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);
52 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
54 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
55 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
56 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
57 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
58 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
59 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
60 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
61 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
62 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
63 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
64 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
65 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
66 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
67 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
68 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
69 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
70 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
71 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
72 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
73 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
75 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
77 #include <linux/drbd_genl_api.h>
78 #include <linux/genl_magic_func.h>
80 /* used blkdev_get_by_path, to claim our meta data device(s) */
81 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
83 /* Configuration is strictly serialized, because generic netlink message
84 * processing is strictly serialized by the genl_lock().
85 * Which means we can use one static global drbd_config_context struct.
87 static struct drbd_config_context {
88 /* assigned from drbd_genlmsghdr */
90 /* assigned from request attributes, if present */
92 #define VOLUME_UNSPECIFIED (-1U)
93 /* pointer into the request skb,
94 * limited lifetime! */
98 struct sk_buff *reply_skb;
99 /* pointer into reply buffer */
100 struct drbd_genlmsghdr *reply_dh;
101 /* resolved from attributes, if possible */
102 struct drbd_conf *mdev;
103 struct drbd_tconn *tconn;
106 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
108 genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
109 if (genlmsg_reply(skb, info))
110 printk(KERN_ERR "drbd: error sending genl reply\n");
113 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
114 * reason it could fail was no space in skb, and there are 4k available. */
115 int drbd_msg_put_info(const char *info)
117 struct sk_buff *skb = adm_ctx.reply_skb;
121 if (!info || !info[0])
124 nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
128 err = nla_put_string(skb, T_info_text, info);
130 nla_nest_cancel(skb, nla);
133 nla_nest_end(skb, nla);
137 /* This would be a good candidate for a "pre_doit" hook,
138 * and per-family private info->pointers.
139 * But we need to stay compatible with older kernels.
140 * If it returns successfully, adm_ctx members are valid.
142 #define DRBD_ADM_NEED_MINOR 1
143 #define DRBD_ADM_NEED_CONN 2
144 static int drbd_adm_prepare(struct sk_buff *skb, struct genl_info *info,
147 struct drbd_genlmsghdr *d_in = info->userhdr;
148 const u8 cmd = info->genlhdr->cmd;
151 memset(&adm_ctx, 0, sizeof(adm_ctx));
153 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
154 if (cmd != DRBD_ADM_GET_STATUS
155 && security_netlink_recv(skb, CAP_SYS_ADMIN))
158 adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
159 if (!adm_ctx.reply_skb)
162 adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
163 info, &drbd_genl_family, 0, cmd);
164 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
166 if (!adm_ctx.reply_dh)
169 adm_ctx.reply_dh->minor = d_in->minor;
170 adm_ctx.reply_dh->ret_code = NO_ERROR;
172 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
174 /* parse and validate only */
175 err = drbd_cfg_context_from_attrs(NULL, info);
179 /* It was present, and valid,
180 * copy it over to the reply skb. */
181 err = nla_put_nohdr(adm_ctx.reply_skb,
182 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
183 info->attrs[DRBD_NLA_CFG_CONTEXT]);
187 /* and assign stuff to the global adm_ctx */
188 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
189 adm_ctx.volume = nla ? nla_get_u32(nla) : VOLUME_UNSPECIFIED;
190 nla = nested_attr_tb[__nla_type(T_ctx_conn_name)];
192 adm_ctx.conn_name = nla_data(nla);
194 adm_ctx.volume = VOLUME_UNSPECIFIED;
196 adm_ctx.minor = d_in->minor;
197 adm_ctx.mdev = minor_to_mdev(d_in->minor);
198 adm_ctx.tconn = conn_by_name(adm_ctx.conn_name);
200 if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
201 drbd_msg_put_info("unknown minor");
202 return ERR_MINOR_INVALID;
204 if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_CONN)) {
205 drbd_msg_put_info("unknown connection");
206 return ERR_INVALID_REQUEST;
209 /* some more paranoia, if the request was over-determined */
210 if (adm_ctx.mdev && adm_ctx.tconn &&
211 adm_ctx.mdev->tconn != adm_ctx.tconn) {
212 pr_warning("request: minor=%u, conn=%s; but that minor belongs to connection %s\n",
213 adm_ctx.minor, adm_ctx.conn_name, adm_ctx.mdev->tconn->name);
214 drbd_msg_put_info("minor exists in different connection");
215 return ERR_INVALID_REQUEST;
218 adm_ctx.volume != VOLUME_UNSPECIFIED &&
219 adm_ctx.volume != adm_ctx.mdev->vnr) {
220 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
221 adm_ctx.minor, adm_ctx.volume,
222 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
223 drbd_msg_put_info("minor exists as different volume");
224 return ERR_INVALID_REQUEST;
226 if (adm_ctx.mdev && !adm_ctx.tconn)
227 adm_ctx.tconn = adm_ctx.mdev->tconn;
231 nlmsg_free(adm_ctx.reply_skb);
232 adm_ctx.reply_skb = NULL;
236 static int drbd_adm_finish(struct genl_info *info, int retcode)
239 const char *conn_name = NULL;
241 if (!adm_ctx.reply_skb)
244 adm_ctx.reply_dh->ret_code = retcode;
246 nla = info->attrs[DRBD_NLA_CFG_CONTEXT];
248 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
250 conn_name = nla_data(nla);
253 drbd_adm_send_reply(adm_ctx.reply_skb, info);
257 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
261 if (get_net_conf(tconn)) {
262 switch (((struct sockaddr *)tconn->net_conf->peer_addr)->sa_family) {
265 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
266 &((struct sockaddr_in6 *)tconn->net_conf->peer_addr)->sin6_addr);
270 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
271 &((struct sockaddr_in *)tconn->net_conf->peer_addr)->sin_addr);
275 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
276 &((struct sockaddr_in *)tconn->net_conf->peer_addr)->sin_addr);
278 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
283 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
285 char *envp[] = { "HOME=/",
287 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
288 (char[20]) { }, /* address family */
289 (char[60]) { }, /* address */
292 char *argv[] = {usermode_helper, cmd, mb, NULL };
296 snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
297 setup_khelper_env(mdev->tconn, envp);
299 /* The helper may take some time.
300 * write out any unsynced meta data changes now */
303 dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
304 sib.sib_reason = SIB_HELPER_PRE;
305 sib.helper_name = cmd;
306 drbd_bcast_event(mdev, &sib);
307 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
309 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
310 usermode_helper, cmd, mb,
311 (ret >> 8) & 0xff, ret);
313 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
314 usermode_helper, cmd, mb,
315 (ret >> 8) & 0xff, ret);
316 sib.sib_reason = SIB_HELPER_POST;
317 sib.helper_exit_code = ret;
318 drbd_bcast_event(mdev, &sib);
320 if (ret < 0) /* Ignore any ERRNOs we got. */
326 static void conn_md_sync(struct drbd_tconn *tconn)
328 struct drbd_conf *mdev;
331 idr_for_each_entry(&tconn->volumes, mdev, vnr)
335 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
337 char *envp[] = { "HOME=/",
339 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
340 (char[20]) { }, /* address family */
341 (char[60]) { }, /* address */
343 char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
346 setup_khelper_env(tconn, envp);
349 conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
350 /* TODO: conn_bcast_event() ?? */
352 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
354 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
355 usermode_helper, cmd, tconn->name,
356 (ret >> 8) & 0xff, ret);
358 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
359 usermode_helper, cmd, tconn->name,
360 (ret >> 8) & 0xff, ret);
361 /* TODO: conn_bcast_event() ?? */
363 if (ret < 0) /* Ignore any ERRNOs we got. */
369 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
371 enum drbd_fencing_p fp = FP_NOT_AVAIL;
372 struct drbd_conf *mdev;
376 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
377 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
378 fp = max_t(enum drbd_fencing_p, fp, mdev->ldev->dc.fencing);
387 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
389 union drbd_state mask = { };
390 union drbd_state val = { };
391 enum drbd_fencing_p fp;
395 if (tconn->cstate >= C_WF_REPORT_PARAMS) {
396 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
400 fp = highest_fencing_policy(tconn);
403 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
410 r = conn_khelper(tconn, "fence-peer");
412 switch ((r>>8) & 0xff) {
413 case 3: /* peer is inconsistent */
414 ex_to_string = "peer is inconsistent or worse";
416 val.pdsk = D_INCONSISTENT;
418 case 4: /* peer got outdated, or was already outdated */
419 ex_to_string = "peer was fenced";
421 val.pdsk = D_OUTDATED;
423 case 5: /* peer was down */
424 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
425 /* we will(have) create(d) a new UUID anyways... */
426 ex_to_string = "peer is unreachable, assumed to be dead";
428 val.pdsk = D_OUTDATED;
430 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
433 case 6: /* Peer is primary, voluntarily outdate myself.
434 * This is useful when an unconnected R_SECONDARY is asked to
435 * become R_PRIMARY, but finds the other peer being active. */
436 ex_to_string = "peer is active";
437 conn_warn(tconn, "Peer is primary, outdating myself.\n");
439 val.disk = D_OUTDATED;
442 if (fp != FP_STONITH)
443 conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
444 ex_to_string = "peer was stonithed";
446 val.pdsk = D_OUTDATED;
449 /* The script is broken ... */
450 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
451 return false; /* Eventually leave IO frozen */
454 conn_info(tconn, "fence-peer helper returned %d (%s)\n",
455 (r>>8) & 0xff, ex_to_string);
460 conn_request_state(tconn, mask, val, CS_VERBOSE);
461 here, because we might were able to re-establish the connection in the
463 spin_lock_irq(&tconn->req_lock);
464 if (tconn->cstate < C_WF_REPORT_PARAMS)
465 _conn_request_state(tconn, mask, val, CS_VERBOSE);
466 spin_unlock_irq(&tconn->req_lock);
468 return conn_highest_pdsk(tconn) <= D_OUTDATED;
471 static int _try_outdate_peer_async(void *data)
473 struct drbd_tconn *tconn = (struct drbd_tconn *)data;
475 conn_try_outdate_peer(tconn);
480 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
482 struct task_struct *opa;
484 opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
486 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
490 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
492 const int max_tries = 4;
493 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
496 union drbd_state mask, val;
498 if (new_role == R_PRIMARY)
499 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
501 mutex_lock(mdev->state_mutex);
503 mask.i = 0; mask.role = R_MASK;
504 val.i = 0; val.role = new_role;
506 while (try++ < max_tries) {
507 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
509 /* in case we first succeeded to outdate,
510 * but now suddenly could establish a connection */
511 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
517 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
518 (mdev->state.disk < D_UP_TO_DATE &&
519 mdev->state.disk >= D_INCONSISTENT)) {
521 val.disk = D_UP_TO_DATE;
526 if (rv == SS_NO_UP_TO_DATE_DISK &&
527 mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
528 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
530 if (conn_try_outdate_peer(mdev->tconn)) {
531 val.disk = D_UP_TO_DATE;
537 if (rv == SS_NOTHING_TO_DO)
539 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
540 if (!conn_try_outdate_peer(mdev->tconn) && force) {
541 dev_warn(DEV, "Forced into split brain situation!\n");
543 val.pdsk = D_OUTDATED;
548 if (rv == SS_TWO_PRIMARIES) {
549 /* Maybe the peer is detected as dead very soon...
550 retry at most once more in this case. */
551 schedule_timeout_interruptible((mdev->tconn->net_conf->ping_timeo+1)*HZ/10);
556 if (rv < SS_SUCCESS) {
557 rv = _drbd_request_state(mdev, mask, val,
558 CS_VERBOSE + CS_WAIT_COMPLETE);
569 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
571 /* Wait until nothing is on the fly :) */
572 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
574 if (new_role == R_SECONDARY) {
575 set_disk_ro(mdev->vdisk, true);
576 if (get_ldev(mdev)) {
577 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
581 if (get_net_conf(mdev->tconn)) {
582 mdev->tconn->net_conf->want_lose = 0;
583 put_net_conf(mdev->tconn);
585 set_disk_ro(mdev->vdisk, false);
586 if (get_ldev(mdev)) {
587 if (((mdev->state.conn < C_CONNECTED ||
588 mdev->state.pdsk <= D_FAILED)
589 && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
590 drbd_uuid_new_current(mdev);
592 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
597 /* writeout of activity log covered areas of the bitmap
598 * to stable storage done in after state change already */
600 if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
601 /* if this was forced, we should consider sync */
603 drbd_send_uuids(mdev);
604 drbd_send_state(mdev);
609 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
611 mutex_unlock(mdev->state_mutex);
615 static const char *from_attrs_err_to_txt(int err)
617 return err == -ENOMSG ? "required attribute missing" :
618 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
619 err == -EEXIST ? "can not change invariant setting" :
620 "invalid attribute value";
623 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
625 struct set_role_parms parms;
627 enum drbd_ret_code retcode;
629 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
630 if (!adm_ctx.reply_skb)
632 if (retcode != NO_ERROR)
635 memset(&parms, 0, sizeof(parms));
636 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
637 err = set_role_parms_from_attrs(&parms, info);
639 retcode = ERR_MANDATORY_TAG;
640 drbd_msg_put_info(from_attrs_err_to_txt(err));
645 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
646 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
648 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
650 drbd_adm_finish(info, retcode);
654 /* initializes the md.*_offset members, so we are able to find
655 * the on disk meta data */
656 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
657 struct drbd_backing_dev *bdev)
659 sector_t md_size_sect = 0;
660 switch (bdev->dc.meta_dev_idx) {
662 /* v07 style fixed size indexed meta data */
663 bdev->md.md_size_sect = MD_RESERVED_SECT;
664 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
665 bdev->md.al_offset = MD_AL_OFFSET;
666 bdev->md.bm_offset = MD_BM_OFFSET;
668 case DRBD_MD_INDEX_FLEX_EXT:
669 /* just occupy the full device; unit: sectors */
670 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
671 bdev->md.md_offset = 0;
672 bdev->md.al_offset = MD_AL_OFFSET;
673 bdev->md.bm_offset = MD_BM_OFFSET;
675 case DRBD_MD_INDEX_INTERNAL:
676 case DRBD_MD_INDEX_FLEX_INT:
677 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
678 /* al size is still fixed */
679 bdev->md.al_offset = -MD_AL_SECTORS;
680 /* we need (slightly less than) ~ this much bitmap sectors: */
681 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
682 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
683 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
684 md_size_sect = ALIGN(md_size_sect, 8);
686 /* plus the "drbd meta data super block",
687 * and the activity log; */
688 md_size_sect += MD_BM_OFFSET;
690 bdev->md.md_size_sect = md_size_sect;
691 /* bitmap offset is adjusted by 'super' block size */
692 bdev->md.bm_offset = -md_size_sect + MD_AL_OFFSET;
697 /* input size is expected to be in KB */
698 char *ppsize(char *buf, unsigned long long size)
700 /* Needs 9 bytes at max including trailing NUL:
701 * -1ULL ==> "16384 EB" */
702 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
704 while (size >= 10000 && base < sizeof(units)-1) {
706 size = (size >> 10) + !!(size & (1<<9));
709 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
714 /* there is still a theoretical deadlock when called from receiver
715 * on an D_INCONSISTENT R_PRIMARY:
716 * remote READ does inc_ap_bio, receiver would need to receive answer
717 * packet from remote to dec_ap_bio again.
718 * receiver receive_sizes(), comes here,
719 * waits for ap_bio_cnt == 0. -> deadlock.
720 * but this cannot happen, actually, because:
721 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
722 * (not connected, or bad/no disk on peer):
723 * see drbd_fail_request_early, ap_bio_cnt is zero.
724 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
725 * peer may not initiate a resize.
727 /* Note these are not to be confused with
728 * drbd_adm_suspend_io/drbd_adm_resume_io,
729 * which are (sub) state changes triggered by admin (drbdsetup),
730 * and can be long lived.
731 * This changes an mdev->flag, is triggered by drbd internals,
732 * and should be short-lived. */
733 void drbd_suspend_io(struct drbd_conf *mdev)
735 set_bit(SUSPEND_IO, &mdev->flags);
736 if (drbd_suspended(mdev))
738 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
741 void drbd_resume_io(struct drbd_conf *mdev)
743 clear_bit(SUSPEND_IO, &mdev->flags);
744 wake_up(&mdev->misc_wait);
748 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
749 * @mdev: DRBD device.
751 * Returns 0 on success, negative return values indicate errors.
752 * You should call drbd_md_sync() after calling this function.
754 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
756 sector_t prev_first_sect, prev_size; /* previous meta location */
761 int md_moved, la_size_changed;
762 enum determine_dev_size rv = unchanged;
765 * application request passes inc_ap_bio,
766 * but then cannot get an AL-reference.
767 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
770 * Suspend IO right here.
771 * still lock the act_log to not trigger ASSERTs there.
773 drbd_suspend_io(mdev);
775 /* no wait necessary anymore, actually we could assert that */
776 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
778 prev_first_sect = drbd_md_first_sector(mdev->ldev);
779 prev_size = mdev->ldev->md.md_size_sect;
780 la_size = mdev->ldev->md.la_size_sect;
782 /* TODO: should only be some assert here, not (re)init... */
783 drbd_md_set_sector_offsets(mdev, mdev->ldev);
785 size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
787 if (drbd_get_capacity(mdev->this_bdev) != size ||
788 drbd_bm_capacity(mdev) != size) {
790 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
792 /* currently there is only one error: ENOMEM! */
793 size = drbd_bm_capacity(mdev)>>1;
795 dev_err(DEV, "OUT OF MEMORY! "
796 "Could not allocate bitmap!\n");
798 dev_err(DEV, "BM resizing failed. "
799 "Leaving size unchanged at size = %lu KB\n",
800 (unsigned long)size);
804 /* racy, see comments above. */
805 drbd_set_my_capacity(mdev, size);
806 mdev->ldev->md.la_size_sect = size;
807 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
808 (unsigned long long)size>>1);
810 if (rv == dev_size_error)
813 la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
815 md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
816 || prev_size != mdev->ldev->md.md_size_sect;
818 if (la_size_changed || md_moved) {
821 drbd_al_shrink(mdev); /* All extents inactive. */
822 dev_info(DEV, "Writing the whole bitmap, %s\n",
823 la_size_changed && md_moved ? "size changed and md moved" :
824 la_size_changed ? "size changed" : "md moved");
825 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
826 err = drbd_bitmap_io(mdev, &drbd_bm_write,
827 "size changed", BM_LOCKED_MASK);
832 drbd_md_mark_dirty(mdev);
840 lc_unlock(mdev->act_log);
841 wake_up(&mdev->al_wait);
842 drbd_resume_io(mdev);
848 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
850 sector_t p_size = mdev->p_size; /* partner's disk size. */
851 sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
852 sector_t m_size; /* my size */
853 sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
856 m_size = drbd_get_max_capacity(bdev);
858 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
859 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
863 if (p_size && m_size) {
864 size = min_t(sector_t, p_size, m_size);
868 if (m_size && m_size < size)
870 if (p_size && p_size < size)
881 dev_err(DEV, "Both nodes diskless!\n");
885 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
886 (unsigned long)u_size>>1, (unsigned long)size>>1);
895 * drbd_check_al_size() - Ensures that the AL is of the right size
896 * @mdev: DRBD device.
898 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
899 * failed, and 0 on success. You should call drbd_md_sync() after you called
902 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
904 struct lru_cache *n, *t;
905 struct lc_element *e;
909 if (!expect(dc->al_extents >= DRBD_AL_EXTENTS_MIN))
910 dc->al_extents = DRBD_AL_EXTENTS_MIN;
913 mdev->act_log->nr_elements == dc->al_extents)
918 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
919 dc->al_extents, sizeof(struct lc_element), 0);
922 dev_err(DEV, "Cannot allocate act_log lru!\n");
925 spin_lock_irq(&mdev->al_lock);
927 for (i = 0; i < t->nr_elements; i++) {
928 e = lc_element_by_index(t, i);
930 dev_err(DEV, "refcnt(%d)==%d\n",
931 e->lc_number, e->refcnt);
937 spin_unlock_irq(&mdev->al_lock);
939 dev_err(DEV, "Activity log still in use!\n");
946 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
950 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
952 struct request_queue * const q = mdev->rq_queue;
953 int max_hw_sectors = max_bio_size >> 9;
954 int max_segments = 0;
956 if (get_ldev_if_state(mdev, D_ATTACHING)) {
957 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
959 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
960 max_segments = mdev->ldev->dc.max_bio_bvecs;
964 blk_queue_logical_block_size(q, 512);
965 blk_queue_max_hw_sectors(q, max_hw_sectors);
966 /* This is the workaround for "bio would need to, but cannot, be split" */
967 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
968 blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
970 if (get_ldev_if_state(mdev, D_ATTACHING)) {
971 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
973 blk_queue_stack_limits(q, b);
975 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
976 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
977 q->backing_dev_info.ra_pages,
978 b->backing_dev_info.ra_pages);
979 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
985 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
987 int now, new, local, peer;
989 now = queue_max_hw_sectors(mdev->rq_queue) << 9;
990 local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
991 peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
993 if (get_ldev_if_state(mdev, D_ATTACHING)) {
994 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
995 mdev->local_max_bio_size = local;
999 /* We may ignore peer limits if the peer is modern enough.
1000 Because new from 8.3.8 onwards the peer can use multiple
1001 BIOs for a single peer_request */
1002 if (mdev->state.conn >= C_CONNECTED) {
1003 if (mdev->tconn->agreed_pro_version < 94)
1004 peer = mdev->peer_max_bio_size;
1005 else if (mdev->tconn->agreed_pro_version == 94)
1006 peer = DRBD_MAX_SIZE_H80_PACKET;
1007 else /* drbd 8.3.8 onwards */
1008 peer = DRBD_MAX_BIO_SIZE;
1011 new = min_t(int, local, peer);
1013 if (mdev->state.role == R_PRIMARY && new < now)
1014 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
1017 dev_info(DEV, "max BIO size = %u\n", new);
1019 drbd_setup_queue_param(mdev, new);
1022 /* serialize deconfig (worker exiting, doing cleanup)
1023 * and reconfig (drbdsetup disk, drbdsetup net)
1025 * Wait for a potentially exiting worker, then restart it,
1026 * or start a new one. Flush any pending work, there may still be an
1027 * after_state_change queued.
1029 static void conn_reconfig_start(struct drbd_tconn *tconn)
1031 wait_event(tconn->ping_wait, !test_and_set_bit(CONFIG_PENDING, &tconn->flags));
1032 wait_event(tconn->ping_wait, !test_bit(OBJECT_DYING, &tconn->flags));
1033 drbd_thread_start(&tconn->worker);
1034 conn_flush_workqueue(tconn);
1037 /* if still unconfigured, stops worker again.
1038 * if configured now, clears CONFIG_PENDING.
1039 * wakes potential waiters */
1040 static void conn_reconfig_done(struct drbd_tconn *tconn)
1042 spin_lock_irq(&tconn->req_lock);
1043 if (conn_all_vols_unconf(tconn)) {
1044 set_bit(OBJECT_DYING, &tconn->flags);
1045 drbd_thread_stop_nowait(&tconn->worker);
1047 clear_bit(CONFIG_PENDING, &tconn->flags);
1048 spin_unlock_irq(&tconn->req_lock);
1049 wake_up(&tconn->ping_wait);
1052 /* Make sure IO is suspended before calling this function(). */
1053 static void drbd_suspend_al(struct drbd_conf *mdev)
1057 if (!lc_try_lock(mdev->act_log)) {
1058 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1062 drbd_al_shrink(mdev);
1063 spin_lock_irq(&mdev->tconn->req_lock);
1064 if (mdev->state.conn < C_CONNECTED)
1065 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1066 spin_unlock_irq(&mdev->tconn->req_lock);
1067 lc_unlock(mdev->act_log);
1070 dev_info(DEV, "Suspended AL updates\n");
1073 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1075 enum drbd_ret_code retcode;
1076 struct drbd_conf *mdev;
1077 struct disk_conf *ndc; /* new disk conf */
1079 int *rs_plan_s = NULL;
1081 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1082 if (!adm_ctx.reply_skb)
1084 if (retcode != NO_ERROR)
1087 mdev = adm_ctx.mdev;
1089 /* we also need a disk
1090 * to change the options on */
1091 if (!get_ldev(mdev)) {
1092 retcode = ERR_NO_DISK;
1096 /* FIXME freeze IO, cluster wide.
1098 * We should make sure no-one uses
1099 * some half-updated struct when we
1100 * assign it later. */
1102 ndc = kmalloc(sizeof(*ndc), GFP_KERNEL);
1104 retcode = ERR_NOMEM;
1108 memcpy(ndc, &mdev->ldev->dc, sizeof(*ndc));
1109 err = disk_conf_from_attrs_for_change(ndc, info);
1111 retcode = ERR_MANDATORY_TAG;
1112 drbd_msg_put_info(from_attrs_err_to_txt(err));
1115 if (!expect(ndc->resync_rate >= 1))
1116 ndc->resync_rate = 1;
1118 /* clip to allowed range */
1119 if (!expect(ndc->al_extents >= DRBD_AL_EXTENTS_MIN))
1120 ndc->al_extents = DRBD_AL_EXTENTS_MIN;
1121 if (!expect(ndc->al_extents <= DRBD_AL_EXTENTS_MAX))
1122 ndc->al_extents = DRBD_AL_EXTENTS_MAX;
1124 /* most sanity checks done, try to assign the new sync-after
1125 * dependency. need to hold the global lock in there,
1126 * to avoid a race in the dependency loop check. */
1127 retcode = drbd_alter_sa(mdev, ndc->resync_after);
1128 if (retcode != NO_ERROR)
1131 fifo_size = (ndc->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1132 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1133 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1135 dev_err(DEV, "kmalloc of fifo_buffer failed");
1136 retcode = ERR_NOMEM;
1141 if (fifo_size != mdev->rs_plan_s.size) {
1142 kfree(mdev->rs_plan_s.values);
1143 mdev->rs_plan_s.values = rs_plan_s;
1144 mdev->rs_plan_s.size = fifo_size;
1145 mdev->rs_planed = 0;
1149 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1150 drbd_al_shrink(mdev);
1151 err = drbd_check_al_size(mdev, ndc);
1152 lc_unlock(mdev->act_log);
1153 wake_up(&mdev->al_wait);
1156 retcode = ERR_NOMEM;
1161 * To avoid someone looking at a half-updated struct, we probably
1162 * should have a rw-semaphor on net_conf and disk_conf.
1164 mdev->ldev->dc = *ndc;
1169 if (mdev->state.conn >= C_CONNECTED)
1170 drbd_send_sync_param(mdev);
1177 drbd_adm_finish(info, retcode);
1181 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1183 struct drbd_conf *mdev;
1185 enum drbd_ret_code retcode;
1186 enum determine_dev_size dd;
1187 sector_t max_possible_sectors;
1188 sector_t min_md_device_sectors;
1189 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1190 struct block_device *bdev;
1191 struct lru_cache *resync_lru = NULL;
1192 union drbd_state ns, os;
1193 enum drbd_state_rv rv;
1194 int cp_discovered = 0;
1196 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1197 if (!adm_ctx.reply_skb)
1199 if (retcode != NO_ERROR)
1202 mdev = adm_ctx.mdev;
1203 conn_reconfig_start(mdev->tconn);
1205 /* if you want to reconfigure, please tear down first */
1206 if (mdev->state.disk > D_DISKLESS) {
1207 retcode = ERR_DISK_CONFIGURED;
1210 /* It may just now have detached because of IO error. Make sure
1211 * drbd_ldev_destroy is done already, we may end up here very fast,
1212 * e.g. if someone calls attach from the on-io-error handler,
1213 * to realize a "hot spare" feature (not that I'd recommend that) */
1214 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1216 /* allocation not in the IO path, drbdsetup context */
1217 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1219 retcode = ERR_NOMEM;
1223 nbc->dc = (struct disk_conf) {
1224 {}, 0, /* backing_dev */
1225 {}, 0, /* meta_dev */
1226 0, /* meta_dev_idx */
1227 DRBD_DISK_SIZE_SECT_DEF, /* disk_size */
1228 DRBD_MAX_BIO_BVECS_DEF, /* max_bio_bvecs */
1229 DRBD_ON_IO_ERROR_DEF, /* on_io_error */
1230 DRBD_FENCING_DEF, /* fencing */
1231 DRBD_RATE_DEF, /* resync_rate */
1232 DRBD_AFTER_DEF, /* resync_after */
1233 DRBD_AL_EXTENTS_DEF, /* al_extents */
1234 DRBD_C_PLAN_AHEAD_DEF, /* c_plan_ahead */
1235 DRBD_C_DELAY_TARGET_DEF, /* c_delay_target */
1236 DRBD_C_FILL_TARGET_DEF, /* c_fill_target */
1237 DRBD_C_MAX_RATE_DEF, /* c_max_rate */
1238 DRBD_C_MIN_RATE_DEF, /* c_min_rate */
1239 0, /* no_disk_barrier */
1240 0, /* no_disk_flush */
1241 0, /* no_disk_drain */
1242 0, /* no_md_flush */
1245 err = disk_conf_from_attrs(&nbc->dc, info);
1247 retcode = ERR_MANDATORY_TAG;
1248 drbd_msg_put_info(from_attrs_err_to_txt(err));
1252 if ((int)nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1253 retcode = ERR_MD_IDX_INVALID;
1257 if (get_net_conf(mdev->tconn)) {
1258 int prot = mdev->tconn->net_conf->wire_protocol;
1259 put_net_conf(mdev->tconn);
1260 if (nbc->dc.fencing == FP_STONITH && prot == DRBD_PROT_A) {
1261 retcode = ERR_STONITH_AND_PROT_A;
1266 bdev = blkdev_get_by_path(nbc->dc.backing_dev,
1267 FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1269 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
1271 retcode = ERR_OPEN_DISK;
1274 nbc->backing_bdev = bdev;
1277 * meta_dev_idx >= 0: external fixed size, possibly multiple
1278 * drbd sharing one meta device. TODO in that case, paranoia
1279 * check that [md_bdev, meta_dev_idx] is not yet used by some
1280 * other drbd minor! (if you use drbd.conf + drbdadm, that
1281 * should check it for you already; but if you don't, or
1282 * someone fooled it, we need to double check here)
1284 bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1285 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1286 ((int)nbc->dc.meta_dev_idx < 0) ?
1287 (void *)mdev : (void *)drbd_m_holder);
1289 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1291 retcode = ERR_OPEN_MD_DISK;
1294 nbc->md_bdev = bdev;
1296 if ((nbc->backing_bdev == nbc->md_bdev) !=
1297 (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1298 nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1299 retcode = ERR_MD_IDX_INVALID;
1303 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1304 1, 61, sizeof(struct bm_extent),
1305 offsetof(struct bm_extent, lce));
1307 retcode = ERR_NOMEM;
1311 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1312 drbd_md_set_sector_offsets(mdev, nbc);
1314 if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1315 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1316 (unsigned long long) drbd_get_max_capacity(nbc),
1317 (unsigned long long) nbc->dc.disk_size);
1318 retcode = ERR_DISK_TO_SMALL;
1322 if ((int)nbc->dc.meta_dev_idx < 0) {
1323 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1324 /* at least one MB, otherwise it does not make sense */
1325 min_md_device_sectors = (2<<10);
1327 max_possible_sectors = DRBD_MAX_SECTORS;
1328 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1331 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1332 retcode = ERR_MD_DISK_TO_SMALL;
1333 dev_warn(DEV, "refusing attach: md-device too small, "
1334 "at least %llu sectors needed for this meta-disk type\n",
1335 (unsigned long long) min_md_device_sectors);
1339 /* Make sure the new disk is big enough
1340 * (we may currently be R_PRIMARY with no local disk...) */
1341 if (drbd_get_max_capacity(nbc) <
1342 drbd_get_capacity(mdev->this_bdev)) {
1343 retcode = ERR_DISK_TO_SMALL;
1347 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1349 if (nbc->known_size > max_possible_sectors) {
1350 dev_warn(DEV, "==> truncating very big lower level device "
1351 "to currently maximum possible %llu sectors <==\n",
1352 (unsigned long long) max_possible_sectors);
1353 if ((int)nbc->dc.meta_dev_idx >= 0)
1354 dev_warn(DEV, "==>> using internal or flexible "
1355 "meta data may help <<==\n");
1358 drbd_suspend_io(mdev);
1359 /* also wait for the last barrier ack. */
1360 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1361 /* and for any other previously queued work */
1362 drbd_flush_workqueue(mdev);
1364 rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1365 retcode = rv; /* FIXME: Type mismatch. */
1366 drbd_resume_io(mdev);
1367 if (rv < SS_SUCCESS)
1370 if (!get_ldev_if_state(mdev, D_ATTACHING))
1371 goto force_diskless;
1373 drbd_md_set_sector_offsets(mdev, nbc);
1375 if (!mdev->bitmap) {
1376 if (drbd_bm_init(mdev)) {
1377 retcode = ERR_NOMEM;
1378 goto force_diskless_dec;
1382 retcode = drbd_md_read(mdev, nbc);
1383 if (retcode != NO_ERROR)
1384 goto force_diskless_dec;
1386 if (mdev->state.conn < C_CONNECTED &&
1387 mdev->state.role == R_PRIMARY &&
1388 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1389 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1390 (unsigned long long)mdev->ed_uuid);
1391 retcode = ERR_DATA_NOT_CURRENT;
1392 goto force_diskless_dec;
1395 /* Since we are diskless, fix the activity log first... */
1396 if (drbd_check_al_size(mdev, &nbc->dc)) {
1397 retcode = ERR_NOMEM;
1398 goto force_diskless_dec;
1401 /* Prevent shrinking of consistent devices ! */
1402 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1403 drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1404 dev_warn(DEV, "refusing to truncate a consistent device\n");
1405 retcode = ERR_DISK_TO_SMALL;
1406 goto force_diskless_dec;
1409 if (!drbd_al_read_log(mdev, nbc)) {
1410 retcode = ERR_IO_MD_DISK;
1411 goto force_diskless_dec;
1414 /* Reset the "barriers don't work" bits here, then force meta data to
1415 * be written, to ensure we determine if barriers are supported. */
1416 if (nbc->dc.no_md_flush)
1417 set_bit(MD_NO_FUA, &mdev->flags);
1419 clear_bit(MD_NO_FUA, &mdev->flags);
1421 /* Point of no return reached.
1422 * Devices and memory are no longer released by error cleanup below.
1423 * now mdev takes over responsibility, and the state engine should
1424 * clean it up somewhere. */
1425 D_ASSERT(mdev->ldev == NULL);
1427 mdev->resync = resync_lru;
1431 mdev->write_ordering = WO_bdev_flush;
1432 drbd_bump_write_ordering(mdev, WO_bdev_flush);
1434 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1435 set_bit(CRASHED_PRIMARY, &mdev->flags);
1437 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1439 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1440 !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod)) {
1441 set_bit(CRASHED_PRIMARY, &mdev->flags);
1450 drbd_reconsider_max_bio_size(mdev);
1452 /* If I am currently not R_PRIMARY,
1453 * but meta data primary indicator is set,
1454 * I just now recover from a hard crash,
1455 * and have been R_PRIMARY before that crash.
1457 * Now, if I had no connection before that crash
1458 * (have been degraded R_PRIMARY), chances are that
1459 * I won't find my peer now either.
1461 * In that case, and _only_ in that case,
1462 * we use the degr-wfc-timeout instead of the default,
1463 * so we can automatically recover from a crash of a
1464 * degraded but active "cluster" after a certain timeout.
1466 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1467 if (mdev->state.role != R_PRIMARY &&
1468 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1469 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1470 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1472 dd = drbd_determine_dev_size(mdev, 0);
1473 if (dd == dev_size_error) {
1474 retcode = ERR_NOMEM_BITMAP;
1475 goto force_diskless_dec;
1476 } else if (dd == grew)
1477 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1479 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1480 dev_info(DEV, "Assuming that all blocks are out of sync "
1481 "(aka FullSync)\n");
1482 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1483 "set_n_write from attaching", BM_LOCKED_MASK)) {
1484 retcode = ERR_IO_MD_DISK;
1485 goto force_diskless_dec;
1488 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1489 "read from attaching", BM_LOCKED_MASK)) {
1490 retcode = ERR_IO_MD_DISK;
1491 goto force_diskless_dec;
1495 if (cp_discovered) {
1496 drbd_al_apply_to_bm(mdev);
1497 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1498 "crashed primary apply AL", BM_LOCKED_MASK)) {
1499 retcode = ERR_IO_MD_DISK;
1500 goto force_diskless_dec;
1504 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1505 drbd_suspend_al(mdev); /* IO is still suspended here... */
1507 spin_lock_irq(&mdev->tconn->req_lock);
1508 os = drbd_read_state(mdev);
1510 /* If MDF_CONSISTENT is not set go into inconsistent state,
1511 otherwise investigate MDF_WasUpToDate...
1512 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1513 otherwise into D_CONSISTENT state.
1515 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1516 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1517 ns.disk = D_CONSISTENT;
1519 ns.disk = D_OUTDATED;
1521 ns.disk = D_INCONSISTENT;
1524 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1525 ns.pdsk = D_OUTDATED;
1527 if ( ns.disk == D_CONSISTENT &&
1528 (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1529 ns.disk = D_UP_TO_DATE;
1531 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1532 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1533 this point, because drbd_request_state() modifies these
1536 /* In case we are C_CONNECTED postpone any decision on the new disk
1537 state after the negotiation phase. */
1538 if (mdev->state.conn == C_CONNECTED) {
1539 mdev->new_state_tmp.i = ns.i;
1541 ns.disk = D_NEGOTIATING;
1543 /* We expect to receive up-to-date UUIDs soon.
1544 To avoid a race in receive_state, free p_uuid while
1545 holding req_lock. I.e. atomic with the state change */
1546 kfree(mdev->p_uuid);
1547 mdev->p_uuid = NULL;
1550 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1551 spin_unlock_irq(&mdev->tconn->req_lock);
1553 if (rv < SS_SUCCESS)
1554 goto force_diskless_dec;
1556 if (mdev->state.role == R_PRIMARY)
1557 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1559 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1561 drbd_md_mark_dirty(mdev);
1564 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1566 conn_reconfig_done(mdev->tconn);
1567 drbd_adm_finish(info, retcode);
1573 drbd_force_state(mdev, NS(disk, D_FAILED));
1576 conn_reconfig_done(mdev->tconn);
1578 if (nbc->backing_bdev)
1579 blkdev_put(nbc->backing_bdev,
1580 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1582 blkdev_put(nbc->md_bdev,
1583 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1586 lc_destroy(resync_lru);
1589 drbd_adm_finish(info, retcode);
1593 static int adm_detach(struct drbd_conf *mdev)
1595 enum drbd_state_rv retcode;
1596 drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1597 retcode = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1598 wait_event(mdev->misc_wait,
1599 mdev->state.disk != D_DISKLESS ||
1600 !atomic_read(&mdev->local_cnt));
1601 drbd_resume_io(mdev);
1605 /* Detaching the disk is a process in multiple stages. First we need to lock
1606 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1607 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1608 * internal references as well.
1609 * Only then we have finally detached. */
1610 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1612 enum drbd_ret_code retcode;
1614 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1615 if (!adm_ctx.reply_skb)
1617 if (retcode != NO_ERROR)
1620 retcode = adm_detach(adm_ctx.mdev);
1622 drbd_adm_finish(info, retcode);
1626 static bool conn_resync_running(struct drbd_tconn *tconn)
1628 struct drbd_conf *mdev;
1633 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1634 if (mdev->state.conn == C_SYNC_SOURCE ||
1635 mdev->state.conn == C_SYNC_TARGET ||
1636 mdev->state.conn == C_PAUSED_SYNC_S ||
1637 mdev->state.conn == C_PAUSED_SYNC_T) {
1647 static bool conn_ov_running(struct drbd_tconn *tconn)
1649 struct drbd_conf *mdev;
1654 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1655 if (mdev->state.conn == C_VERIFY_S ||
1656 mdev->state.conn == C_VERIFY_T) {
1666 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1668 enum drbd_ret_code retcode;
1669 struct drbd_tconn *tconn;
1670 struct net_conf *new_conf = NULL;
1672 int ovr; /* online verify running */
1673 int rsr; /* re-sync running */
1674 struct crypto_hash *verify_tfm = NULL;
1675 struct crypto_hash *csums_tfm = NULL;
1678 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1679 if (!adm_ctx.reply_skb)
1681 if (retcode != NO_ERROR)
1684 tconn = adm_ctx.tconn;
1686 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1688 retcode = ERR_NOMEM;
1692 /* we also need a net config
1693 * to change the options on */
1694 if (!get_net_conf(tconn)) {
1695 drbd_msg_put_info("net conf missing, try connect");
1696 retcode = ERR_INVALID_REQUEST;
1700 conn_reconfig_start(tconn);
1702 memcpy(new_conf, tconn->net_conf, sizeof(*new_conf));
1703 err = net_conf_from_attrs_for_change(new_conf, info);
1705 retcode = ERR_MANDATORY_TAG;
1706 drbd_msg_put_info(from_attrs_err_to_txt(err));
1710 /* re-sync running */
1711 rsr = conn_resync_running(tconn);
1712 if (rsr && strcmp(new_conf->csums_alg, tconn->net_conf->csums_alg)) {
1713 retcode = ERR_CSUMS_RESYNC_RUNNING;
1717 if (!rsr && new_conf->csums_alg[0]) {
1718 csums_tfm = crypto_alloc_hash(new_conf->csums_alg, 0, CRYPTO_ALG_ASYNC);
1719 if (IS_ERR(csums_tfm)) {
1721 retcode = ERR_CSUMS_ALG;
1725 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1726 retcode = ERR_CSUMS_ALG_ND;
1731 /* online verify running */
1732 ovr = conn_ov_running(tconn);
1734 if (strcmp(new_conf->verify_alg, tconn->net_conf->verify_alg)) {
1735 retcode = ERR_VERIFY_RUNNING;
1740 if (!ovr && new_conf->verify_alg[0]) {
1741 verify_tfm = crypto_alloc_hash(new_conf->verify_alg, 0, CRYPTO_ALG_ASYNC);
1742 if (IS_ERR(verify_tfm)) {
1744 retcode = ERR_VERIFY_ALG;
1748 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1749 retcode = ERR_VERIFY_ALG_ND;
1755 /* For now, use struct assignment, not pointer assignment.
1756 * We don't have any means to determine who might still
1757 * keep a local alias into the struct,
1758 * so we cannot just free it and hope for the best :(
1760 * To avoid someone looking at a half-updated struct, we probably
1761 * should have a rw-semaphor on net_conf and disk_conf.
1763 *tconn->net_conf = *new_conf;
1766 crypto_free_hash(tconn->csums_tfm);
1767 tconn->csums_tfm = csums_tfm;
1771 crypto_free_hash(tconn->verify_tfm);
1772 tconn->verify_tfm = verify_tfm;
1776 if (tconn->cstate >= C_WF_REPORT_PARAMS)
1777 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1780 crypto_free_hash(csums_tfm);
1781 crypto_free_hash(verify_tfm);
1783 put_net_conf(tconn);
1784 conn_reconfig_done(tconn);
1786 drbd_adm_finish(info, retcode);
1790 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1792 char hmac_name[CRYPTO_MAX_ALG_NAME];
1793 struct drbd_conf *mdev;
1794 struct net_conf *new_conf = NULL;
1795 struct crypto_hash *tfm = NULL;
1796 struct crypto_hash *integrity_w_tfm = NULL;
1797 struct crypto_hash *integrity_r_tfm = NULL;
1798 void *int_dig_in = NULL;
1799 void *int_dig_vv = NULL;
1800 struct drbd_tconn *oconn;
1801 struct drbd_tconn *tconn;
1802 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1803 enum drbd_ret_code retcode;
1807 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1808 if (!adm_ctx.reply_skb)
1810 if (retcode != NO_ERROR)
1813 tconn = adm_ctx.tconn;
1814 conn_reconfig_start(tconn);
1816 if (tconn->cstate > C_STANDALONE) {
1817 retcode = ERR_NET_CONFIGURED;
1821 /* allocation not in the IO path, cqueue thread context */
1822 new_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL);
1824 retcode = ERR_NOMEM;
1828 *new_conf = (struct net_conf) {
1829 {}, 0, /* my_addr */
1830 {}, 0, /* peer_addr */
1831 {}, 0, /* shared_secret */
1832 {}, 0, /* cram_hmac_alg */
1833 {}, 0, /* integrity_alg */
1834 {}, 0, /* verify_alg */
1835 {}, 0, /* csums_alg */
1836 DRBD_PROTOCOL_DEF, /* wire_protocol */
1837 DRBD_CONNECT_INT_DEF, /* try_connect_int */
1838 DRBD_TIMEOUT_DEF, /* timeout */
1839 DRBD_PING_INT_DEF, /* ping_int */
1840 DRBD_PING_TIMEO_DEF, /* ping_timeo */
1841 DRBD_SNDBUF_SIZE_DEF, /* sndbuf_size */
1842 DRBD_RCVBUF_SIZE_DEF, /* rcvbuf_size */
1843 DRBD_KO_COUNT_DEF, /* ko_count */
1844 DRBD_MAX_BUFFERS_DEF, /* max_buffers */
1845 DRBD_MAX_EPOCH_SIZE_DEF, /* max_epoch_size */
1846 DRBD_UNPLUG_WATERMARK_DEF, /* unplug_watermark */
1847 DRBD_AFTER_SB_0P_DEF, /* after_sb_0p */
1848 DRBD_AFTER_SB_1P_DEF, /* after_sb_1p */
1849 DRBD_AFTER_SB_2P_DEF, /* after_sb_2p */
1850 DRBD_RR_CONFLICT_DEF, /* rr_conflict */
1851 DRBD_ON_CONGESTION_DEF, /* on_congestion */
1852 DRBD_CONG_FILL_DEF, /* cong_fill */
1853 DRBD_CONG_EXTENTS_DEF, /* cong_extents */
1854 0, /* two_primaries */
1857 0, /* always_asbp */
1862 err = net_conf_from_attrs(new_conf, info);
1864 retcode = ERR_MANDATORY_TAG;
1865 drbd_msg_put_info(from_attrs_err_to_txt(err));
1869 if (new_conf->two_primaries
1870 && (new_conf->wire_protocol != DRBD_PROT_C)) {
1871 retcode = ERR_NOT_PROTO_C;
1876 idr_for_each_entry(&tconn->volumes, mdev, i) {
1877 if (get_ldev(mdev)) {
1878 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1880 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH) {
1881 retcode = ERR_STONITH_AND_PROT_A;
1882 goto fail_rcu_unlock;
1885 if (mdev->state.role == R_PRIMARY && new_conf->want_lose) {
1886 retcode = ERR_DISCARD;
1887 goto fail_rcu_unlock;
1889 if (!mdev->bitmap) {
1890 if(drbd_bm_init(mdev)) {
1891 retcode = ERR_NOMEM;
1892 goto fail_rcu_unlock;
1898 if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A) {
1899 retcode = ERR_CONG_NOT_PROTO_A;
1905 new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1906 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1908 /* No need to take drbd_cfg_mutex here. All reconfiguration is
1909 * strictly serialized on genl_lock(). We are protected against
1910 * concurrent reconfiguration/addition/deletion */
1911 list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
1914 if (get_net_conf(oconn)) {
1915 taken_addr = (struct sockaddr *)&oconn->net_conf->my_addr;
1916 if (new_conf->my_addr_len == oconn->net_conf->my_addr_len &&
1917 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1918 retcode = ERR_LOCAL_ADDR;
1920 taken_addr = (struct sockaddr *)&oconn->net_conf->peer_addr;
1921 if (new_conf->peer_addr_len == oconn->net_conf->peer_addr_len &&
1922 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1923 retcode = ERR_PEER_ADDR;
1925 put_net_conf(oconn);
1926 if (retcode != NO_ERROR)
1931 if (new_conf->cram_hmac_alg[0] != 0) {
1932 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1933 new_conf->cram_hmac_alg);
1934 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1937 retcode = ERR_AUTH_ALG;
1941 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1942 retcode = ERR_AUTH_ALG_ND;
1947 if (new_conf->integrity_alg[0]) {
1948 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1949 if (IS_ERR(integrity_w_tfm)) {
1950 integrity_w_tfm = NULL;
1951 retcode=ERR_INTEGRITY_ALG;
1955 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1956 retcode=ERR_INTEGRITY_ALG_ND;
1960 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1961 if (IS_ERR(integrity_r_tfm)) {
1962 integrity_r_tfm = NULL;
1963 retcode=ERR_INTEGRITY_ALG;
1968 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
1970 /* allocation not in the IO path, cqueue thread context */
1971 if (integrity_w_tfm) {
1972 i = crypto_hash_digestsize(integrity_w_tfm);
1973 int_dig_in = kmalloc(i, GFP_KERNEL);
1975 retcode = ERR_NOMEM;
1978 int_dig_vv = kmalloc(i, GFP_KERNEL);
1980 retcode = ERR_NOMEM;
1985 conn_flush_workqueue(tconn);
1986 spin_lock_irq(&tconn->req_lock);
1987 if (tconn->net_conf != NULL) {
1988 retcode = ERR_NET_CONFIGURED;
1989 spin_unlock_irq(&tconn->req_lock);
1992 tconn->net_conf = new_conf;
1994 crypto_free_hash(tconn->cram_hmac_tfm);
1995 tconn->cram_hmac_tfm = tfm;
1997 crypto_free_hash(tconn->integrity_w_tfm);
1998 tconn->integrity_w_tfm = integrity_w_tfm;
2000 crypto_free_hash(tconn->integrity_r_tfm);
2001 tconn->integrity_r_tfm = integrity_r_tfm;
2003 kfree(tconn->int_dig_in);
2004 kfree(tconn->int_dig_vv);
2005 tconn->int_dig_in=int_dig_in;
2006 tconn->int_dig_vv=int_dig_vv;
2007 retcode = _conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2008 spin_unlock_irq(&tconn->req_lock);
2011 idr_for_each_entry(&tconn->volumes, mdev, i) {
2014 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
2017 conn_reconfig_done(tconn);
2018 drbd_adm_finish(info, retcode);
2026 crypto_free_hash(tfm);
2027 crypto_free_hash(integrity_w_tfm);
2028 crypto_free_hash(integrity_r_tfm);
2031 conn_reconfig_done(tconn);
2033 drbd_adm_finish(info, retcode);
2037 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2039 enum drbd_state_rv rv;
2041 spin_lock_irq(&tconn->req_lock);
2042 if (tconn->cstate >= C_WF_CONNECTION)
2043 _conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
2044 spin_unlock_irq(&tconn->req_lock);
2048 rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING), 0);
2051 case SS_NOTHING_TO_DO:
2052 case SS_ALREADY_STANDALONE:
2054 case SS_PRIMARY_NOP:
2055 /* Our state checking code wants to see the peer outdated. */
2056 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2057 pdsk, D_OUTDATED), CS_VERBOSE);
2059 case SS_CW_FAILED_BY_PEER:
2060 /* The peer probably wants to see us outdated. */
2061 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2062 disk, D_OUTDATED), 0);
2063 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2064 conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
2069 /* no special handling necessary */
2075 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2077 struct disconnect_parms parms;
2078 struct drbd_tconn *tconn;
2079 enum drbd_state_rv rv;
2080 enum drbd_ret_code retcode;
2083 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2084 if (!adm_ctx.reply_skb)
2086 if (retcode != NO_ERROR)
2089 tconn = adm_ctx.tconn;
2090 memset(&parms, 0, sizeof(parms));
2091 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2092 err = disconnect_parms_from_attrs(&parms, info);
2094 retcode = ERR_MANDATORY_TAG;
2095 drbd_msg_put_info(from_attrs_err_to_txt(err));
2100 rv = conn_try_disconnect(tconn, parms.force_disconnect);
2101 if (rv < SS_SUCCESS)
2104 if (wait_event_interruptible(tconn->ping_wait,
2105 tconn->cstate != C_DISCONNECTING)) {
2106 /* Do not test for mdev->state.conn == C_STANDALONE, since
2107 someone else might connect us in the mean time! */
2114 drbd_adm_finish(info, retcode);
2118 void resync_after_online_grow(struct drbd_conf *mdev)
2120 int iass; /* I am sync source */
2122 dev_info(DEV, "Resync of new storage after online grow\n");
2123 if (mdev->state.role != mdev->state.peer)
2124 iass = (mdev->state.role == R_PRIMARY);
2126 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2129 drbd_start_resync(mdev, C_SYNC_SOURCE);
2131 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2134 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2136 struct resize_parms rs;
2137 struct drbd_conf *mdev;
2138 enum drbd_ret_code retcode;
2139 enum determine_dev_size dd;
2140 enum dds_flags ddsf;
2143 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2144 if (!adm_ctx.reply_skb)
2146 if (retcode != NO_ERROR)
2149 memset(&rs, 0, sizeof(struct resize_parms));
2150 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2151 err = resize_parms_from_attrs(&rs, info);
2153 retcode = ERR_MANDATORY_TAG;
2154 drbd_msg_put_info(from_attrs_err_to_txt(err));
2159 mdev = adm_ctx.mdev;
2160 if (mdev->state.conn > C_CONNECTED) {
2161 retcode = ERR_RESIZE_RESYNC;
2165 if (mdev->state.role == R_SECONDARY &&
2166 mdev->state.peer == R_SECONDARY) {
2167 retcode = ERR_NO_PRIMARY;
2171 if (!get_ldev(mdev)) {
2172 retcode = ERR_NO_DISK;
2176 if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2177 retcode = ERR_NEED_APV_93;
2181 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2182 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2184 mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
2185 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2186 dd = drbd_determine_dev_size(mdev, ddsf);
2189 if (dd == dev_size_error) {
2190 retcode = ERR_NOMEM_BITMAP;
2194 if (mdev->state.conn == C_CONNECTED) {
2196 set_bit(RESIZE_PENDING, &mdev->flags);
2198 drbd_send_uuids(mdev);
2199 drbd_send_sizes(mdev, 1, ddsf);
2203 drbd_adm_finish(info, retcode);
2207 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2209 enum drbd_ret_code retcode;
2210 cpumask_var_t new_cpu_mask;
2211 struct drbd_tconn *tconn;
2212 int *rs_plan_s = NULL;
2216 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2217 if (!adm_ctx.reply_skb)
2219 if (retcode != NO_ERROR)
2221 tconn = adm_ctx.tconn;
2223 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
2224 retcode = ERR_NOMEM;
2225 drbd_msg_put_info("unable to allocate cpumask");
2229 if (((struct drbd_genlmsghdr*)info->userhdr)->flags
2230 & DRBD_GENL_F_SET_DEFAULTS) {
2231 memset(&sc, 0, sizeof(struct res_opts));
2232 sc.on_no_data = DRBD_ON_NO_DATA_DEF;
2234 sc = tconn->res_opts;
2236 err = res_opts_from_attrs(&sc, info);
2238 retcode = ERR_MANDATORY_TAG;
2239 drbd_msg_put_info(from_attrs_err_to_txt(err));
2243 /* silently ignore cpu mask on UP kernel */
2244 if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
2245 err = __bitmap_parse(sc.cpu_mask, 32, 0,
2246 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2248 conn_warn(tconn, "__bitmap_parse() failed with %d\n", err);
2249 retcode = ERR_CPU_MASK_PARSE;
2255 tconn->res_opts = sc;
2257 if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2258 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2259 drbd_calc_cpu_mask(tconn);
2260 tconn->receiver.reset_cpu_mask = 1;
2261 tconn->asender.reset_cpu_mask = 1;
2262 tconn->worker.reset_cpu_mask = 1;
2267 free_cpumask_var(new_cpu_mask);
2269 drbd_adm_finish(info, retcode);
2273 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2275 struct drbd_conf *mdev;
2276 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2278 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2279 if (!adm_ctx.reply_skb)
2281 if (retcode != NO_ERROR)
2284 mdev = adm_ctx.mdev;
2286 /* If there is still bitmap IO pending, probably because of a previous
2287 * resync just being finished, wait for it before requesting a new resync. */
2288 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2290 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2292 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2293 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2295 while (retcode == SS_NEED_CONNECTION) {
2296 spin_lock_irq(&mdev->tconn->req_lock);
2297 if (mdev->state.conn < C_CONNECTED)
2298 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2299 spin_unlock_irq(&mdev->tconn->req_lock);
2301 if (retcode != SS_NEED_CONNECTION)
2304 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2308 drbd_adm_finish(info, retcode);
2312 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2316 rv = drbd_bmio_set_n_write(mdev);
2317 drbd_suspend_al(mdev);
2321 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2322 union drbd_state mask, union drbd_state val)
2324 enum drbd_ret_code retcode;
2326 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2327 if (!adm_ctx.reply_skb)
2329 if (retcode != NO_ERROR)
2332 retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2334 drbd_adm_finish(info, retcode);
2338 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2340 return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2343 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2345 enum drbd_ret_code retcode;
2347 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2348 if (!adm_ctx.reply_skb)
2350 if (retcode != NO_ERROR)
2353 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2354 retcode = ERR_PAUSE_IS_SET;
2356 drbd_adm_finish(info, retcode);
2360 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2362 union drbd_dev_state s;
2363 enum drbd_ret_code retcode;
2365 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2366 if (!adm_ctx.reply_skb)
2368 if (retcode != NO_ERROR)
2371 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2372 s = adm_ctx.mdev->state;
2373 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2374 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2375 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2377 retcode = ERR_PAUSE_IS_CLEAR;
2382 drbd_adm_finish(info, retcode);
2386 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2388 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2391 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2393 struct drbd_conf *mdev;
2394 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2396 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2397 if (!adm_ctx.reply_skb)
2399 if (retcode != NO_ERROR)
2402 mdev = adm_ctx.mdev;
2403 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2404 drbd_uuid_new_current(mdev);
2405 clear_bit(NEW_CUR_UUID, &mdev->flags);
2407 drbd_suspend_io(mdev);
2408 retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2409 if (retcode == SS_SUCCESS) {
2410 if (mdev->state.conn < C_CONNECTED)
2411 tl_clear(mdev->tconn);
2412 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2413 tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2415 drbd_resume_io(mdev);
2418 drbd_adm_finish(info, retcode);
2422 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2424 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2427 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *conn_name, unsigned vnr)
2430 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2432 goto nla_put_failure;
2433 if (vnr != VOLUME_UNSPECIFIED)
2434 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2435 NLA_PUT_STRING(skb, T_ctx_conn_name, conn_name);
2436 nla_nest_end(skb, nla);
2441 nla_nest_cancel(skb, nla);
2445 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2446 const struct sib_info *sib)
2448 struct state_info *si = NULL; /* for sizeof(si->member); */
2453 int exclude_sensitive;
2455 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2456 * to. So we better exclude_sensitive information.
2458 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2459 * in the context of the requesting user process. Exclude sensitive
2460 * information, unless current has superuser.
2462 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2463 * relies on the current implementation of netlink_dump(), which
2464 * executes the dump callback successively from netlink_recvmsg(),
2465 * always in the context of the receiving process */
2466 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2468 got_ldev = get_ldev(mdev);
2469 got_net = get_net_conf(mdev->tconn);
2471 /* We need to add connection name and volume number information still.
2472 * Minor number is in drbd_genlmsghdr. */
2473 if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2474 goto nla_put_failure;
2476 if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2477 goto nla_put_failure;
2480 if (disk_conf_to_skb(skb, &mdev->ldev->dc, exclude_sensitive))
2481 goto nla_put_failure;
2483 if (net_conf_to_skb(skb, mdev->tconn->net_conf, exclude_sensitive))
2484 goto nla_put_failure;
2486 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2488 goto nla_put_failure;
2489 NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2490 NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2491 NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2492 NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2495 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2496 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2497 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2498 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2499 if (C_SYNC_SOURCE <= mdev->state.conn &&
2500 C_PAUSED_SYNC_T >= mdev->state.conn) {
2501 NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2502 NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2507 switch(sib->sib_reason) {
2508 case SIB_SYNC_PROGRESS:
2509 case SIB_GET_STATUS_REPLY:
2511 case SIB_STATE_CHANGE:
2512 NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2513 NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2515 case SIB_HELPER_POST:
2517 T_helper_exit_code, sib->helper_exit_code);
2519 case SIB_HELPER_PRE:
2520 NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2524 nla_nest_end(skb, nla);
2532 put_net_conf(mdev->tconn);
2536 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2538 enum drbd_ret_code retcode;
2541 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2542 if (!adm_ctx.reply_skb)
2544 if (retcode != NO_ERROR)
2547 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2549 nlmsg_free(adm_ctx.reply_skb);
2553 drbd_adm_finish(info, retcode);
2557 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2559 struct drbd_conf *mdev;
2560 struct drbd_genlmsghdr *dh;
2561 struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2562 struct drbd_tconn *tconn = NULL;
2563 struct drbd_tconn *tmp;
2564 unsigned volume = cb->args[1];
2566 /* Open coded, deferred, iteration:
2567 * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2568 * idr_for_each_entry(&tconn->volumes, mdev, i) {
2572 * where tconn is cb->args[0];
2573 * and i is cb->args[1];
2575 * This may miss entries inserted after this dump started,
2576 * or entries deleted before they are reached.
2578 * We need to make sure the mdev won't disappear while
2579 * we are looking at it, and revalidate our iterators
2580 * on each iteration.
2583 /* synchronize with drbd_new_tconn/drbd_free_tconn */
2584 mutex_lock(&drbd_cfg_mutex);
2586 /* revalidate iterator position */
2587 list_for_each_entry(tmp, &drbd_tconns, all_tconn) {
2589 /* first iteration */
2600 mdev = idr_get_next(&tconn->volumes, &volume);
2602 /* No more volumes to dump on this tconn.
2603 * Advance tconn iterator. */
2604 pos = list_entry(tconn->all_tconn.next,
2605 struct drbd_tconn, all_tconn);
2606 /* But, did we dump any volume on this tconn yet? */
2614 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2615 cb->nlh->nlmsg_seq, &drbd_genl_family,
2616 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2621 /* this is a tconn without a single volume */
2623 dh->ret_code = NO_ERROR;
2624 if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2625 genlmsg_cancel(skb, dh);
2627 genlmsg_end(skb, dh);
2631 D_ASSERT(mdev->vnr == volume);
2632 D_ASSERT(mdev->tconn == tconn);
2634 dh->minor = mdev_to_minor(mdev);
2635 dh->ret_code = NO_ERROR;
2637 if (nla_put_status_info(skb, mdev, NULL)) {
2638 genlmsg_cancel(skb, dh);
2641 genlmsg_end(skb, dh);
2645 mutex_unlock(&drbd_cfg_mutex);
2646 /* where to start the next iteration */
2647 cb->args[0] = (long)pos;
2648 cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2650 /* No more tconns/volumes/minors found results in an empty skb.
2651 * Which will terminate the dump. */
2655 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2657 enum drbd_ret_code retcode;
2658 struct timeout_parms tp;
2661 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2662 if (!adm_ctx.reply_skb)
2664 if (retcode != NO_ERROR)
2668 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2669 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2672 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2674 nlmsg_free(adm_ctx.reply_skb);
2678 drbd_adm_finish(info, retcode);
2682 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2684 struct drbd_conf *mdev;
2685 enum drbd_ret_code retcode;
2687 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2688 if (!adm_ctx.reply_skb)
2690 if (retcode != NO_ERROR)
2693 mdev = adm_ctx.mdev;
2694 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2695 /* resume from last known position, if possible */
2696 struct start_ov_parms parms =
2697 { .ov_start_sector = mdev->ov_start_sector };
2698 int err = start_ov_parms_from_attrs(&parms, info);
2700 retcode = ERR_MANDATORY_TAG;
2701 drbd_msg_put_info(from_attrs_err_to_txt(err));
2704 /* w_make_ov_request expects position to be aligned */
2705 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2707 /* If there is still bitmap IO pending, e.g. previous resync or verify
2708 * just being finished, wait for it before requesting a new resync. */
2709 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2710 retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2712 drbd_adm_finish(info, retcode);
2717 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2719 struct drbd_conf *mdev;
2720 enum drbd_ret_code retcode;
2721 int skip_initial_sync = 0;
2723 struct new_c_uuid_parms args;
2725 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2726 if (!adm_ctx.reply_skb)
2728 if (retcode != NO_ERROR)
2731 mdev = adm_ctx.mdev;
2732 memset(&args, 0, sizeof(args));
2733 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2734 err = new_c_uuid_parms_from_attrs(&args, info);
2736 retcode = ERR_MANDATORY_TAG;
2737 drbd_msg_put_info(from_attrs_err_to_txt(err));
2742 mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2744 if (!get_ldev(mdev)) {
2745 retcode = ERR_NO_DISK;
2749 /* this is "skip initial sync", assume to be clean */
2750 if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2751 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2752 dev_info(DEV, "Preparing to skip initial sync\n");
2753 skip_initial_sync = 1;
2754 } else if (mdev->state.conn != C_STANDALONE) {
2755 retcode = ERR_CONNECTED;
2759 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2760 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2762 if (args.clear_bm) {
2763 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2764 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2766 dev_err(DEV, "Writing bitmap failed with %d\n",err);
2767 retcode = ERR_IO_MD_DISK;
2769 if (skip_initial_sync) {
2770 drbd_send_uuids_skip_initial_sync(mdev);
2771 _drbd_uuid_set(mdev, UI_BITMAP, 0);
2772 drbd_print_uuids(mdev, "cleared bitmap UUID");
2773 spin_lock_irq(&mdev->tconn->req_lock);
2774 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2776 spin_unlock_irq(&mdev->tconn->req_lock);
2784 mutex_unlock(mdev->state_mutex);
2786 drbd_adm_finish(info, retcode);
2790 static enum drbd_ret_code
2791 drbd_check_conn_name(const char *name)
2793 if (!name || !name[0]) {
2794 drbd_msg_put_info("connection name missing");
2795 return ERR_MANDATORY_TAG;
2797 /* if we want to use these in sysfs/configfs/debugfs some day,
2798 * we must not allow slashes */
2799 if (strchr(name, '/')) {
2800 drbd_msg_put_info("invalid connection name");
2801 return ERR_INVALID_REQUEST;
2806 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info)
2808 enum drbd_ret_code retcode;
2810 retcode = drbd_adm_prepare(skb, info, 0);
2811 if (!adm_ctx.reply_skb)
2813 if (retcode != NO_ERROR)
2816 retcode = drbd_check_conn_name(adm_ctx.conn_name);
2817 if (retcode != NO_ERROR)
2820 if (adm_ctx.tconn) {
2821 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
2822 retcode = ERR_INVALID_REQUEST;
2823 drbd_msg_put_info("connection exists");
2825 /* else: still NO_ERROR */
2829 if (!drbd_new_tconn(adm_ctx.conn_name))
2830 retcode = ERR_NOMEM;
2832 drbd_adm_finish(info, retcode);
2836 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
2838 struct drbd_genlmsghdr *dh = info->userhdr;
2839 enum drbd_ret_code retcode;
2841 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2842 if (!adm_ctx.reply_skb)
2844 if (retcode != NO_ERROR)
2847 /* FIXME drop minor_count parameter, limit to MINORMASK */
2848 if (dh->minor >= minor_count) {
2849 drbd_msg_put_info("requested minor out of range");
2850 retcode = ERR_INVALID_REQUEST;
2853 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
2854 drbd_msg_put_info("requested volume id out of range");
2855 retcode = ERR_INVALID_REQUEST;
2859 /* drbd_adm_prepare made sure already
2860 * that mdev->tconn and mdev->vnr match the request. */
2862 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
2863 retcode = ERR_MINOR_EXISTS;
2864 /* else: still NO_ERROR */
2868 retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
2870 drbd_adm_finish(info, retcode);
2874 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
2876 if (mdev->state.disk == D_DISKLESS &&
2877 /* no need to be mdev->state.conn == C_STANDALONE &&
2878 * we may want to delete a minor from a live replication group.
2880 mdev->state.role == R_SECONDARY) {
2881 drbd_delete_device(mdev);
2884 return ERR_MINOR_CONFIGURED;
2887 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
2889 enum drbd_ret_code retcode;
2891 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2892 if (!adm_ctx.reply_skb)
2894 if (retcode != NO_ERROR)
2897 mutex_lock(&drbd_cfg_mutex);
2898 retcode = adm_delete_minor(adm_ctx.mdev);
2899 mutex_unlock(&drbd_cfg_mutex);
2900 /* if this was the last volume of this connection,
2901 * this will terminate all threads */
2902 if (retcode == NO_ERROR)
2903 conn_reconfig_done(adm_ctx.tconn);
2905 drbd_adm_finish(info, retcode);
2909 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
2911 enum drbd_ret_code retcode;
2912 enum drbd_state_rv rv;
2913 struct drbd_conf *mdev;
2916 retcode = drbd_adm_prepare(skb, info, 0);
2917 if (!adm_ctx.reply_skb)
2919 if (retcode != NO_ERROR)
2922 if (!adm_ctx.tconn) {
2923 retcode = ERR_CONN_NOT_KNOWN;
2927 mutex_lock(&drbd_cfg_mutex);
2929 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
2930 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
2931 if (retcode < SS_SUCCESS) {
2932 drbd_msg_put_info("failed to demote");
2938 rv = conn_try_disconnect(adm_ctx.tconn, 0);
2939 if (rv < SS_SUCCESS) {
2940 retcode = rv; /* enum type mismatch! */
2941 drbd_msg_put_info("failed to disconnect");
2946 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
2947 rv = adm_detach(mdev);
2948 if (rv < SS_SUCCESS) {
2949 retcode = rv; /* enum type mismatch! */
2950 drbd_msg_put_info("failed to detach");
2955 /* delete volumes */
2956 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
2957 retcode = adm_delete_minor(mdev);
2958 if (retcode != NO_ERROR) {
2959 /* "can not happen" */
2960 drbd_msg_put_info("failed to delete volume");
2965 /* stop all threads */
2966 conn_reconfig_done(adm_ctx.tconn);
2968 /* delete connection */
2969 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
2970 drbd_free_tconn(adm_ctx.tconn);
2973 /* "can not happen" */
2974 retcode = ERR_CONN_IN_USE;
2975 drbd_msg_put_info("failed to delete connection");
2979 mutex_unlock(&drbd_cfg_mutex);
2981 drbd_adm_finish(info, retcode);
2985 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info)
2987 enum drbd_ret_code retcode;
2989 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2990 if (!adm_ctx.reply_skb)
2992 if (retcode != NO_ERROR)
2995 mutex_lock(&drbd_cfg_mutex);
2996 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
2997 drbd_free_tconn(adm_ctx.tconn);
3000 retcode = ERR_CONN_IN_USE;
3002 mutex_unlock(&drbd_cfg_mutex);
3005 drbd_adm_finish(info, retcode);
3009 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3011 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3012 struct sk_buff *msg;
3013 struct drbd_genlmsghdr *d_out;
3017 seq = atomic_inc_return(&drbd_genl_seq);
3018 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3023 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3024 if (!d_out) /* cannot happen, but anyways. */
3025 goto nla_put_failure;
3026 d_out->minor = mdev_to_minor(mdev);
3027 d_out->ret_code = 0;
3029 if (nla_put_status_info(msg, mdev, sib))
3030 goto nla_put_failure;
3031 genlmsg_end(msg, d_out);
3032 err = drbd_genl_multicast_events(msg, 0);
3033 /* msg has been consumed or freed in netlink_broadcast() */
3034 if (err && err != -ESRCH)
3042 dev_err(DEV, "Error %d while broadcasting event. "
3043 "Event seq:%u sib_reason:%u\n",
3044 err, seq, sib->sib_reason);