd_resume_io(mdev);
2112 static int drbd_nl_outdate(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2113 struct drbd_nl_cfg_reply *reply)
2115 reply->ret_code = drbd_request_state(mdev, NS(disk, D_OUTDATED));
2119 static int drbd_nl_get_config(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2120 struct drbd_nl_cfg_reply *reply)
2124 tl = reply->tag_list;
2126 if (get_ldev(mdev)) {
2127 tl = disk_conf_to_tags(mdev, &mdev->ldev->dc, tl);
2131 if (get_net_conf(mdev)) {
2132 tl = net_conf_to_tags(mdev, mdev->net_conf, tl);
2135 tl = syncer_conf_to_tags(mdev, &mdev->sync_conf, tl);
2137 put_unaligned(TT_END, tl++); /* Close the tag list */
2139 return (int)((char *)tl - (char *)reply->tag_list);
2142 static int drbd_nl_get_state(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2143 struct drbd_nl_cfg_reply *reply)
2145 unsigned short *tl = reply->tag_list;
2146 union drbd_state s = mdev->state;
2147 unsigned long rs_left;
2150 tl = get_state_to_tags(mdev, (struct get_state *)&s, tl);
2152 /* no local ref, no bitmap, no syncer progress. */
2153 if (s.conn >= C_SYNC_SOURCE && s.conn <= C_PAUSED_SYNC_T) {
2154 if (get_ldev(mdev)) {
2155 drbd_get_syncer_progress(mdev, &rs_left, &res);
2156 tl = tl_add_int(tl, T_sync_progress, &res);
2160 put_unaligned(TT_END, tl++); /* Close the tag list */
2162 return (int)((char *)tl - (char *)reply->tag_list);
2165 static int drbd_nl_get_uuids(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2166 struct drbd_nl_cfg_reply *reply)
2170 tl = reply->tag_list;
2172 if (get_ldev(mdev)) {
2173 tl = tl_add_blob(tl, T_uuids, mdev->ldev->md.uuid, UI_SIZE*sizeof(u64));
2174 tl = tl_add_int(tl, T_uuids_flags, &mdev->ldev->md.flags);
2177 put_unaligned(TT_END, tl++); /* Close the tag list */
2179 return (int)((char *)tl - (char *)reply->tag_list);
2183 * drbd_nl_get_timeout_flag() - Used by drbdsetup to find out which timeout value to use
2184 * @mdev: DRBD device.
2185 * @nlp: Netlink/connector packet from drbdsetup
2186 * @reply: Reply packet for drbdsetup
2188 static int drbd_nl_get_timeout_flag(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2189 struct drbd_nl_cfg_reply *reply)
2194 tl = reply->tag_list;
2196 rv = mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2197 test_bit(USE_DEGR_WFC_T, &mdev->flags) ? UT_DEGRADED : UT_DEFAULT;
2199 tl = tl_add_blob(tl, T_use_degraded, &rv, sizeof(rv));
2200 put_unaligned(TT_END, tl++); /* Close the tag list */
2202 return (int)((char *)tl - (char *)reply->tag_list);
2205 static int drbd_nl_start_ov(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2206 struct drbd_nl_cfg_reply *reply)
2208 /* default to resume from last known position, if possible */
2209 struct start_ov args =
2210 { .start_sector = mdev->ov_start_sector };
2212 if (!start_ov_from_tags(mdev, nlp->tag_list, &args)) {
2213 reply->ret_code = ERR_MANDATORY_TAG;
2217 /* If there is still bitmap IO pending, e.g. previous resync or verify
2218 * just being finished, wait for it before requesting a new resync. */
2219 drbd_suspend_io(mdev);
2220 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2222 /* w_make_ov_request expects position to be aligned */
2223 mdev->ov_start_sector = args.start_sector & ~BM_SECT_PER_BIT;
2224 reply->ret_code = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2225 drbd_resume_io(mdev);
2230 static int drbd_nl_new_c_uuid(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp,
2231 struct drbd_nl_cfg_reply *reply)
2233 int retcode = NO_ERROR;
2234 int skip_initial_sync = 0;
2237 struct new_c_uuid args;
2239 memset(&args, 0, sizeof(struct new_c_uuid));
2240 if (!new_c_uuid_from_tags(mdev, nlp->tag_list, &args)) {
2241 reply->ret_code = ERR_MANDATORY_TAG;
2245 mutex_lock(&mdev->state_mutex); /* Protects us against serialized state changes. */
2247 if (!get_ldev(mdev)) {
2248 retcode = ERR_NO_DISK;
2252 /* this is "skip initial sync", assume to be clean */
2253 if (mdev->state.conn == C_CONNECTED && mdev->agreed_pro_version >= 90 &&
2254 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2255 dev_info(DEV, "Preparing to skip initial sync\n");
2256 skip_initial_sync = 1;
2257 } else if (mdev->state.conn != C_STANDALONE) {
2258 retcode = ERR_CONNECTED;
2262 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2263 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2265 if (args.clear_bm) {
2266 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2267 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2269 dev_err(DEV, "Writing bitmap failed with %d\n",err);
2270 retcode = ERR_IO_MD_DISK;
2272 if (skip_initial_sync) {
2273 drbd_send_uuids_skip_initial_sync(mdev);
2274 _drbd_uuid_set(mdev, UI_BITMAP, 0);
2275 drbd_print_uuids(mdev, "cleared bitmap UUID");
2276 spin_lock_irq(&mdev->req_lock);
2277 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2279 spin_unlock_irq(&mdev->req_lock);
2287 mutex_unlock(&mdev->state_mutex);
2289 reply->ret_code = retcode;
2293 struct cn_handler_struct {
2294 int (*function)(struct drbd_conf *,
2295 struct drbd_nl_cfg_req *,
2296 struct drbd_nl_cfg_reply *);
2297 int reply_body_size;
2300 static struct cn_handler_struct cnd_table[] = {
2301 [ P_primary ] = { &drbd_nl_primary, 0 },
2302 [ P_secondary ] = { &drbd_nl_secondary, 0 },
2303 [ P_disk_conf ] = { &drbd_nl_disk_conf, 0 },
2304 [ P_detach ] = { &drbd_nl_detach, 0 },
2305 [ P_net_conf ] = { &drbd_nl_net_conf, 0 },
2306 [ P_disconnect ] = { &drbd_nl_disconnect, 0 },
2307 [ P_resize ] = { &drbd_nl_resize, 0 },
2308 [ P_syncer_conf ] = { &drbd_nl_syncer_conf, 0 },
2309 [ P_invalidate ] = { &drbd_nl_invalidate, 0 },
2310 [ P_invalidate_peer ] = { &drbd_nl_invalidate_peer, 0 },
2311 [ P_pause_sync ] = { &drbd_nl_pause_sync, 0 },
2312 [ P_resume_sync ] = { &drbd_nl_resume_sync, 0 },
2313 [ P_suspend_io ] = { &drbd_nl_suspend_io, 0 },
2314 [ P_resume_io ] = { &drbd_nl_resume_io, 0 },
2315 [ P_outdate ] = { &drbd_nl_outdate, 0 },
2316 [ P_get_config ] = { &drbd_nl_get_config,
2317 sizeof(struct syncer_conf_tag_len_struct) +
2318 sizeof(struct disk_conf_tag_len_struct) +
2319 sizeof(struct net_conf_tag_len_struct) },
2320 [ P_get_state ] = { &drbd_nl_get_state,
2321 sizeof(struct get_state_tag_len_struct) +
2322 sizeof(struct sync_progress_tag_len_struct) },
2323 [ P_get_uuids ] = { &drbd_nl_get_uuids,
2324 sizeof(struct get_uuids_tag_len_struct) },
2325 [ P_get_timeout_flag ] = { &drbd_nl_get_timeout_flag,
2326 sizeof(struct get_timeout_flag_tag_len_struct)},
2327 [ P_start_ov ] = { &drbd_nl_start_ov, 0 },
2328 [ P_new_c_uuid ] = { &drbd_nl_new_c_uuid, 0 },
2331 static void drbd_connector_callback(struct cn_msg *req, struct netlink_skb_parms *nsp)
2333 struct drbd_nl_cfg_req *nlp = (struct drbd_nl_cfg_req *)req->data;
2334 struct cn_handler_struct *cm;
2335 struct cn_msg *cn_reply;
2336 struct drbd_nl_cfg_reply *reply;
2337 struct drbd_conf *mdev;
2339 int reply_size = sizeof(struct cn_msg)
2340 + sizeof(struct drbd_nl_cfg_reply)
2341 + sizeof(short int);
2343 if (!try_module_get(THIS_MODULE)) {
2344 printk(KERN_ERR "drbd: try_module_get() failed!\n");
2348 if (!capable(CAP_SYS_ADMIN)) {
2353 mdev = ensure_mdev(nlp->drbd_minor,
2354 (nlp->flags & DRBD_NL_CREATE_DEVICE));
2356 retcode = ERR_MINOR_INVALID;
2360 if (nlp->packet_type >= P_nl_after_last_packet ||
2361 nlp->packet_type == P_return_code_only) {
2362 retcode = ERR_PACKET_NR;
2366 cm = cnd_table + nlp->packet_type;
2368 /* This may happen if packet number is 0: */
2369 if (cm->function == NULL) {
2370 retcode = ERR_PACKET_NR;
2374 reply_size += cm->reply_body_size;
2376 /* allocation not in the IO path, cqueue thread context */
2377 cn_reply = kzalloc(reply_size, GFP_KERNEL);
2379 retcode = ERR_NOMEM;
2382 reply = (struct drbd_nl_cfg_reply *) cn_reply->data;
2384 reply->packet_type =
2385 cm->reply_body_size ? nlp->packet_type : P_return_code_only;
2386 reply->minor = nlp->drbd_minor;
2387 reply->ret_code = NO_ERROR; /* Might by modified by cm->function. */
2388 /* reply->tag_list; might be modified by cm->function. */
2390 rr = cm->function(mdev, nlp, reply);
2392 cn_reply->id = req->id;
2393 cn_reply->seq = req->seq;
2394 cn_reply->ack = req->ack + 1;
2395 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) + rr;
2396 cn_reply->flags = 0;
2398 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_KERNEL);
2399 if (rr && rr != -ESRCH)
2400 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);
2403 module_put(THIS_MODULE);
2406 drbd_nl_send_reply(req, retcode);
2407 module_put(THIS_MODULE);
2410 static atomic_t drbd_nl_seq = ATOMIC_INIT(2); /* two. */
2412 static unsigned short *
2413 __tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data,
2414 unsigned short len, int nul_terminated)
2416 unsigned short l = tag_descriptions[tag_number(tag)].max_len;
2417 len = (len < l) ? len : l;
2418 put_unaligned(tag, tl++);
2419 put_unaligned(len, tl++);
2420 memcpy(tl, data, len);
2421 tl = (unsigned short*)((char*)tl + len);
2423 *((char*)tl - 1) = 0;
2427 static unsigned short *
2428 tl_add_blob(unsigned short *tl, enum drbd_tags tag, const void *data, int len)
2430 return __tl_add_blob(tl, tag, data, len, 0);
2433 static unsigned short *
2434 tl_add_str(unsigned short *tl, enum drbd_tags tag, const char *str)
2436 return __tl_add_blob(tl, tag, str, strlen(str)+1, 0);
2439 static unsigned short *
2440 tl_add_int(unsigned short *tl, enum drbd_tags tag, const void *val)
2442 put_unaligned(tag, tl++);
2443 switch(tag_type(tag)) {
2445 put_unaligned(sizeof(int), tl++);
2446 put_unaligned(*(int *)val, (int *)tl);
2447 tl = (unsigned short*)((char*)tl+sizeof(int));
2450 put_unaligned(sizeof(u64), tl++);
2451 put_unaligned(*(u64 *)val, (u64 *)tl);
2452 tl = (unsigned short*)((char*)tl+sizeof(u64));
2455 /* someone did something stupid. */
2461 void drbd_bcast_state(struct drbd_conf *mdev, union drbd_state state)
2463 char buffer[sizeof(struct cn_msg)+
2464 sizeof(struct drbd_nl_cfg_reply)+
2465 sizeof(struct get_state_tag_len_struct)+
2467 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2468 struct drbd_nl_cfg_reply *reply =
2469 (struct drbd_nl_cfg_reply *)cn_reply->data;
2470 unsigned short *tl = reply->tag_list;
2472 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2474 tl = get_state_to_tags(mdev, (struct get_state *)&state, tl);
2476 put_unaligned(TT_END, tl++); /* Close the tag list */
2478 cn_reply->id.idx = CN_IDX_DRBD;
2479 cn_reply->id.val = CN_VAL_DRBD;
2481 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2482 cn_reply->ack = 0; /* not used here. */
2483 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2484 (int)((char *)tl - (char *)reply->tag_list);
2485 cn_reply->flags = 0;
2487 reply->packet_type = P_get_state;
2488 reply->minor = mdev_to_minor(mdev);
2489 reply->ret_code = NO_ERROR;
2491 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2494 void drbd_bcast_ev_helper(struct drbd_conf *mdev, char *helper_name)
2496 char buffer[sizeof(struct cn_msg)+
2497 sizeof(struct drbd_nl_cfg_reply)+
2498 sizeof(struct call_helper_tag_len_struct)+
2500 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2501 struct drbd_nl_cfg_reply *reply =
2502 (struct drbd_nl_cfg_reply *)cn_reply->data;
2503 unsigned short *tl = reply->tag_list;
2505 /* dev_warn(DEV, "drbd_bcast_state() got called\n"); */
2507 tl = tl_add_str(tl, T_helper, helper_name);
2508 put_unaligned(TT_END, tl++); /* Close the tag list */
2510 cn_reply->id.idx = CN_IDX_DRBD;
2511 cn_reply->id.val = CN_VAL_DRBD;
2513 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2514 cn_reply->ack = 0; /* not used here. */
2515 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2516 (int)((char *)tl - (char *)reply->tag_list);
2517 cn_reply->flags = 0;
2519 reply->packet_type = P_call_helper;
2520 reply->minor = mdev_to_minor(mdev);
2521 reply->ret_code = NO_ERROR;
2523 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2526 void drbd_bcast_ee(struct drbd_conf *mdev,
2527 const char *reason, const int dgs,
2528 const char* seen_hash, const char* calc_hash,
2529 const struct drbd_epoch_entry* e)
2531 struct cn_msg *cn_reply;
2532 struct drbd_nl_cfg_reply *reply;
2539 if (!reason || !reason[0])
2542 /* apparently we have to memcpy twice, first to prepare the data for the
2543 * struct cn_msg, then within cn_netlink_send from the cn_msg to the
2545 /* receiver thread context, which is not in the writeout path (of this node),
2546 * but may be in the writeout path of the _other_ node.
2547 * GFP_NOIO to avoid potential "distributed deadlock". */
2549 sizeof(struct cn_msg)+
2550 sizeof(struct drbd_nl_cfg_reply)+
2551 sizeof(struct dump_ee_tag_len_struct)+
2556 dev_err(DEV, "could not kmalloc buffer for drbd_bcast_ee, sector %llu, size %u\n",
2557 (unsigned long long)e->sector, e->size);
2561 reply = (struct drbd_nl_cfg_reply*)cn_reply->data;
2562 tl = reply->tag_list;
2564 tl = tl_add_str(tl, T_dump_ee_reason, reason);
2565 tl = tl_add_blob(tl, T_seen_digest, seen_hash, dgs);
2566 tl = tl_add_blob(tl, T_calc_digest, calc_hash, dgs);
2567 tl = tl_add_int(tl, T_ee_sector, &e->sector);
2568 tl = tl_add_int(tl, T_ee_block_id, &e->block_id);
2570 /* dump the first 32k */
2571 len = min_t(unsigned, e->size, 32 << 10);
2572 put_unaligned(T_ee_data, tl++);
2573 put_unaligned(len, tl++);
2576 page_chain_for_each(page) {
2577 void *d = kmap_atomic(page);
2578 unsigned l = min_t(unsigned, len, PAGE_SIZE);
2581 tl = (unsigned short*)((char*)tl + l);
2586 put_unaligned(TT_END, tl++); /* Close the tag list */
2588 cn_reply->id.idx = CN_IDX_DRBD;
2589 cn_reply->id.val = CN_VAL_DRBD;
2591 cn_reply->seq = atomic_add_return(1,&drbd_nl_seq);
2592 cn_reply->ack = 0; // not used here.
2593 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2594 (int)((char*)tl - (char*)reply->tag_list);
2595 cn_reply->flags = 0;
2597 reply->packet_type = P_dump_ee;
2598 reply->minor = mdev_to_minor(mdev);
2599 reply->ret_code = NO_ERROR;
2601 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2605 void drbd_bcast_sync_progress(struct drbd_conf *mdev)
2607 char buffer[sizeof(struct cn_msg)+
2608 sizeof(struct drbd_nl_cfg_reply)+
2609 sizeof(struct sync_progress_tag_len_struct)+
2611 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2612 struct drbd_nl_cfg_reply *reply =
2613 (struct drbd_nl_cfg_reply *)cn_reply->data;
2614 unsigned short *tl = reply->tag_list;
2615 unsigned long rs_left;
2618 /* no local ref, no bitmap, no syncer progress, no broadcast. */
2619 if (!get_ldev(mdev))
2621 drbd_get_syncer_progress(mdev, &rs_left, &res);
2624 tl = tl_add_int(tl, T_sync_progress, &res);
2625 put_unaligned(TT_END, tl++); /* Close the tag list */
2627 cn_reply->id.idx = CN_IDX_DRBD;
2628 cn_reply->id.val = CN_VAL_DRBD;
2630 cn_reply->seq = atomic_add_return(1, &drbd_nl_seq);
2631 cn_reply->ack = 0; /* not used here. */
2632 cn_reply->len = sizeof(struct drbd_nl_cfg_reply) +
2633 (int)((char *)tl - (char *)reply->tag_list);
2634 cn_reply->flags = 0;
2636 reply->packet_type = P_sync_progress;
2637 reply->minor = mdev_to_minor(mdev);
2638 reply->ret_code = NO_ERROR;
2640 cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2643 int __init drbd_nl_init(void)
2645 static struct cb_id cn_id_drbd;
2648 cn_id_drbd.val = CN_VAL_DRBD;
2650 cn_id_drbd.idx = cn_idx;
2651 err = cn_add_callback(&cn_id_drbd, "cn_drbd", &drbd_connector_callback);
2654 cn_idx = (cn_idx + CN_IDX_STEP);
2658 printk(KERN_ERR "drbd: cn_drbd failed to register\n");
2665 void drbd_nl_cleanup(void)
2667 static struct cb_id cn_id_drbd;
2669 cn_id_drbd.idx = cn_idx;
2670 cn_id_drbd.val = CN_VAL_DRBD;
2672 cn_del_callback(&cn_id_drbd);
2675 void drbd_nl_send_reply(struct cn_msg *req, int ret_code)
2677 char buffer[sizeof(struct cn_msg)+sizeof(struct drbd_nl_cfg_reply)];
2678 struct cn_msg *cn_reply = (struct cn_msg *) buffer;
2679 struct drbd_nl_cfg_reply *reply =
2680 (struct drbd_nl_cfg_reply *)cn_reply->data;
2683 memset(buffer, 0, sizeof(buffer));
2684 cn_reply->id = req->id;
2686 cn_reply->seq = req->seq;
2687 cn_reply->ack = req->ack + 1;
2688 cn_reply->len = sizeof(struct drbd_nl_cfg_reply);
2689 cn_reply->flags = 0;
2691 reply->packet_type = P_return_code_only;
2692 reply->minor = ((struct drbd_nl_cfg_req *)req->data)->drbd_minor;
2693 reply->ret_code = ret_code;
2695 rr = cn_netlink_send(cn_reply, CN_IDX_DRBD, GFP_NOIO);
2696 if (rr && rr != -ESRCH)
2697 printk(KERN_INFO "drbd: cn_netlink_send()=%d\n", rr);