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/blkpg.h>
33 #include <linux/cpumask.h>
36 #include "drbd_wrappers.h"
37 #include <asm/unaligned.h>
38 #include <linux/drbd_limits.h>
39 #include <linux/kthread.h>
41 #include <net/genetlink.h>
44 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
45 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
47 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info);
48 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info);
50 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info);
51 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info);
53 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
54 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
55 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
56 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
57 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
58 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
59 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
60 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
61 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
62 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
63 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
64 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
65 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
66 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
67 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
68 int drbd_adm_syncer(struct sk_buff *skb, struct genl_info *info);
69 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
70 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
72 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
74 #include <linux/drbd_genl_api.h>
75 #include <linux/genl_magic_func.h>
77 /* used blkdev_get_by_path, to claim our meta data device(s) */
78 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
80 /* Configuration is strictly serialized, because generic netlink message
81 * processing is strictly serialized by the genl_lock().
82 * Which means we can use one static global drbd_config_context struct.
84 static struct drbd_config_context {
85 /* assigned from drbd_genlmsghdr */
87 /* assigned from request attributes, if present */
89 #define VOLUME_UNSPECIFIED (-1U)
90 /* pointer into the request skb,
91 * limited lifetime! */
95 struct sk_buff *reply_skb;
96 /* pointer into reply buffer */
97 struct drbd_genlmsghdr *reply_dh;
98 /* resolved from attributes, if possible */
99 struct drbd_conf *mdev;
100 struct drbd_tconn *tconn;
103 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
105 genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
106 if (genlmsg_reply(skb, info))
107 printk(KERN_ERR "drbd: error sending genl reply\n");
110 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
111 * reason it could fail was no space in skb, and there are 4k available. */
112 int drbd_msg_put_info(const char *info)
114 struct sk_buff *skb = adm_ctx.reply_skb;
118 if (!info || !info[0])
121 nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
125 err = nla_put_string(skb, T_info_text, info);
127 nla_nest_cancel(skb, nla);
130 nla_nest_end(skb, nla);
134 /* This would be a good candidate for a "pre_doit" hook,
135 * and per-family private info->pointers.
136 * But we need to stay compatible with older kernels.
137 * If it returns successfully, adm_ctx members are valid.
139 #define DRBD_ADM_NEED_MINOR 1
140 #define DRBD_ADM_NEED_CONN 2
141 static int drbd_adm_prepare(struct sk_buff *skb, struct genl_info *info,
144 struct drbd_genlmsghdr *d_in = info->userhdr;
145 const u8 cmd = info->genlhdr->cmd;
148 memset(&adm_ctx, 0, sizeof(adm_ctx));
150 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
151 if (cmd != DRBD_ADM_GET_STATUS
152 && security_netlink_recv(skb, CAP_SYS_ADMIN))
155 adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
156 if (!adm_ctx.reply_skb)
159 adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
160 info, &drbd_genl_family, 0, cmd);
161 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
163 if (!adm_ctx.reply_dh)
166 adm_ctx.reply_dh->minor = d_in->minor;
167 adm_ctx.reply_dh->ret_code = NO_ERROR;
169 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
171 /* parse and validate only */
172 err = drbd_cfg_context_from_attrs(NULL, info->attrs);
176 /* It was present, and valid,
177 * copy it over to the reply skb. */
178 err = nla_put_nohdr(adm_ctx.reply_skb,
179 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
180 info->attrs[DRBD_NLA_CFG_CONTEXT]);
184 /* and assign stuff to the global adm_ctx */
185 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
186 adm_ctx.volume = nla ? nla_get_u32(nla) : VOLUME_UNSPECIFIED;
187 nla = nested_attr_tb[__nla_type(T_ctx_conn_name)];
189 adm_ctx.conn_name = nla_data(nla);
191 adm_ctx.volume = VOLUME_UNSPECIFIED;
193 adm_ctx.minor = d_in->minor;
194 adm_ctx.mdev = minor_to_mdev(d_in->minor);
195 adm_ctx.tconn = conn_by_name(adm_ctx.conn_name);
197 if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
198 drbd_msg_put_info("unknown minor");
199 return ERR_MINOR_INVALID;
201 if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_CONN)) {
202 drbd_msg_put_info("unknown connection");
203 return ERR_INVALID_REQUEST;
206 /* some more paranoia, if the request was over-determined */
208 adm_ctx.volume != VOLUME_UNSPECIFIED &&
209 adm_ctx.volume != adm_ctx.mdev->vnr) {
210 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
211 adm_ctx.minor, adm_ctx.volume,
212 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
213 drbd_msg_put_info("over-determined configuration context mismatch");
214 return ERR_INVALID_REQUEST;
216 if (adm_ctx.mdev && adm_ctx.tconn &&
217 adm_ctx.mdev->tconn != adm_ctx.tconn) {
218 pr_warning("request: minor=%u, conn=%s; but that minor belongs to connection %s\n",
219 adm_ctx.minor, adm_ctx.conn_name, adm_ctx.mdev->tconn->name);
220 drbd_msg_put_info("over-determined configuration context mismatch");
221 return ERR_INVALID_REQUEST;
223 if (adm_ctx.mdev && !adm_ctx.tconn)
224 adm_ctx.tconn = adm_ctx.mdev->tconn;
228 nlmsg_free(adm_ctx.reply_skb);
229 adm_ctx.reply_skb = NULL;
233 static int drbd_adm_finish(struct genl_info *info, int retcode)
236 const char *conn_name = NULL;
238 if (!adm_ctx.reply_skb)
241 adm_ctx.reply_dh->ret_code = retcode;
243 nla = info->attrs[DRBD_NLA_CFG_CONTEXT];
245 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
247 conn_name = nla_data(nla);
250 drbd_adm_send_reply(adm_ctx.reply_skb, info);
254 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
256 char *envp[] = { "HOME=/",
258 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
259 NULL, /* Will be set to address family */
260 NULL, /* Will be set to address */
262 char mb[12], af[20], ad[60], *afs;
263 char *argv[] = {usermode_helper, cmd, mb, NULL };
267 snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
269 if (get_net_conf(mdev->tconn)) {
270 switch (((struct sockaddr *)mdev->tconn->net_conf->peer_addr)->sa_family) {
273 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI6",
274 &((struct sockaddr_in6 *)mdev->tconn->net_conf->peer_addr)->sin6_addr);
278 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
279 &((struct sockaddr_in *)mdev->tconn->net_conf->peer_addr)->sin_addr);
283 snprintf(ad, 60, "DRBD_PEER_ADDRESS=%pI4",
284 &((struct sockaddr_in *)mdev->tconn->net_conf->peer_addr)->sin_addr);
286 snprintf(af, 20, "DRBD_PEER_AF=%s", afs);
289 put_net_conf(mdev->tconn);
292 /* The helper may take some time.
293 * write out any unsynced meta data changes now */
296 dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
297 sib.sib_reason = SIB_HELPER_PRE;
298 sib.helper_name = cmd;
299 drbd_bcast_event(mdev, &sib);
300 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
302 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
303 usermode_helper, cmd, mb,
304 (ret >> 8) & 0xff, ret);
306 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
307 usermode_helper, cmd, mb,
308 (ret >> 8) & 0xff, ret);
309 sib.sib_reason = SIB_HELPER_POST;
310 sib.helper_exit_code = ret;
311 drbd_bcast_event(mdev, &sib);
313 if (ret < 0) /* Ignore any ERRNOs we got. */
319 enum drbd_disk_state drbd_try_outdate_peer(struct drbd_conf *mdev)
323 enum drbd_disk_state nps;
324 enum drbd_fencing_p fp;
326 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
328 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
329 fp = mdev->ldev->dc.fencing;
332 dev_warn(DEV, "Not fencing peer, I'm not even Consistent myself.\n");
333 nps = mdev->state.pdsk;
337 r = drbd_khelper(mdev, "fence-peer");
339 switch ((r>>8) & 0xff) {
340 case 3: /* peer is inconsistent */
341 ex_to_string = "peer is inconsistent or worse";
342 nps = D_INCONSISTENT;
344 case 4: /* peer got outdated, or was already outdated */
345 ex_to_string = "peer was fenced";
348 case 5: /* peer was down */
349 if (mdev->state.disk == D_UP_TO_DATE) {
350 /* we will(have) create(d) a new UUID anyways... */
351 ex_to_string = "peer is unreachable, assumed to be dead";
354 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
355 nps = mdev->state.pdsk;
358 case 6: /* Peer is primary, voluntarily outdate myself.
359 * This is useful when an unconnected R_SECONDARY is asked to
360 * become R_PRIMARY, but finds the other peer being active. */
361 ex_to_string = "peer is active";
362 dev_warn(DEV, "Peer is primary, outdating myself.\n");
364 _drbd_request_state(mdev, NS(disk, D_OUTDATED), CS_WAIT_COMPLETE);
367 if (fp != FP_STONITH)
368 dev_err(DEV, "fence-peer() = 7 && fencing != Stonith !!!\n");
369 ex_to_string = "peer was stonithed";
373 /* The script is broken ... */
375 dev_err(DEV, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
379 dev_info(DEV, "fence-peer helper returned %d (%s)\n",
380 (r>>8) & 0xff, ex_to_string);
383 if (mdev->state.susp_fen && nps >= D_UNKNOWN) {
384 /* The handler was not successful... unfreeze here, the
385 state engine can not unfreeze... */
386 _drbd_request_state(mdev, NS(susp_fen, 0), CS_VERBOSE);
392 static int _try_outdate_peer_async(void *data)
394 struct drbd_conf *mdev = (struct drbd_conf *)data;
395 enum drbd_disk_state nps;
398 nps = drbd_try_outdate_peer(mdev);
401 drbd_request_state(mdev, NS(pdsk, nps));
402 here, because we might were able to re-establish the connection
403 in the meantime. This can only partially be solved in the state's
404 engine is_valid_state() and is_valid_state_transition()
407 nps can be D_INCONSISTENT, D_OUTDATED or D_UNKNOWN.
408 pdsk == D_INCONSISTENT while conn >= C_CONNECTED is valid,
409 therefore we have to have the pre state change check here.
411 spin_lock_irq(&mdev->tconn->req_lock);
413 if (ns.conn < C_WF_REPORT_PARAMS) {
415 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
417 spin_unlock_irq(&mdev->tconn->req_lock);
422 void drbd_try_outdate_peer_async(struct drbd_conf *mdev)
424 struct task_struct *opa;
426 opa = kthread_run(_try_outdate_peer_async, mdev, "drbd%d_a_helper", mdev_to_minor(mdev));
428 dev_err(DEV, "out of mem, failed to invoke fence-peer helper\n");
432 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
434 const int max_tries = 4;
435 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
438 union drbd_state mask, val;
439 enum drbd_disk_state nps;
441 if (new_role == R_PRIMARY)
442 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
444 mutex_lock(mdev->state_mutex);
446 mask.i = 0; mask.role = R_MASK;
447 val.i = 0; val.role = new_role;
449 while (try++ < max_tries) {
450 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
452 /* in case we first succeeded to outdate,
453 * but now suddenly could establish a connection */
454 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
460 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
461 (mdev->state.disk < D_UP_TO_DATE &&
462 mdev->state.disk >= D_INCONSISTENT)) {
464 val.disk = D_UP_TO_DATE;
469 if (rv == SS_NO_UP_TO_DATE_DISK &&
470 mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
471 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
472 nps = drbd_try_outdate_peer(mdev);
474 if (nps == D_OUTDATED || nps == D_INCONSISTENT) {
475 val.disk = D_UP_TO_DATE;
485 if (rv == SS_NOTHING_TO_DO)
487 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
488 nps = drbd_try_outdate_peer(mdev);
490 if (force && nps > D_OUTDATED) {
491 dev_warn(DEV, "Forced into split brain situation!\n");
500 if (rv == SS_TWO_PRIMARIES) {
501 /* Maybe the peer is detected as dead very soon...
502 retry at most once more in this case. */
503 schedule_timeout_interruptible((mdev->tconn->net_conf->ping_timeo+1)*HZ/10);
508 if (rv < SS_SUCCESS) {
509 rv = _drbd_request_state(mdev, mask, val,
510 CS_VERBOSE + CS_WAIT_COMPLETE);
521 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
523 /* Wait until nothing is on the fly :) */
524 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
526 if (new_role == R_SECONDARY) {
527 set_disk_ro(mdev->vdisk, true);
528 if (get_ldev(mdev)) {
529 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
533 if (get_net_conf(mdev->tconn)) {
534 mdev->tconn->net_conf->want_lose = 0;
535 put_net_conf(mdev->tconn);
537 set_disk_ro(mdev->vdisk, false);
538 if (get_ldev(mdev)) {
539 if (((mdev->state.conn < C_CONNECTED ||
540 mdev->state.pdsk <= D_FAILED)
541 && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
542 drbd_uuid_new_current(mdev);
544 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
549 /* writeout of activity log covered areas of the bitmap
550 * to stable storage done in after state change already */
552 if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
553 /* if this was forced, we should consider sync */
555 drbd_send_uuids(mdev);
556 drbd_send_state(mdev);
561 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
563 mutex_unlock(mdev->state_mutex);
567 static const char *from_attrs_err_to_txt(int err)
569 return err == -ENOMSG ? "required attribute missing" :
570 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
571 "invalid attribute value";
574 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
576 struct set_role_parms parms;
578 enum drbd_ret_code retcode;
580 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
581 if (!adm_ctx.reply_skb)
583 if (retcode != NO_ERROR)
586 memset(&parms, 0, sizeof(parms));
587 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
588 err = set_role_parms_from_attrs(&parms, info->attrs);
590 retcode = ERR_MANDATORY_TAG;
591 drbd_msg_put_info(from_attrs_err_to_txt(err));
596 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
597 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
599 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
601 drbd_adm_finish(info, retcode);
605 /* initializes the md.*_offset members, so we are able to find
606 * the on disk meta data */
607 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
608 struct drbd_backing_dev *bdev)
610 sector_t md_size_sect = 0;
611 switch (bdev->dc.meta_dev_idx) {
613 /* v07 style fixed size indexed meta data */
614 bdev->md.md_size_sect = MD_RESERVED_SECT;
615 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
616 bdev->md.al_offset = MD_AL_OFFSET;
617 bdev->md.bm_offset = MD_BM_OFFSET;
619 case DRBD_MD_INDEX_FLEX_EXT:
620 /* just occupy the full device; unit: sectors */
621 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
622 bdev->md.md_offset = 0;
623 bdev->md.al_offset = MD_AL_OFFSET;
624 bdev->md.bm_offset = MD_BM_OFFSET;
626 case DRBD_MD_INDEX_INTERNAL:
627 case DRBD_MD_INDEX_FLEX_INT:
628 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
629 /* al size is still fixed */
630 bdev->md.al_offset = -MD_AL_SECTORS;
631 /* we need (slightly less than) ~ this much bitmap sectors: */
632 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
633 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
634 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
635 md_size_sect = ALIGN(md_size_sect, 8);
637 /* plus the "drbd meta data super block",
638 * and the activity log; */
639 md_size_sect += MD_BM_OFFSET;
641 bdev->md.md_size_sect = md_size_sect;
642 /* bitmap offset is adjusted by 'super' block size */
643 bdev->md.bm_offset = -md_size_sect + MD_AL_OFFSET;
648 /* input size is expected to be in KB */
649 char *ppsize(char *buf, unsigned long long size)
651 /* Needs 9 bytes at max including trailing NUL:
652 * -1ULL ==> "16384 EB" */
653 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
655 while (size >= 10000 && base < sizeof(units)-1) {
657 size = (size >> 10) + !!(size & (1<<9));
660 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
665 /* there is still a theoretical deadlock when called from receiver
666 * on an D_INCONSISTENT R_PRIMARY:
667 * remote READ does inc_ap_bio, receiver would need to receive answer
668 * packet from remote to dec_ap_bio again.
669 * receiver receive_sizes(), comes here,
670 * waits for ap_bio_cnt == 0. -> deadlock.
671 * but this cannot happen, actually, because:
672 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
673 * (not connected, or bad/no disk on peer):
674 * see drbd_fail_request_early, ap_bio_cnt is zero.
675 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
676 * peer may not initiate a resize.
678 /* Note these are not to be confused with
679 * drbd_adm_suspend_io/drbd_adm_resume_io,
680 * which are (sub) state changes triggered by admin (drbdsetup),
681 * and can be long lived.
682 * This changes an mdev->flag, is triggered by drbd internals,
683 * and should be short-lived. */
684 void drbd_suspend_io(struct drbd_conf *mdev)
686 set_bit(SUSPEND_IO, &mdev->flags);
687 if (is_susp(mdev->state))
689 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
692 void drbd_resume_io(struct drbd_conf *mdev)
694 clear_bit(SUSPEND_IO, &mdev->flags);
695 wake_up(&mdev->misc_wait);
699 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
700 * @mdev: DRBD device.
702 * Returns 0 on success, negative return values indicate errors.
703 * You should call drbd_md_sync() after calling this function.
705 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
707 sector_t prev_first_sect, prev_size; /* previous meta location */
712 int md_moved, la_size_changed;
713 enum determine_dev_size rv = unchanged;
716 * application request passes inc_ap_bio,
717 * but then cannot get an AL-reference.
718 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
721 * Suspend IO right here.
722 * still lock the act_log to not trigger ASSERTs there.
724 drbd_suspend_io(mdev);
726 /* no wait necessary anymore, actually we could assert that */
727 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
729 prev_first_sect = drbd_md_first_sector(mdev->ldev);
730 prev_size = mdev->ldev->md.md_size_sect;
731 la_size = mdev->ldev->md.la_size_sect;
733 /* TODO: should only be some assert here, not (re)init... */
734 drbd_md_set_sector_offsets(mdev, mdev->ldev);
736 size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
738 if (drbd_get_capacity(mdev->this_bdev) != size ||
739 drbd_bm_capacity(mdev) != size) {
741 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
743 /* currently there is only one error: ENOMEM! */
744 size = drbd_bm_capacity(mdev)>>1;
746 dev_err(DEV, "OUT OF MEMORY! "
747 "Could not allocate bitmap!\n");
749 dev_err(DEV, "BM resizing failed. "
750 "Leaving size unchanged at size = %lu KB\n",
751 (unsigned long)size);
755 /* racy, see comments above. */
756 drbd_set_my_capacity(mdev, size);
757 mdev->ldev->md.la_size_sect = size;
758 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
759 (unsigned long long)size>>1);
761 if (rv == dev_size_error)
764 la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
766 md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
767 || prev_size != mdev->ldev->md.md_size_sect;
769 if (la_size_changed || md_moved) {
772 drbd_al_shrink(mdev); /* All extents inactive. */
773 dev_info(DEV, "Writing the whole bitmap, %s\n",
774 la_size_changed && md_moved ? "size changed and md moved" :
775 la_size_changed ? "size changed" : "md moved");
776 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
777 err = drbd_bitmap_io(mdev, &drbd_bm_write,
778 "size changed", BM_LOCKED_MASK);
783 drbd_md_mark_dirty(mdev);
791 lc_unlock(mdev->act_log);
792 wake_up(&mdev->al_wait);
793 drbd_resume_io(mdev);
799 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
801 sector_t p_size = mdev->p_size; /* partner's disk size. */
802 sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
803 sector_t m_size; /* my size */
804 sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
807 m_size = drbd_get_max_capacity(bdev);
809 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
810 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
814 if (p_size && m_size) {
815 size = min_t(sector_t, p_size, m_size);
819 if (m_size && m_size < size)
821 if (p_size && p_size < size)
832 dev_err(DEV, "Both nodes diskless!\n");
836 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
837 (unsigned long)u_size>>1, (unsigned long)size>>1);
846 * drbd_check_al_size() - Ensures that the AL is of the right size
847 * @mdev: DRBD device.
849 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
850 * failed, and 0 on success. You should call drbd_md_sync() after you called
853 static int drbd_check_al_size(struct drbd_conf *mdev)
855 struct lru_cache *n, *t;
856 struct lc_element *e;
860 if (!expect(mdev->sync_conf.al_extents >= DRBD_AL_EXTENTS_MIN))
861 mdev->sync_conf.al_extents = DRBD_AL_EXTENTS_MIN;
864 mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
869 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
870 mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
873 dev_err(DEV, "Cannot allocate act_log lru!\n");
876 spin_lock_irq(&mdev->al_lock);
878 for (i = 0; i < t->nr_elements; i++) {
879 e = lc_element_by_index(t, i);
881 dev_err(DEV, "refcnt(%d)==%d\n",
882 e->lc_number, e->refcnt);
888 spin_unlock_irq(&mdev->al_lock);
890 dev_err(DEV, "Activity log still in use!\n");
897 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
901 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
903 struct request_queue * const q = mdev->rq_queue;
904 int max_hw_sectors = max_bio_size >> 9;
905 int max_segments = 0;
907 if (get_ldev_if_state(mdev, D_ATTACHING)) {
908 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
910 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
911 max_segments = mdev->ldev->dc.max_bio_bvecs;
915 blk_queue_logical_block_size(q, 512);
916 blk_queue_max_hw_sectors(q, max_hw_sectors);
917 /* This is the workaround for "bio would need to, but cannot, be split" */
918 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
919 blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
921 if (get_ldev_if_state(mdev, D_ATTACHING)) {
922 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
924 blk_queue_stack_limits(q, b);
926 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
927 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
928 q->backing_dev_info.ra_pages,
929 b->backing_dev_info.ra_pages);
930 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
936 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
938 int now, new, local, peer;
940 now = queue_max_hw_sectors(mdev->rq_queue) << 9;
941 local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
942 peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
944 if (get_ldev_if_state(mdev, D_ATTACHING)) {
945 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
946 mdev->local_max_bio_size = local;
950 /* We may ignore peer limits if the peer is modern enough.
951 Because new from 8.3.8 onwards the peer can use multiple
952 BIOs for a single peer_request */
953 if (mdev->state.conn >= C_CONNECTED) {
954 if (mdev->tconn->agreed_pro_version < 94)
955 peer = mdev->peer_max_bio_size;
956 else if (mdev->tconn->agreed_pro_version == 94)
957 peer = DRBD_MAX_SIZE_H80_PACKET;
958 else /* drbd 8.3.8 onwards */
959 peer = DRBD_MAX_BIO_SIZE;
962 new = min_t(int, local, peer);
964 if (mdev->state.role == R_PRIMARY && new < now)
965 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
968 dev_info(DEV, "max BIO size = %u\n", new);
970 drbd_setup_queue_param(mdev, new);
973 /* serialize deconfig (worker exiting, doing cleanup)
974 * and reconfig (drbdsetup disk, drbdsetup net)
976 * Wait for a potentially exiting worker, then restart it,
977 * or start a new one. Flush any pending work, there may still be an
978 * after_state_change queued.
980 static void conn_reconfig_start(struct drbd_tconn *tconn)
982 wait_event(tconn->ping_wait, !test_and_set_bit(CONFIG_PENDING, &tconn->flags));
983 wait_event(tconn->ping_wait, !test_bit(OBJECT_DYING, &tconn->flags));
984 drbd_thread_start(&tconn->worker);
985 conn_flush_workqueue(tconn);
988 /* if still unconfigured, stops worker again.
989 * if configured now, clears CONFIG_PENDING.
990 * wakes potential waiters */
991 static void conn_reconfig_done(struct drbd_tconn *tconn)
993 spin_lock_irq(&tconn->req_lock);
994 if (conn_all_vols_unconf(tconn)) {
995 set_bit(OBJECT_DYING, &tconn->flags);
996 drbd_thread_stop_nowait(&tconn->worker);
998 clear_bit(CONFIG_PENDING, &tconn->flags);
999 spin_unlock_irq(&tconn->req_lock);
1000 wake_up(&tconn->ping_wait);
1003 /* Make sure IO is suspended before calling this function(). */
1004 static void drbd_suspend_al(struct drbd_conf *mdev)
1008 if (!lc_try_lock(mdev->act_log)) {
1009 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1013 drbd_al_shrink(mdev);
1014 spin_lock_irq(&mdev->tconn->req_lock);
1015 if (mdev->state.conn < C_CONNECTED)
1016 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1017 spin_unlock_irq(&mdev->tconn->req_lock);
1018 lc_unlock(mdev->act_log);
1021 dev_info(DEV, "Suspended AL updates\n");
1024 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1026 struct drbd_conf *mdev;
1028 enum drbd_ret_code retcode;
1029 enum determine_dev_size dd;
1030 sector_t max_possible_sectors;
1031 sector_t min_md_device_sectors;
1032 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1033 struct block_device *bdev;
1034 struct lru_cache *resync_lru = NULL;
1035 union drbd_state ns, os;
1036 enum drbd_state_rv rv;
1037 int cp_discovered = 0;
1039 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1040 if (!adm_ctx.reply_skb)
1042 if (retcode != NO_ERROR)
1045 mdev = adm_ctx.mdev;
1046 conn_reconfig_start(mdev->tconn);
1048 /* if you want to reconfigure, please tear down first */
1049 if (mdev->state.disk > D_DISKLESS) {
1050 retcode = ERR_DISK_CONFIGURED;
1053 /* It may just now have detached because of IO error. Make sure
1054 * drbd_ldev_destroy is done already, we may end up here very fast,
1055 * e.g. if someone calls attach from the on-io-error handler,
1056 * to realize a "hot spare" feature (not that I'd recommend that) */
1057 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1059 /* allocation not in the IO path, drbdsetup context */
1060 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1062 retcode = ERR_NOMEM;
1066 nbc->dc.disk_size = DRBD_DISK_SIZE_SECT_DEF;
1067 nbc->dc.on_io_error = DRBD_ON_IO_ERROR_DEF;
1068 nbc->dc.fencing = DRBD_FENCING_DEF;
1069 nbc->dc.max_bio_bvecs = DRBD_MAX_BIO_BVECS_DEF;
1071 err = disk_conf_from_attrs(&nbc->dc, info->attrs);
1073 retcode = ERR_MANDATORY_TAG;
1074 drbd_msg_put_info(from_attrs_err_to_txt(err));
1078 if ((int)nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1079 retcode = ERR_MD_IDX_INVALID;
1083 if (get_net_conf(mdev->tconn)) {
1084 int prot = mdev->tconn->net_conf->wire_protocol;
1085 put_net_conf(mdev->tconn);
1086 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
1087 retcode = ERR_STONITH_AND_PROT_A;
1092 bdev = blkdev_get_by_path(nbc->dc.backing_dev,
1093 FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1095 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
1097 retcode = ERR_OPEN_DISK;
1100 nbc->backing_bdev = bdev;
1103 * meta_dev_idx >= 0: external fixed size, possibly multiple
1104 * drbd sharing one meta device. TODO in that case, paranoia
1105 * check that [md_bdev, meta_dev_idx] is not yet used by some
1106 * other drbd minor! (if you use drbd.conf + drbdadm, that
1107 * should check it for you already; but if you don't, or
1108 * someone fooled it, we need to double check here)
1110 bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1111 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1112 ((int)nbc->dc.meta_dev_idx < 0) ?
1113 (void *)mdev : (void *)drbd_m_holder);
1115 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1117 retcode = ERR_OPEN_MD_DISK;
1120 nbc->md_bdev = bdev;
1122 if ((nbc->backing_bdev == nbc->md_bdev) !=
1123 (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1124 nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1125 retcode = ERR_MD_IDX_INVALID;
1129 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1130 1, 61, sizeof(struct bm_extent),
1131 offsetof(struct bm_extent, lce));
1133 retcode = ERR_NOMEM;
1137 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1138 drbd_md_set_sector_offsets(mdev, nbc);
1140 if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1141 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1142 (unsigned long long) drbd_get_max_capacity(nbc),
1143 (unsigned long long) nbc->dc.disk_size);
1144 retcode = ERR_DISK_TO_SMALL;
1148 if ((int)nbc->dc.meta_dev_idx < 0) {
1149 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1150 /* at least one MB, otherwise it does not make sense */
1151 min_md_device_sectors = (2<<10);
1153 max_possible_sectors = DRBD_MAX_SECTORS;
1154 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1157 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1158 retcode = ERR_MD_DISK_TO_SMALL;
1159 dev_warn(DEV, "refusing attach: md-device too small, "
1160 "at least %llu sectors needed for this meta-disk type\n",
1161 (unsigned long long) min_md_device_sectors);
1165 /* Make sure the new disk is big enough
1166 * (we may currently be R_PRIMARY with no local disk...) */
1167 if (drbd_get_max_capacity(nbc) <
1168 drbd_get_capacity(mdev->this_bdev)) {
1169 retcode = ERR_DISK_TO_SMALL;
1173 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1175 if (nbc->known_size > max_possible_sectors) {
1176 dev_warn(DEV, "==> truncating very big lower level device "
1177 "to currently maximum possible %llu sectors <==\n",
1178 (unsigned long long) max_possible_sectors);
1179 if ((int)nbc->dc.meta_dev_idx >= 0)
1180 dev_warn(DEV, "==>> using internal or flexible "
1181 "meta data may help <<==\n");
1184 drbd_suspend_io(mdev);
1185 /* also wait for the last barrier ack. */
1186 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || is_susp(mdev->state));
1187 /* and for any other previously queued work */
1188 drbd_flush_workqueue(mdev);
1190 rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1191 retcode = rv; /* FIXME: Type mismatch. */
1192 drbd_resume_io(mdev);
1193 if (rv < SS_SUCCESS)
1196 if (!get_ldev_if_state(mdev, D_ATTACHING))
1197 goto force_diskless;
1199 drbd_md_set_sector_offsets(mdev, nbc);
1201 if (!mdev->bitmap) {
1202 if (drbd_bm_init(mdev)) {
1203 retcode = ERR_NOMEM;
1204 goto force_diskless_dec;
1208 retcode = drbd_md_read(mdev, nbc);
1209 if (retcode != NO_ERROR)
1210 goto force_diskless_dec;
1212 if (mdev->state.conn < C_CONNECTED &&
1213 mdev->state.role == R_PRIMARY &&
1214 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1215 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1216 (unsigned long long)mdev->ed_uuid);
1217 retcode = ERR_DATA_NOT_CURRENT;
1218 goto force_diskless_dec;
1221 /* Since we are diskless, fix the activity log first... */
1222 if (drbd_check_al_size(mdev)) {
1223 retcode = ERR_NOMEM;
1224 goto force_diskless_dec;
1227 /* Prevent shrinking of consistent devices ! */
1228 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1229 drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1230 dev_warn(DEV, "refusing to truncate a consistent device\n");
1231 retcode = ERR_DISK_TO_SMALL;
1232 goto force_diskless_dec;
1235 if (!drbd_al_read_log(mdev, nbc)) {
1236 retcode = ERR_IO_MD_DISK;
1237 goto force_diskless_dec;
1240 /* Reset the "barriers don't work" bits here, then force meta data to
1241 * be written, to ensure we determine if barriers are supported. */
1242 if (nbc->dc.no_md_flush)
1243 set_bit(MD_NO_FUA, &mdev->flags);
1245 clear_bit(MD_NO_FUA, &mdev->flags);
1247 /* Point of no return reached.
1248 * Devices and memory are no longer released by error cleanup below.
1249 * now mdev takes over responsibility, and the state engine should
1250 * clean it up somewhere. */
1251 D_ASSERT(mdev->ldev == NULL);
1253 mdev->resync = resync_lru;
1257 mdev->write_ordering = WO_bdev_flush;
1258 drbd_bump_write_ordering(mdev, WO_bdev_flush);
1260 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1261 set_bit(CRASHED_PRIMARY, &mdev->flags);
1263 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1265 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1266 !(mdev->state.role == R_PRIMARY && mdev->state.susp_nod)) {
1267 set_bit(CRASHED_PRIMARY, &mdev->flags);
1276 drbd_reconsider_max_bio_size(mdev);
1278 /* If I am currently not R_PRIMARY,
1279 * but meta data primary indicator is set,
1280 * I just now recover from a hard crash,
1281 * and have been R_PRIMARY before that crash.
1283 * Now, if I had no connection before that crash
1284 * (have been degraded R_PRIMARY), chances are that
1285 * I won't find my peer now either.
1287 * In that case, and _only_ in that case,
1288 * we use the degr-wfc-timeout instead of the default,
1289 * so we can automatically recover from a crash of a
1290 * degraded but active "cluster" after a certain timeout.
1292 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1293 if (mdev->state.role != R_PRIMARY &&
1294 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1295 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1296 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1298 dd = drbd_determine_dev_size(mdev, 0);
1299 if (dd == dev_size_error) {
1300 retcode = ERR_NOMEM_BITMAP;
1301 goto force_diskless_dec;
1302 } else if (dd == grew)
1303 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1305 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1306 dev_info(DEV, "Assuming that all blocks are out of sync "
1307 "(aka FullSync)\n");
1308 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1309 "set_n_write from attaching", BM_LOCKED_MASK)) {
1310 retcode = ERR_IO_MD_DISK;
1311 goto force_diskless_dec;
1314 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1315 "read from attaching", BM_LOCKED_MASK) < 0) {
1316 retcode = ERR_IO_MD_DISK;
1317 goto force_diskless_dec;
1321 if (cp_discovered) {
1322 drbd_al_apply_to_bm(mdev);
1323 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1324 "crashed primary apply AL", BM_LOCKED_MASK)) {
1325 retcode = ERR_IO_MD_DISK;
1326 goto force_diskless_dec;
1330 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1331 drbd_suspend_al(mdev); /* IO is still suspended here... */
1333 spin_lock_irq(&mdev->tconn->req_lock);
1336 /* If MDF_CONSISTENT is not set go into inconsistent state,
1337 otherwise investigate MDF_WasUpToDate...
1338 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1339 otherwise into D_CONSISTENT state.
1341 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1342 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1343 ns.disk = D_CONSISTENT;
1345 ns.disk = D_OUTDATED;
1347 ns.disk = D_INCONSISTENT;
1350 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1351 ns.pdsk = D_OUTDATED;
1353 if ( ns.disk == D_CONSISTENT &&
1354 (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1355 ns.disk = D_UP_TO_DATE;
1357 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1358 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1359 this point, because drbd_request_state() modifies these
1362 /* In case we are C_CONNECTED postpone any decision on the new disk
1363 state after the negotiation phase. */
1364 if (mdev->state.conn == C_CONNECTED) {
1365 mdev->new_state_tmp.i = ns.i;
1367 ns.disk = D_NEGOTIATING;
1369 /* We expect to receive up-to-date UUIDs soon.
1370 To avoid a race in receive_state, free p_uuid while
1371 holding req_lock. I.e. atomic with the state change */
1372 kfree(mdev->p_uuid);
1373 mdev->p_uuid = NULL;
1376 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1378 spin_unlock_irq(&mdev->tconn->req_lock);
1380 if (rv < SS_SUCCESS)
1381 goto force_diskless_dec;
1383 if (mdev->state.role == R_PRIMARY)
1384 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1386 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1388 drbd_md_mark_dirty(mdev);
1391 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1393 conn_reconfig_done(mdev->tconn);
1394 drbd_adm_finish(info, retcode);
1400 drbd_force_state(mdev, NS(disk, D_FAILED));
1402 conn_reconfig_done(mdev->tconn);
1405 if (nbc->backing_bdev)
1406 blkdev_put(nbc->backing_bdev,
1407 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1409 blkdev_put(nbc->md_bdev,
1410 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1413 lc_destroy(resync_lru);
1415 drbd_adm_finish(info, retcode);
1419 /* Detaching the disk is a process in multiple stages. First we need to lock
1420 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1421 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1422 * internal references as well.
1423 * Only then we have finally detached. */
1424 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1426 struct drbd_conf *mdev;
1427 enum drbd_ret_code retcode;
1429 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1430 if (!adm_ctx.reply_skb)
1432 if (retcode != NO_ERROR)
1435 mdev = adm_ctx.mdev;
1436 drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1437 retcode = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1438 wait_event(mdev->misc_wait,
1439 mdev->state.disk != D_DISKLESS ||
1440 !atomic_read(&mdev->local_cnt));
1441 drbd_resume_io(mdev);
1443 drbd_adm_finish(info, retcode);
1447 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1449 char hmac_name[CRYPTO_MAX_ALG_NAME];
1450 struct drbd_conf *mdev;
1451 struct net_conf *new_conf = NULL;
1452 struct crypto_hash *tfm = NULL;
1453 struct crypto_hash *integrity_w_tfm = NULL;
1454 struct crypto_hash *integrity_r_tfm = NULL;
1455 void *int_dig_out = NULL;
1456 void *int_dig_in = NULL;
1457 void *int_dig_vv = NULL;
1458 struct drbd_tconn *oconn;
1459 struct drbd_tconn *tconn;
1460 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1461 enum drbd_ret_code retcode;
1465 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1466 if (!adm_ctx.reply_skb)
1468 if (retcode != NO_ERROR)
1471 tconn = adm_ctx.tconn;
1472 conn_reconfig_start(tconn);
1474 if (tconn->cstate > C_STANDALONE) {
1475 retcode = ERR_NET_CONFIGURED;
1479 /* allocation not in the IO path, cqueue thread context */
1480 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1482 retcode = ERR_NOMEM;
1486 new_conf->timeout = DRBD_TIMEOUT_DEF;
1487 new_conf->try_connect_int = DRBD_CONNECT_INT_DEF;
1488 new_conf->ping_int = DRBD_PING_INT_DEF;
1489 new_conf->max_epoch_size = DRBD_MAX_EPOCH_SIZE_DEF;
1490 new_conf->max_buffers = DRBD_MAX_BUFFERS_DEF;
1491 new_conf->unplug_watermark = DRBD_UNPLUG_WATERMARK_DEF;
1492 new_conf->sndbuf_size = DRBD_SNDBUF_SIZE_DEF;
1493 new_conf->rcvbuf_size = DRBD_RCVBUF_SIZE_DEF;
1494 new_conf->ko_count = DRBD_KO_COUNT_DEF;
1495 new_conf->after_sb_0p = DRBD_AFTER_SB_0P_DEF;
1496 new_conf->after_sb_1p = DRBD_AFTER_SB_1P_DEF;
1497 new_conf->after_sb_2p = DRBD_AFTER_SB_2P_DEF;
1498 new_conf->want_lose = 0;
1499 new_conf->two_primaries = 0;
1500 new_conf->wire_protocol = DRBD_PROT_C;
1501 new_conf->ping_timeo = DRBD_PING_TIMEO_DEF;
1502 new_conf->rr_conflict = DRBD_RR_CONFLICT_DEF;
1503 new_conf->on_congestion = DRBD_ON_CONGESTION_DEF;
1504 new_conf->cong_extents = DRBD_CONG_EXTENTS_DEF;
1506 err = net_conf_from_attrs(new_conf, info->attrs);
1508 retcode = ERR_MANDATORY_TAG;
1509 drbd_msg_put_info(from_attrs_err_to_txt(err));
1513 if (new_conf->two_primaries
1514 && (new_conf->wire_protocol != DRBD_PROT_C)) {
1515 retcode = ERR_NOT_PROTO_C;
1519 idr_for_each_entry(&tconn->volumes, mdev, i) {
1520 if (get_ldev(mdev)) {
1521 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1523 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1524 retcode = ERR_STONITH_AND_PROT_A;
1528 if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1529 retcode = ERR_DISCARD;
1532 if (!mdev->bitmap) {
1533 if(drbd_bm_init(mdev)) {
1534 retcode = ERR_NOMEM;
1540 if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1541 retcode = ERR_CONG_NOT_PROTO_A;
1547 new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1548 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1550 /* No need to take drbd_cfg_mutex here. All reconfiguration is
1551 * strictly serialized on genl_lock(). We are protected against
1552 * concurrent reconfiguration/addition/deletion */
1553 list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
1556 if (get_net_conf(oconn)) {
1557 taken_addr = (struct sockaddr *)&oconn->net_conf->my_addr;
1558 if (new_conf->my_addr_len == oconn->net_conf->my_addr_len &&
1559 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1560 retcode = ERR_LOCAL_ADDR;
1562 taken_addr = (struct sockaddr *)&oconn->net_conf->peer_addr;
1563 if (new_conf->peer_addr_len == oconn->net_conf->peer_addr_len &&
1564 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1565 retcode = ERR_PEER_ADDR;
1567 put_net_conf(oconn);
1568 if (retcode != NO_ERROR)
1573 if (new_conf->cram_hmac_alg[0] != 0) {
1574 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1575 new_conf->cram_hmac_alg);
1576 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1579 retcode = ERR_AUTH_ALG;
1583 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1584 retcode = ERR_AUTH_ALG_ND;
1589 if (new_conf->integrity_alg[0]) {
1590 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1591 if (IS_ERR(integrity_w_tfm)) {
1592 integrity_w_tfm = NULL;
1593 retcode=ERR_INTEGRITY_ALG;
1597 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1598 retcode=ERR_INTEGRITY_ALG_ND;
1602 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1603 if (IS_ERR(integrity_r_tfm)) {
1604 integrity_r_tfm = NULL;
1605 retcode=ERR_INTEGRITY_ALG;
1610 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1612 /* allocation not in the IO path, cqueue thread context */
1613 if (integrity_w_tfm) {
1614 i = crypto_hash_digestsize(integrity_w_tfm);
1615 int_dig_out = kmalloc(i, GFP_KERNEL);
1617 retcode = ERR_NOMEM;
1620 int_dig_in = kmalloc(i, GFP_KERNEL);
1622 retcode = ERR_NOMEM;
1625 int_dig_vv = kmalloc(i, GFP_KERNEL);
1627 retcode = ERR_NOMEM;
1632 conn_flush_workqueue(tconn);
1633 spin_lock_irq(&tconn->req_lock);
1634 if (tconn->net_conf != NULL) {
1635 retcode = ERR_NET_CONFIGURED;
1636 spin_unlock_irq(&tconn->req_lock);
1639 tconn->net_conf = new_conf;
1641 crypto_free_hash(tconn->cram_hmac_tfm);
1642 tconn->cram_hmac_tfm = tfm;
1644 crypto_free_hash(tconn->integrity_w_tfm);
1645 tconn->integrity_w_tfm = integrity_w_tfm;
1647 crypto_free_hash(tconn->integrity_r_tfm);
1648 tconn->integrity_r_tfm = integrity_r_tfm;
1650 kfree(tconn->int_dig_out);
1651 kfree(tconn->int_dig_in);
1652 kfree(tconn->int_dig_vv);
1653 tconn->int_dig_out=int_dig_out;
1654 tconn->int_dig_in=int_dig_in;
1655 tconn->int_dig_vv=int_dig_vv;
1656 retcode = _conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
1657 spin_unlock_irq(&tconn->req_lock);
1659 idr_for_each_entry(&tconn->volumes, mdev, i) {
1662 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1664 conn_reconfig_done(tconn);
1665 drbd_adm_finish(info, retcode);
1672 crypto_free_hash(tfm);
1673 crypto_free_hash(integrity_w_tfm);
1674 crypto_free_hash(integrity_r_tfm);
1677 conn_reconfig_done(tconn);
1679 drbd_adm_finish(info, retcode);
1683 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
1685 struct disconnect_parms parms;
1686 struct drbd_tconn *tconn;
1687 enum drbd_ret_code retcode;
1690 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1691 if (!adm_ctx.reply_skb)
1693 if (retcode != NO_ERROR)
1696 tconn = adm_ctx.tconn;
1697 memset(&parms, 0, sizeof(parms));
1698 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
1699 err = disconnect_parms_from_attrs(&parms, info->attrs);
1701 retcode = ERR_MANDATORY_TAG;
1702 drbd_msg_put_info(from_attrs_err_to_txt(err));
1707 if (parms.force_disconnect) {
1708 spin_lock_irq(&tconn->req_lock);
1709 if (tconn->cstate >= C_WF_CONNECTION)
1710 _conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
1711 spin_unlock_irq(&tconn->req_lock);
1715 retcode = conn_request_state(tconn, NS(conn, C_DISCONNECTING), 0);
1717 if (retcode == SS_NOTHING_TO_DO)
1719 else if (retcode == SS_ALREADY_STANDALONE)
1721 else if (retcode == SS_PRIMARY_NOP) {
1722 /* Our state checking code wants to see the peer outdated. */
1723 retcode = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
1724 pdsk, D_OUTDATED), CS_VERBOSE);
1725 } else if (retcode == SS_CW_FAILED_BY_PEER) {
1726 /* The peer probably wants to see us outdated. */
1727 retcode = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
1728 disk, D_OUTDATED), 0);
1729 if (retcode == SS_IS_DISKLESS || retcode == SS_LOWER_THAN_OUTDATED) {
1730 conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
1731 retcode = SS_SUCCESS;
1735 if (retcode < SS_SUCCESS)
1738 if (wait_event_interruptible(tconn->ping_wait,
1739 tconn->cstate != C_DISCONNECTING)) {
1740 /* Do not test for mdev->state.conn == C_STANDALONE, since
1741 someone else might connect us in the mean time! */
1749 drbd_adm_finish(info, retcode);
1753 void resync_after_online_grow(struct drbd_conf *mdev)
1755 int iass; /* I am sync source */
1757 dev_info(DEV, "Resync of new storage after online grow\n");
1758 if (mdev->state.role != mdev->state.peer)
1759 iass = (mdev->state.role == R_PRIMARY);
1761 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
1764 drbd_start_resync(mdev, C_SYNC_SOURCE);
1766 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
1769 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
1771 struct resize_parms rs;
1772 struct drbd_conf *mdev;
1773 enum drbd_ret_code retcode;
1774 enum determine_dev_size dd;
1775 enum dds_flags ddsf;
1778 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1779 if (!adm_ctx.reply_skb)
1781 if (retcode != NO_ERROR)
1784 memset(&rs, 0, sizeof(struct resize_parms));
1785 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
1786 err = resize_parms_from_attrs(&rs, info->attrs);
1788 retcode = ERR_MANDATORY_TAG;
1789 drbd_msg_put_info(from_attrs_err_to_txt(err));
1794 mdev = adm_ctx.mdev;
1795 if (mdev->state.conn > C_CONNECTED) {
1796 retcode = ERR_RESIZE_RESYNC;
1800 if (mdev->state.role == R_SECONDARY &&
1801 mdev->state.peer == R_SECONDARY) {
1802 retcode = ERR_NO_PRIMARY;
1806 if (!get_ldev(mdev)) {
1807 retcode = ERR_NO_DISK;
1811 if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
1812 retcode = ERR_NEED_APV_93;
1816 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
1817 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
1819 mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
1820 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
1821 dd = drbd_determine_dev_size(mdev, ddsf);
1824 if (dd == dev_size_error) {
1825 retcode = ERR_NOMEM_BITMAP;
1829 if (mdev->state.conn == C_CONNECTED) {
1831 set_bit(RESIZE_PENDING, &mdev->flags);
1833 drbd_send_uuids(mdev);
1834 drbd_send_sizes(mdev, 1, ddsf);
1838 drbd_adm_finish(info, retcode);
1842 int drbd_adm_syncer(struct sk_buff *skb, struct genl_info *info)
1844 struct drbd_conf *mdev;
1845 enum drbd_ret_code retcode;
1847 int ovr; /* online verify running */
1848 int rsr; /* re-sync running */
1849 struct crypto_hash *verify_tfm = NULL;
1850 struct crypto_hash *csums_tfm = NULL;
1851 struct syncer_conf sc;
1852 cpumask_var_t new_cpu_mask;
1853 int *rs_plan_s = NULL;
1856 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1857 if (!adm_ctx.reply_skb)
1859 if (retcode != NO_ERROR)
1861 mdev = adm_ctx.mdev;
1863 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
1864 retcode = ERR_NOMEM;
1865 drbd_msg_put_info("unable to allocate cpumask");
1869 if (((struct drbd_genlmsghdr*)info->userhdr)->flags
1870 & DRBD_GENL_F_SET_DEFAULTS) {
1871 memset(&sc, 0, sizeof(struct syncer_conf));
1872 sc.rate = DRBD_RATE_DEF;
1873 sc.after = DRBD_AFTER_DEF;
1874 sc.al_extents = DRBD_AL_EXTENTS_DEF;
1875 sc.on_no_data = DRBD_ON_NO_DATA_DEF;
1876 sc.c_plan_ahead = DRBD_C_PLAN_AHEAD_DEF;
1877 sc.c_delay_target = DRBD_C_DELAY_TARGET_DEF;
1878 sc.c_fill_target = DRBD_C_FILL_TARGET_DEF;
1879 sc.c_max_rate = DRBD_C_MAX_RATE_DEF;
1880 sc.c_min_rate = DRBD_C_MIN_RATE_DEF;
1882 memcpy(&sc, &mdev->sync_conf, sizeof(struct syncer_conf));
1884 err = syncer_conf_from_attrs(&sc, info->attrs);
1886 retcode = ERR_MANDATORY_TAG;
1887 drbd_msg_put_info(from_attrs_err_to_txt(err));
1891 /* re-sync running */
1892 rsr = ( mdev->state.conn == C_SYNC_SOURCE ||
1893 mdev->state.conn == C_SYNC_TARGET ||
1894 mdev->state.conn == C_PAUSED_SYNC_S ||
1895 mdev->state.conn == C_PAUSED_SYNC_T );
1897 if (rsr && strcmp(sc.csums_alg, mdev->sync_conf.csums_alg)) {
1898 retcode = ERR_CSUMS_RESYNC_RUNNING;
1902 if (!rsr && sc.csums_alg[0]) {
1903 csums_tfm = crypto_alloc_hash(sc.csums_alg, 0, CRYPTO_ALG_ASYNC);
1904 if (IS_ERR(csums_tfm)) {
1906 retcode = ERR_CSUMS_ALG;
1910 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1911 retcode = ERR_CSUMS_ALG_ND;
1916 /* online verify running */
1917 ovr = (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T);
1920 if (strcmp(sc.verify_alg, mdev->sync_conf.verify_alg)) {
1921 retcode = ERR_VERIFY_RUNNING;
1926 if (!ovr && sc.verify_alg[0]) {
1927 verify_tfm = crypto_alloc_hash(sc.verify_alg, 0, CRYPTO_ALG_ASYNC);
1928 if (IS_ERR(verify_tfm)) {
1930 retcode = ERR_VERIFY_ALG;
1934 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1935 retcode = ERR_VERIFY_ALG_ND;
1940 /* silently ignore cpu mask on UP kernel */
1941 if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
1942 err = __bitmap_parse(sc.cpu_mask, 32, 0,
1943 cpumask_bits(new_cpu_mask), nr_cpu_ids);
1945 dev_warn(DEV, "__bitmap_parse() failed with %d\n", err);
1946 retcode = ERR_CPU_MASK_PARSE;
1951 if (!expect(sc.rate >= 1))
1954 /* clip to allowed range */
1955 if (!expect(sc.al_extents >= DRBD_AL_EXTENTS_MIN))
1956 sc.al_extents = DRBD_AL_EXTENTS_MIN;
1957 if (!expect(sc.al_extents <= DRBD_AL_EXTENTS_MAX))
1958 sc.al_extents = DRBD_AL_EXTENTS_MAX;
1960 /* most sanity checks done, try to assign the new sync-after
1961 * dependency. need to hold the global lock in there,
1962 * to avoid a race in the dependency loop check. */
1963 retcode = drbd_alter_sa(mdev, sc.after);
1964 if (retcode != NO_ERROR)
1967 fifo_size = (sc.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1968 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1969 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1971 dev_err(DEV, "kmalloc of fifo_buffer failed");
1972 retcode = ERR_NOMEM;
1977 /* ok, assign the rest of it as well.
1978 * lock against receive_SyncParam() */
1979 spin_lock(&mdev->peer_seq_lock);
1980 mdev->sync_conf = sc;
1983 crypto_free_hash(mdev->csums_tfm);
1984 mdev->csums_tfm = csums_tfm;
1989 crypto_free_hash(mdev->verify_tfm);
1990 mdev->verify_tfm = verify_tfm;
1994 if (fifo_size != mdev->rs_plan_s.size) {
1995 kfree(mdev->rs_plan_s.values);
1996 mdev->rs_plan_s.values = rs_plan_s;
1997 mdev->rs_plan_s.size = fifo_size;
1998 mdev->rs_planed = 0;
2002 spin_unlock(&mdev->peer_seq_lock);
2004 if (get_ldev(mdev)) {
2005 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
2006 drbd_al_shrink(mdev);
2007 err = drbd_check_al_size(mdev);
2008 lc_unlock(mdev->act_log);
2009 wake_up(&mdev->al_wait);
2015 retcode = ERR_NOMEM;
2020 if (mdev->state.conn >= C_CONNECTED)
2021 drbd_send_sync_param(mdev, &sc);
2023 if (!cpumask_equal(mdev->tconn->cpu_mask, new_cpu_mask)) {
2024 cpumask_copy(mdev->tconn->cpu_mask, new_cpu_mask);
2025 drbd_calc_cpu_mask(mdev->tconn);
2026 mdev->tconn->receiver.reset_cpu_mask = 1;
2027 mdev->tconn->asender.reset_cpu_mask = 1;
2028 mdev->tconn->worker.reset_cpu_mask = 1;
2031 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
2034 free_cpumask_var(new_cpu_mask);
2035 crypto_free_hash(csums_tfm);
2036 crypto_free_hash(verify_tfm);
2038 drbd_adm_finish(info, retcode);
2042 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2044 struct drbd_conf *mdev;
2045 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2047 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2048 if (!adm_ctx.reply_skb)
2050 if (retcode != NO_ERROR)
2053 mdev = adm_ctx.mdev;
2055 /* If there is still bitmap IO pending, probably because of a previous
2056 * resync just being finished, wait for it before requesting a new resync. */
2057 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2059 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2061 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2062 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2064 while (retcode == SS_NEED_CONNECTION) {
2065 spin_lock_irq(&mdev->tconn->req_lock);
2066 if (mdev->state.conn < C_CONNECTED)
2067 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2068 spin_unlock_irq(&mdev->tconn->req_lock);
2070 if (retcode != SS_NEED_CONNECTION)
2073 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2077 drbd_adm_finish(info, retcode);
2081 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2085 rv = drbd_bmio_set_n_write(mdev);
2086 drbd_suspend_al(mdev);
2090 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2091 union drbd_state mask, union drbd_state val)
2093 enum drbd_ret_code retcode;
2095 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2096 if (!adm_ctx.reply_skb)
2098 if (retcode != NO_ERROR)
2101 retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2103 drbd_adm_finish(info, retcode);
2107 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2109 return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2112 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2114 enum drbd_ret_code retcode;
2116 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2117 if (!adm_ctx.reply_skb)
2119 if (retcode != NO_ERROR)
2122 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2123 retcode = ERR_PAUSE_IS_SET;
2125 drbd_adm_finish(info, retcode);
2129 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2132 enum drbd_ret_code retcode;
2134 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2135 if (!adm_ctx.reply_skb)
2137 if (retcode != NO_ERROR)
2140 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2141 s = adm_ctx.mdev->state;
2142 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2143 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2144 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2146 retcode = ERR_PAUSE_IS_CLEAR;
2151 drbd_adm_finish(info, retcode);
2155 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2157 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2160 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2162 struct drbd_conf *mdev;
2163 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2165 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2166 if (!adm_ctx.reply_skb)
2168 if (retcode != NO_ERROR)
2171 mdev = adm_ctx.mdev;
2172 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2173 drbd_uuid_new_current(mdev);
2174 clear_bit(NEW_CUR_UUID, &mdev->flags);
2176 drbd_suspend_io(mdev);
2177 retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2178 if (retcode == SS_SUCCESS) {
2179 if (mdev->state.conn < C_CONNECTED)
2180 tl_clear(mdev->tconn);
2181 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2182 tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2184 drbd_resume_io(mdev);
2187 drbd_adm_finish(info, retcode);
2191 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2193 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2196 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *conn_name, unsigned vnr)
2199 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2201 goto nla_put_failure;
2202 if (vnr != VOLUME_UNSPECIFIED)
2203 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2204 NLA_PUT_STRING(skb, T_ctx_conn_name, conn_name);
2205 nla_nest_end(skb, nla);
2210 nla_nest_cancel(skb, nla);
2214 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2215 const struct sib_info *sib)
2217 struct state_info *si = NULL; /* for sizeof(si->member); */
2222 int exclude_sensitive;
2224 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2225 * to. So we better exclude_sensitive information.
2227 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2228 * in the context of the requesting user process. Exclude sensitive
2229 * information, unless current has superuser.
2231 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2232 * relies on the current implementation of netlink_dump(), which
2233 * executes the dump callback successively from netlink_recvmsg(),
2234 * always in the context of the receiving process */
2235 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2237 got_ldev = get_ldev(mdev);
2238 got_net = get_net_conf(mdev->tconn);
2240 /* We need to add connection name and volume number information still.
2241 * Minor number is in drbd_genlmsghdr. */
2242 if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2243 goto nla_put_failure;
2246 if (disk_conf_to_skb(skb, &mdev->ldev->dc, exclude_sensitive))
2247 goto nla_put_failure;
2249 if (net_conf_to_skb(skb, mdev->tconn->net_conf, exclude_sensitive))
2250 goto nla_put_failure;
2252 if (syncer_conf_to_skb(skb, &mdev->sync_conf, exclude_sensitive))
2253 goto nla_put_failure;
2255 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2257 goto nla_put_failure;
2258 NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2259 NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2260 NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2261 NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2264 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2265 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2266 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2267 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2268 if (C_SYNC_SOURCE <= mdev->state.conn &&
2269 C_PAUSED_SYNC_T >= mdev->state.conn) {
2270 NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2271 NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2276 switch(sib->sib_reason) {
2277 case SIB_SYNC_PROGRESS:
2278 case SIB_GET_STATUS_REPLY:
2280 case SIB_STATE_CHANGE:
2281 NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2282 NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2284 case SIB_HELPER_POST:
2286 T_helper_exit_code, sib->helper_exit_code);
2288 case SIB_HELPER_PRE:
2289 NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2293 nla_nest_end(skb, nla);
2301 put_net_conf(mdev->tconn);
2305 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2307 enum drbd_ret_code retcode;
2310 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2311 if (!adm_ctx.reply_skb)
2313 if (retcode != NO_ERROR)
2316 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2318 nlmsg_free(adm_ctx.reply_skb);
2322 drbd_adm_finish(info, retcode);
2326 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2328 struct drbd_conf *mdev;
2329 struct drbd_genlmsghdr *dh;
2330 struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2331 struct drbd_tconn *tconn = NULL;
2332 struct drbd_tconn *tmp;
2333 unsigned volume = cb->args[1];
2335 /* Open coded, deferred, iteration:
2336 * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2337 * idr_for_each_entry(&tconn->volumes, mdev, i) {
2341 * where tconn is cb->args[0];
2342 * and i is cb->args[1];
2344 * This may miss entries inserted after this dump started,
2345 * or entries deleted before they are reached.
2347 * We need to make sure the mdev won't disappear while
2348 * we are looking at it, and revalidate our iterators
2349 * on each iteration.
2352 /* synchronize with drbd_new_tconn/drbd_free_tconn */
2353 mutex_lock(&drbd_cfg_mutex);
2354 /* synchronize with drbd_delete_device */
2357 /* revalidate iterator position */
2358 list_for_each_entry(tmp, &drbd_tconns, all_tconn) {
2360 /* first iteration */
2371 mdev = idr_get_next(&tconn->volumes, &volume);
2373 /* No more volumes to dump on this tconn.
2374 * Advance tconn iterator. */
2375 pos = list_entry(tconn->all_tconn.next,
2376 struct drbd_tconn, all_tconn);
2377 /* But, did we dump any volume on this tconn yet? */
2385 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2386 cb->nlh->nlmsg_seq, &drbd_genl_family,
2387 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2392 /* this is a tconn without a single volume */
2394 dh->ret_code = NO_ERROR;
2395 if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2396 genlmsg_cancel(skb, dh);
2398 genlmsg_end(skb, dh);
2402 D_ASSERT(mdev->vnr == volume);
2403 D_ASSERT(mdev->tconn == tconn);
2405 dh->minor = mdev_to_minor(mdev);
2406 dh->ret_code = NO_ERROR;
2408 if (nla_put_status_info(skb, mdev, NULL)) {
2409 genlmsg_cancel(skb, dh);
2412 genlmsg_end(skb, dh);
2417 mutex_unlock(&drbd_cfg_mutex);
2418 /* where to start the next iteration */
2419 cb->args[0] = (long)pos;
2420 cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2422 /* No more tconns/volumes/minors found results in an empty skb.
2423 * Which will terminate the dump. */
2427 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2429 enum drbd_ret_code retcode;
2430 struct timeout_parms tp;
2433 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2434 if (!adm_ctx.reply_skb)
2436 if (retcode != NO_ERROR)
2440 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2441 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2444 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2446 nlmsg_free(adm_ctx.reply_skb);
2450 drbd_adm_finish(info, retcode);
2454 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2456 struct drbd_conf *mdev;
2457 enum drbd_ret_code retcode;
2459 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2460 if (!adm_ctx.reply_skb)
2462 if (retcode != NO_ERROR)
2465 mdev = adm_ctx.mdev;
2466 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2467 /* resume from last known position, if possible */
2468 struct start_ov_parms parms =
2469 { .ov_start_sector = mdev->ov_start_sector };
2470 int err = start_ov_parms_from_attrs(&parms, info->attrs);
2472 retcode = ERR_MANDATORY_TAG;
2473 drbd_msg_put_info(from_attrs_err_to_txt(err));
2476 /* w_make_ov_request expects position to be aligned */
2477 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2479 /* If there is still bitmap IO pending, e.g. previous resync or verify
2480 * just being finished, wait for it before requesting a new resync. */
2481 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2482 retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2484 drbd_adm_finish(info, retcode);
2489 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2491 struct drbd_conf *mdev;
2492 enum drbd_ret_code retcode;
2493 int skip_initial_sync = 0;
2495 struct new_c_uuid_parms args;
2497 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2498 if (!adm_ctx.reply_skb)
2500 if (retcode != NO_ERROR)
2503 mdev = adm_ctx.mdev;
2504 memset(&args, 0, sizeof(args));
2505 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2506 err = new_c_uuid_parms_from_attrs(&args, info->attrs);
2508 retcode = ERR_MANDATORY_TAG;
2509 drbd_msg_put_info(from_attrs_err_to_txt(err));
2514 mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2516 if (!get_ldev(mdev)) {
2517 retcode = ERR_NO_DISK;
2521 /* this is "skip initial sync", assume to be clean */
2522 if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2523 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2524 dev_info(DEV, "Preparing to skip initial sync\n");
2525 skip_initial_sync = 1;
2526 } else if (mdev->state.conn != C_STANDALONE) {
2527 retcode = ERR_CONNECTED;
2531 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2532 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2534 if (args.clear_bm) {
2535 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2536 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2538 dev_err(DEV, "Writing bitmap failed with %d\n",err);
2539 retcode = ERR_IO_MD_DISK;
2541 if (skip_initial_sync) {
2542 drbd_send_uuids_skip_initial_sync(mdev);
2543 _drbd_uuid_set(mdev, UI_BITMAP, 0);
2544 drbd_print_uuids(mdev, "cleared bitmap UUID");
2545 spin_lock_irq(&mdev->tconn->req_lock);
2546 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2548 spin_unlock_irq(&mdev->tconn->req_lock);
2556 mutex_unlock(mdev->state_mutex);
2558 drbd_adm_finish(info, retcode);
2562 static enum drbd_ret_code
2563 drbd_check_conn_name(const char *name)
2565 if (!name || !name[0]) {
2566 drbd_msg_put_info("connection name missing");
2567 return ERR_MANDATORY_TAG;
2569 /* if we want to use these in sysfs/configfs/debugfs some day,
2570 * we must not allow slashes */
2571 if (strchr(name, '/')) {
2572 drbd_msg_put_info("invalid connection name");
2573 return ERR_INVALID_REQUEST;
2578 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info)
2580 enum drbd_ret_code retcode;
2582 retcode = drbd_adm_prepare(skb, info, 0);
2583 if (!adm_ctx.reply_skb)
2585 if (retcode != NO_ERROR)
2588 retcode = drbd_check_conn_name(adm_ctx.conn_name);
2589 if (retcode != NO_ERROR)
2592 if (adm_ctx.tconn) {
2593 retcode = ERR_INVALID_REQUEST;
2594 drbd_msg_put_info("connection exists");
2598 if (!drbd_new_tconn(adm_ctx.conn_name))
2599 retcode = ERR_NOMEM;
2601 drbd_adm_finish(info, retcode);
2605 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
2607 struct drbd_genlmsghdr *dh = info->userhdr;
2608 enum drbd_ret_code retcode;
2610 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2611 if (!adm_ctx.reply_skb)
2613 if (retcode != NO_ERROR)
2616 /* FIXME drop minor_count parameter, limit to MINORMASK */
2617 if (dh->minor >= minor_count) {
2618 drbd_msg_put_info("requested minor out of range");
2619 retcode = ERR_INVALID_REQUEST;
2622 /* FIXME we need a define here */
2623 if (adm_ctx.volume >= 256) {
2624 drbd_msg_put_info("requested volume id out of range");
2625 retcode = ERR_INVALID_REQUEST;
2629 retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
2631 drbd_adm_finish(info, retcode);
2635 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
2637 struct drbd_conf *mdev;
2638 enum drbd_ret_code retcode;
2640 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2641 if (!adm_ctx.reply_skb)
2643 if (retcode != NO_ERROR)
2646 mdev = adm_ctx.mdev;
2647 if (mdev->state.disk == D_DISKLESS &&
2648 /* no need to be mdev->state.conn == C_STANDALONE &&
2649 * we may want to delete a minor from a live replication group.
2651 mdev->state.role == R_SECONDARY) {
2652 drbd_delete_device(mdev_to_minor(mdev));
2654 /* if this was the last volume of this connection,
2655 * this will terminate all threads */
2656 conn_reconfig_done(adm_ctx.tconn);
2658 retcode = ERR_MINOR_CONFIGURED;
2660 drbd_adm_finish(info, retcode);
2664 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info)
2666 enum drbd_ret_code retcode;
2668 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2669 if (!adm_ctx.reply_skb)
2671 if (retcode != NO_ERROR)
2674 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
2675 drbd_free_tconn(adm_ctx.tconn);
2678 retcode = ERR_CONN_IN_USE;
2682 drbd_adm_finish(info, retcode);
2686 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
2688 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
2689 struct sk_buff *msg;
2690 struct drbd_genlmsghdr *d_out;
2694 seq = atomic_inc_return(&drbd_genl_seq);
2695 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
2700 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
2701 if (!d_out) /* cannot happen, but anyways. */
2702 goto nla_put_failure;
2703 d_out->minor = mdev_to_minor(mdev);
2704 d_out->ret_code = 0;
2706 if (nla_put_status_info(msg, mdev, sib))
2707 goto nla_put_failure;
2708 genlmsg_end(msg, d_out);
2709 err = drbd_genl_multicast_events(msg, 0);
2710 /* msg has been consumed or freed in netlink_broadcast() */
2711 if (err && err != -ESRCH)
2719 dev_err(DEV, "Error %d while broadcasting event. "
2720 "Event seq:%u sib_reason:%u\n",
2721 err, seq, sib->sib_reason);