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_get_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;
230 nlmsg_free(adm_ctx.reply_skb);
231 adm_ctx.reply_skb = NULL;
235 static int drbd_adm_finish(struct genl_info *info, int retcode)
238 const char *conn_name = NULL;
241 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
242 adm_ctx.tconn = NULL;
245 if (!adm_ctx.reply_skb)
248 adm_ctx.reply_dh->ret_code = retcode;
250 nla = info->attrs[DRBD_NLA_CFG_CONTEXT];
252 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
254 conn_name = nla_data(nla);
257 drbd_adm_send_reply(adm_ctx.reply_skb, info);
261 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
267 nc = rcu_dereference(tconn->net_conf);
269 switch (((struct sockaddr *)nc->peer_addr)->sa_family) {
272 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
273 &((struct sockaddr_in6 *)nc->peer_addr)->sin6_addr);
277 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
278 &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
282 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
283 &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
285 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
290 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
292 char *envp[] = { "HOME=/",
294 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
295 (char[20]) { }, /* address family */
296 (char[60]) { }, /* address */
299 char *argv[] = {usermode_helper, cmd, mb, NULL };
303 snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
304 setup_khelper_env(mdev->tconn, envp);
306 /* The helper may take some time.
307 * write out any unsynced meta data changes now */
310 dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
311 sib.sib_reason = SIB_HELPER_PRE;
312 sib.helper_name = cmd;
313 drbd_bcast_event(mdev, &sib);
314 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
316 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
317 usermode_helper, cmd, mb,
318 (ret >> 8) & 0xff, ret);
320 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
321 usermode_helper, cmd, mb,
322 (ret >> 8) & 0xff, ret);
323 sib.sib_reason = SIB_HELPER_POST;
324 sib.helper_exit_code = ret;
325 drbd_bcast_event(mdev, &sib);
327 if (ret < 0) /* Ignore any ERRNOs we got. */
333 static void conn_md_sync(struct drbd_tconn *tconn)
335 struct drbd_conf *mdev;
338 down_read(&drbd_cfg_rwsem);
339 idr_for_each_entry(&tconn->volumes, mdev, vnr)
341 up_read(&drbd_cfg_rwsem);
344 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
346 char *envp[] = { "HOME=/",
348 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
349 (char[20]) { }, /* address family */
350 (char[60]) { }, /* address */
352 char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
355 setup_khelper_env(tconn, envp);
358 conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
359 /* TODO: conn_bcast_event() ?? */
361 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
363 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
364 usermode_helper, cmd, tconn->name,
365 (ret >> 8) & 0xff, ret);
367 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
368 usermode_helper, cmd, tconn->name,
369 (ret >> 8) & 0xff, ret);
370 /* TODO: conn_bcast_event() ?? */
372 if (ret < 0) /* Ignore any ERRNOs we got. */
378 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
380 enum drbd_fencing_p fp = FP_NOT_AVAIL;
381 struct drbd_conf *mdev;
385 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
386 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
387 fp = max_t(enum drbd_fencing_p, fp, mdev->ldev->dc.fencing);
396 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
398 union drbd_state mask = { };
399 union drbd_state val = { };
400 enum drbd_fencing_p fp;
404 if (tconn->cstate >= C_WF_REPORT_PARAMS) {
405 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
409 fp = highest_fencing_policy(tconn);
412 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
419 r = conn_khelper(tconn, "fence-peer");
421 switch ((r>>8) & 0xff) {
422 case 3: /* peer is inconsistent */
423 ex_to_string = "peer is inconsistent or worse";
425 val.pdsk = D_INCONSISTENT;
427 case 4: /* peer got outdated, or was already outdated */
428 ex_to_string = "peer was fenced";
430 val.pdsk = D_OUTDATED;
432 case 5: /* peer was down */
433 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
434 /* we will(have) create(d) a new UUID anyways... */
435 ex_to_string = "peer is unreachable, assumed to be dead";
437 val.pdsk = D_OUTDATED;
439 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
442 case 6: /* Peer is primary, voluntarily outdate myself.
443 * This is useful when an unconnected R_SECONDARY is asked to
444 * become R_PRIMARY, but finds the other peer being active. */
445 ex_to_string = "peer is active";
446 conn_warn(tconn, "Peer is primary, outdating myself.\n");
448 val.disk = D_OUTDATED;
451 if (fp != FP_STONITH)
452 conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
453 ex_to_string = "peer was stonithed";
455 val.pdsk = D_OUTDATED;
458 /* The script is broken ... */
459 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
460 return false; /* Eventually leave IO frozen */
463 conn_info(tconn, "fence-peer helper returned %d (%s)\n",
464 (r>>8) & 0xff, ex_to_string);
469 conn_request_state(tconn, mask, val, CS_VERBOSE);
470 here, because we might were able to re-establish the connection in the
472 spin_lock_irq(&tconn->req_lock);
473 if (tconn->cstate < C_WF_REPORT_PARAMS)
474 _conn_request_state(tconn, mask, val, CS_VERBOSE);
475 spin_unlock_irq(&tconn->req_lock);
477 return conn_highest_pdsk(tconn) <= D_OUTDATED;
480 static int _try_outdate_peer_async(void *data)
482 struct drbd_tconn *tconn = (struct drbd_tconn *)data;
484 conn_try_outdate_peer(tconn);
486 kref_put(&tconn->kref, &conn_destroy);
490 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
492 struct task_struct *opa;
494 kref_get(&tconn->kref);
495 opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
497 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
498 kref_put(&tconn->kref, &conn_destroy);
503 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
505 const int max_tries = 4;
506 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
510 union drbd_state mask, val;
512 if (new_role == R_PRIMARY)
513 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
515 mutex_lock(mdev->state_mutex);
517 mask.i = 0; mask.role = R_MASK;
518 val.i = 0; val.role = new_role;
520 while (try++ < max_tries) {
521 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
523 /* in case we first succeeded to outdate,
524 * but now suddenly could establish a connection */
525 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
531 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
532 (mdev->state.disk < D_UP_TO_DATE &&
533 mdev->state.disk >= D_INCONSISTENT)) {
535 val.disk = D_UP_TO_DATE;
540 if (rv == SS_NO_UP_TO_DATE_DISK &&
541 mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
542 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
544 if (conn_try_outdate_peer(mdev->tconn)) {
545 val.disk = D_UP_TO_DATE;
551 if (rv == SS_NOTHING_TO_DO)
553 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
554 if (!conn_try_outdate_peer(mdev->tconn) && force) {
555 dev_warn(DEV, "Forced into split brain situation!\n");
557 val.pdsk = D_OUTDATED;
562 if (rv == SS_TWO_PRIMARIES) {
563 /* Maybe the peer is detected as dead very soon...
564 retry at most once more in this case. */
567 nc = rcu_dereference(mdev->tconn->net_conf);
568 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
570 schedule_timeout_interruptible(timeo);
575 if (rv < SS_SUCCESS) {
576 rv = _drbd_request_state(mdev, mask, val,
577 CS_VERBOSE + CS_WAIT_COMPLETE);
588 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
590 /* Wait until nothing is on the fly :) */
591 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
593 if (new_role == R_SECONDARY) {
594 set_disk_ro(mdev->vdisk, true);
595 if (get_ldev(mdev)) {
596 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
600 mutex_lock(&mdev->tconn->net_conf_update);
601 nc = mdev->tconn->net_conf;
603 nc->want_lose = 0; /* without copy; single bit op is atomic */
604 mutex_unlock(&mdev->tconn->net_conf_update);
606 set_disk_ro(mdev->vdisk, false);
607 if (get_ldev(mdev)) {
608 if (((mdev->state.conn < C_CONNECTED ||
609 mdev->state.pdsk <= D_FAILED)
610 && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
611 drbd_uuid_new_current(mdev);
613 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
618 /* writeout of activity log covered areas of the bitmap
619 * to stable storage done in after state change already */
621 if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
622 /* if this was forced, we should consider sync */
624 drbd_send_uuids(mdev);
625 drbd_send_state(mdev);
630 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
632 mutex_unlock(mdev->state_mutex);
636 static const char *from_attrs_err_to_txt(int err)
638 return err == -ENOMSG ? "required attribute missing" :
639 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
640 err == -EEXIST ? "can not change invariant setting" :
641 "invalid attribute value";
644 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
646 struct set_role_parms parms;
648 enum drbd_ret_code retcode;
650 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
651 if (!adm_ctx.reply_skb)
653 if (retcode != NO_ERROR)
656 memset(&parms, 0, sizeof(parms));
657 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
658 err = set_role_parms_from_attrs(&parms, info);
660 retcode = ERR_MANDATORY_TAG;
661 drbd_msg_put_info(from_attrs_err_to_txt(err));
666 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
667 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
669 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
671 drbd_adm_finish(info, retcode);
675 /* initializes the md.*_offset members, so we are able to find
676 * the on disk meta data */
677 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
678 struct drbd_backing_dev *bdev)
680 sector_t md_size_sect = 0;
681 switch (bdev->dc.meta_dev_idx) {
683 /* v07 style fixed size indexed meta data */
684 bdev->md.md_size_sect = MD_RESERVED_SECT;
685 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
686 bdev->md.al_offset = MD_AL_OFFSET;
687 bdev->md.bm_offset = MD_BM_OFFSET;
689 case DRBD_MD_INDEX_FLEX_EXT:
690 /* just occupy the full device; unit: sectors */
691 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
692 bdev->md.md_offset = 0;
693 bdev->md.al_offset = MD_AL_OFFSET;
694 bdev->md.bm_offset = MD_BM_OFFSET;
696 case DRBD_MD_INDEX_INTERNAL:
697 case DRBD_MD_INDEX_FLEX_INT:
698 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
699 /* al size is still fixed */
700 bdev->md.al_offset = -MD_AL_SECTORS;
701 /* we need (slightly less than) ~ this much bitmap sectors: */
702 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
703 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
704 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
705 md_size_sect = ALIGN(md_size_sect, 8);
707 /* plus the "drbd meta data super block",
708 * and the activity log; */
709 md_size_sect += MD_BM_OFFSET;
711 bdev->md.md_size_sect = md_size_sect;
712 /* bitmap offset is adjusted by 'super' block size */
713 bdev->md.bm_offset = -md_size_sect + MD_AL_OFFSET;
718 /* input size is expected to be in KB */
719 char *ppsize(char *buf, unsigned long long size)
721 /* Needs 9 bytes at max including trailing NUL:
722 * -1ULL ==> "16384 EB" */
723 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
725 while (size >= 10000 && base < sizeof(units)-1) {
727 size = (size >> 10) + !!(size & (1<<9));
730 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
735 /* there is still a theoretical deadlock when called from receiver
736 * on an D_INCONSISTENT R_PRIMARY:
737 * remote READ does inc_ap_bio, receiver would need to receive answer
738 * packet from remote to dec_ap_bio again.
739 * receiver receive_sizes(), comes here,
740 * waits for ap_bio_cnt == 0. -> deadlock.
741 * but this cannot happen, actually, because:
742 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
743 * (not connected, or bad/no disk on peer):
744 * see drbd_fail_request_early, ap_bio_cnt is zero.
745 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
746 * peer may not initiate a resize.
748 /* Note these are not to be confused with
749 * drbd_adm_suspend_io/drbd_adm_resume_io,
750 * which are (sub) state changes triggered by admin (drbdsetup),
751 * and can be long lived.
752 * This changes an mdev->flag, is triggered by drbd internals,
753 * and should be short-lived. */
754 void drbd_suspend_io(struct drbd_conf *mdev)
756 set_bit(SUSPEND_IO, &mdev->flags);
757 if (drbd_suspended(mdev))
759 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
762 void drbd_resume_io(struct drbd_conf *mdev)
764 clear_bit(SUSPEND_IO, &mdev->flags);
765 wake_up(&mdev->misc_wait);
769 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
770 * @mdev: DRBD device.
772 * Returns 0 on success, negative return values indicate errors.
773 * You should call drbd_md_sync() after calling this function.
775 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
777 sector_t prev_first_sect, prev_size; /* previous meta location */
782 int md_moved, la_size_changed;
783 enum determine_dev_size rv = unchanged;
786 * application request passes inc_ap_bio,
787 * but then cannot get an AL-reference.
788 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
791 * Suspend IO right here.
792 * still lock the act_log to not trigger ASSERTs there.
794 drbd_suspend_io(mdev);
796 /* no wait necessary anymore, actually we could assert that */
797 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
799 prev_first_sect = drbd_md_first_sector(mdev->ldev);
800 prev_size = mdev->ldev->md.md_size_sect;
801 la_size = mdev->ldev->md.la_size_sect;
803 /* TODO: should only be some assert here, not (re)init... */
804 drbd_md_set_sector_offsets(mdev, mdev->ldev);
806 size = drbd_new_dev_size(mdev, mdev->ldev, flags & DDSF_FORCED);
808 if (drbd_get_capacity(mdev->this_bdev) != size ||
809 drbd_bm_capacity(mdev) != size) {
811 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
813 /* currently there is only one error: ENOMEM! */
814 size = drbd_bm_capacity(mdev)>>1;
816 dev_err(DEV, "OUT OF MEMORY! "
817 "Could not allocate bitmap!\n");
819 dev_err(DEV, "BM resizing failed. "
820 "Leaving size unchanged at size = %lu KB\n",
821 (unsigned long)size);
825 /* racy, see comments above. */
826 drbd_set_my_capacity(mdev, size);
827 mdev->ldev->md.la_size_sect = size;
828 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
829 (unsigned long long)size>>1);
831 if (rv == dev_size_error)
834 la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
836 md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
837 || prev_size != mdev->ldev->md.md_size_sect;
839 if (la_size_changed || md_moved) {
842 drbd_al_shrink(mdev); /* All extents inactive. */
843 dev_info(DEV, "Writing the whole bitmap, %s\n",
844 la_size_changed && md_moved ? "size changed and md moved" :
845 la_size_changed ? "size changed" : "md moved");
846 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
847 err = drbd_bitmap_io(mdev, &drbd_bm_write,
848 "size changed", BM_LOCKED_MASK);
853 drbd_md_mark_dirty(mdev);
861 lc_unlock(mdev->act_log);
862 wake_up(&mdev->al_wait);
863 drbd_resume_io(mdev);
869 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev, int assume_peer_has_space)
871 sector_t p_size = mdev->p_size; /* partner's disk size. */
872 sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
873 sector_t m_size; /* my size */
874 sector_t u_size = bdev->dc.disk_size; /* size requested by user. */
877 m_size = drbd_get_max_capacity(bdev);
879 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
880 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
884 if (p_size && m_size) {
885 size = min_t(sector_t, p_size, m_size);
889 if (m_size && m_size < size)
891 if (p_size && p_size < size)
902 dev_err(DEV, "Both nodes diskless!\n");
906 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
907 (unsigned long)u_size>>1, (unsigned long)size>>1);
916 * drbd_check_al_size() - Ensures that the AL is of the right size
917 * @mdev: DRBD device.
919 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
920 * failed, and 0 on success. You should call drbd_md_sync() after you called
923 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
925 struct lru_cache *n, *t;
926 struct lc_element *e;
930 if (!expect(dc->al_extents >= DRBD_AL_EXTENTS_MIN))
931 dc->al_extents = DRBD_AL_EXTENTS_MIN;
934 mdev->act_log->nr_elements == dc->al_extents)
939 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
940 dc->al_extents, sizeof(struct lc_element), 0);
943 dev_err(DEV, "Cannot allocate act_log lru!\n");
946 spin_lock_irq(&mdev->al_lock);
948 for (i = 0; i < t->nr_elements; i++) {
949 e = lc_element_by_index(t, i);
951 dev_err(DEV, "refcnt(%d)==%d\n",
952 e->lc_number, e->refcnt);
958 spin_unlock_irq(&mdev->al_lock);
960 dev_err(DEV, "Activity log still in use!\n");
967 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
971 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
973 struct request_queue * const q = mdev->rq_queue;
974 int max_hw_sectors = max_bio_size >> 9;
975 int max_segments = 0;
977 if (get_ldev_if_state(mdev, D_ATTACHING)) {
978 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
980 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
981 max_segments = mdev->ldev->dc.max_bio_bvecs;
985 blk_queue_logical_block_size(q, 512);
986 blk_queue_max_hw_sectors(q, max_hw_sectors);
987 /* This is the workaround for "bio would need to, but cannot, be split" */
988 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
989 blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
991 if (get_ldev_if_state(mdev, D_ATTACHING)) {
992 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
994 blk_queue_stack_limits(q, b);
996 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
997 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
998 q->backing_dev_info.ra_pages,
999 b->backing_dev_info.ra_pages);
1000 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1006 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1008 int now, new, local, peer;
1010 now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1011 local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1012 peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1014 if (get_ldev_if_state(mdev, D_ATTACHING)) {
1015 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1016 mdev->local_max_bio_size = local;
1020 /* We may ignore peer limits if the peer is modern enough.
1021 Because new from 8.3.8 onwards the peer can use multiple
1022 BIOs for a single peer_request */
1023 if (mdev->state.conn >= C_CONNECTED) {
1024 if (mdev->tconn->agreed_pro_version < 94)
1025 peer = mdev->peer_max_bio_size;
1026 else if (mdev->tconn->agreed_pro_version == 94)
1027 peer = DRBD_MAX_SIZE_H80_PACKET;
1028 else /* drbd 8.3.8 onwards */
1029 peer = DRBD_MAX_BIO_SIZE;
1032 new = min_t(int, local, peer);
1034 if (mdev->state.role == R_PRIMARY && new < now)
1035 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
1038 dev_info(DEV, "max BIO size = %u\n", new);
1040 drbd_setup_queue_param(mdev, new);
1043 /* Starts the worker thread */
1044 static void conn_reconfig_start(struct drbd_tconn *tconn)
1046 drbd_thread_start(&tconn->worker);
1047 conn_flush_workqueue(tconn);
1050 /* if still unconfigured, stops worker again. */
1051 static void conn_reconfig_done(struct drbd_tconn *tconn)
1053 spin_lock_irq(&tconn->req_lock);
1054 if (conn_all_vols_unconf(tconn))
1055 drbd_thread_stop_nowait(&tconn->worker);
1056 spin_unlock_irq(&tconn->req_lock);
1059 /* Make sure IO is suspended before calling this function(). */
1060 static void drbd_suspend_al(struct drbd_conf *mdev)
1064 if (!lc_try_lock(mdev->act_log)) {
1065 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1069 drbd_al_shrink(mdev);
1070 spin_lock_irq(&mdev->tconn->req_lock);
1071 if (mdev->state.conn < C_CONNECTED)
1072 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1073 spin_unlock_irq(&mdev->tconn->req_lock);
1074 lc_unlock(mdev->act_log);
1077 dev_info(DEV, "Suspended AL updates\n");
1080 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1082 enum drbd_ret_code retcode;
1083 struct drbd_conf *mdev;
1084 struct disk_conf *ndc; /* new disk conf */
1086 int *rs_plan_s = NULL;
1088 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1089 if (!adm_ctx.reply_skb)
1091 if (retcode != NO_ERROR)
1094 mdev = adm_ctx.mdev;
1096 /* we also need a disk
1097 * to change the options on */
1098 if (!get_ldev(mdev)) {
1099 retcode = ERR_NO_DISK;
1103 /* FIXME freeze IO, cluster wide.
1105 * We should make sure no-one uses
1106 * some half-updated struct when we
1107 * assign it later. */
1109 ndc = kmalloc(sizeof(*ndc), GFP_KERNEL);
1111 retcode = ERR_NOMEM;
1115 memcpy(ndc, &mdev->ldev->dc, sizeof(*ndc));
1116 err = disk_conf_from_attrs_for_change(ndc, info);
1118 retcode = ERR_MANDATORY_TAG;
1119 drbd_msg_put_info(from_attrs_err_to_txt(err));
1122 if (!expect(ndc->resync_rate >= 1))
1123 ndc->resync_rate = 1;
1125 /* clip to allowed range */
1126 if (!expect(ndc->al_extents >= DRBD_AL_EXTENTS_MIN))
1127 ndc->al_extents = DRBD_AL_EXTENTS_MIN;
1128 if (!expect(ndc->al_extents <= DRBD_AL_EXTENTS_MAX))
1129 ndc->al_extents = DRBD_AL_EXTENTS_MAX;
1131 /* most sanity checks done, try to assign the new sync-after
1132 * dependency. need to hold the global lock in there,
1133 * to avoid a race in the dependency loop check. */
1134 retcode = drbd_alter_sa(mdev, ndc->resync_after);
1135 if (retcode != NO_ERROR)
1138 fifo_size = (ndc->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1139 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
1140 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
1142 dev_err(DEV, "kmalloc of fifo_buffer failed");
1143 retcode = ERR_NOMEM;
1148 if (fifo_size != mdev->rs_plan_s.size) {
1149 kfree(mdev->rs_plan_s.values);
1150 mdev->rs_plan_s.values = rs_plan_s;
1151 mdev->rs_plan_s.size = fifo_size;
1152 mdev->rs_planed = 0;
1156 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1157 drbd_al_shrink(mdev);
1158 err = drbd_check_al_size(mdev, ndc);
1159 lc_unlock(mdev->act_log);
1160 wake_up(&mdev->al_wait);
1163 retcode = ERR_NOMEM;
1168 * To avoid someone looking at a half-updated struct, we probably
1169 * should have a rw-semaphor on net_conf and disk_conf.
1171 mdev->ldev->dc = *ndc;
1176 if (mdev->state.conn >= C_CONNECTED)
1177 drbd_send_sync_param(mdev);
1184 drbd_adm_finish(info, retcode);
1188 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1190 struct drbd_conf *mdev;
1192 enum drbd_ret_code retcode;
1193 enum determine_dev_size dd;
1194 sector_t max_possible_sectors;
1195 sector_t min_md_device_sectors;
1196 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1197 struct block_device *bdev;
1198 struct lru_cache *resync_lru = NULL;
1199 union drbd_state ns, os;
1200 enum drbd_state_rv rv;
1201 struct net_conf *nc;
1202 int cp_discovered = 0;
1204 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1205 if (!adm_ctx.reply_skb)
1207 if (retcode != NO_ERROR)
1210 mdev = adm_ctx.mdev;
1211 conn_reconfig_start(mdev->tconn);
1213 /* if you want to reconfigure, please tear down first */
1214 if (mdev->state.disk > D_DISKLESS) {
1215 retcode = ERR_DISK_CONFIGURED;
1218 /* It may just now have detached because of IO error. Make sure
1219 * drbd_ldev_destroy is done already, we may end up here very fast,
1220 * e.g. if someone calls attach from the on-io-error handler,
1221 * to realize a "hot spare" feature (not that I'd recommend that) */
1222 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1224 /* allocation not in the IO path, drbdsetup context */
1225 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1227 retcode = ERR_NOMEM;
1231 nbc->dc = (struct disk_conf) {
1232 {}, 0, /* backing_dev */
1233 {}, 0, /* meta_dev */
1234 0, /* meta_dev_idx */
1235 DRBD_DISK_SIZE_SECT_DEF, /* disk_size */
1236 DRBD_MAX_BIO_BVECS_DEF, /* max_bio_bvecs */
1237 DRBD_ON_IO_ERROR_DEF, /* on_io_error */
1238 DRBD_FENCING_DEF, /* fencing */
1239 DRBD_RATE_DEF, /* resync_rate */
1240 DRBD_AFTER_DEF, /* resync_after */
1241 DRBD_AL_EXTENTS_DEF, /* al_extents */
1242 DRBD_C_PLAN_AHEAD_DEF, /* c_plan_ahead */
1243 DRBD_C_DELAY_TARGET_DEF, /* c_delay_target */
1244 DRBD_C_FILL_TARGET_DEF, /* c_fill_target */
1245 DRBD_C_MAX_RATE_DEF, /* c_max_rate */
1246 DRBD_C_MIN_RATE_DEF, /* c_min_rate */
1247 0, /* no_disk_barrier */
1248 0, /* no_disk_flush */
1249 0, /* no_disk_drain */
1250 0, /* no_md_flush */
1253 err = disk_conf_from_attrs(&nbc->dc, info);
1255 retcode = ERR_MANDATORY_TAG;
1256 drbd_msg_put_info(from_attrs_err_to_txt(err));
1260 if ((int)nbc->dc.meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1261 retcode = ERR_MD_IDX_INVALID;
1266 nc = rcu_dereference(mdev->tconn->net_conf);
1268 if (nbc->dc.fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1270 retcode = ERR_STONITH_AND_PROT_A;
1276 bdev = blkdev_get_by_path(nbc->dc.backing_dev,
1277 FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1279 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.backing_dev,
1281 retcode = ERR_OPEN_DISK;
1284 nbc->backing_bdev = bdev;
1287 * meta_dev_idx >= 0: external fixed size, possibly multiple
1288 * drbd sharing one meta device. TODO in that case, paranoia
1289 * check that [md_bdev, meta_dev_idx] is not yet used by some
1290 * other drbd minor! (if you use drbd.conf + drbdadm, that
1291 * should check it for you already; but if you don't, or
1292 * someone fooled it, we need to double check here)
1294 bdev = blkdev_get_by_path(nbc->dc.meta_dev,
1295 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1296 ((int)nbc->dc.meta_dev_idx < 0) ?
1297 (void *)mdev : (void *)drbd_m_holder);
1299 dev_err(DEV, "open(\"%s\") failed with %ld\n", nbc->dc.meta_dev,
1301 retcode = ERR_OPEN_MD_DISK;
1304 nbc->md_bdev = bdev;
1306 if ((nbc->backing_bdev == nbc->md_bdev) !=
1307 (nbc->dc.meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1308 nbc->dc.meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1309 retcode = ERR_MD_IDX_INVALID;
1313 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1314 1, 61, sizeof(struct bm_extent),
1315 offsetof(struct bm_extent, lce));
1317 retcode = ERR_NOMEM;
1321 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1322 drbd_md_set_sector_offsets(mdev, nbc);
1324 if (drbd_get_max_capacity(nbc) < nbc->dc.disk_size) {
1325 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1326 (unsigned long long) drbd_get_max_capacity(nbc),
1327 (unsigned long long) nbc->dc.disk_size);
1328 retcode = ERR_DISK_TO_SMALL;
1332 if ((int)nbc->dc.meta_dev_idx < 0) {
1333 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1334 /* at least one MB, otherwise it does not make sense */
1335 min_md_device_sectors = (2<<10);
1337 max_possible_sectors = DRBD_MAX_SECTORS;
1338 min_md_device_sectors = MD_RESERVED_SECT * (nbc->dc.meta_dev_idx + 1);
1341 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1342 retcode = ERR_MD_DISK_TO_SMALL;
1343 dev_warn(DEV, "refusing attach: md-device too small, "
1344 "at least %llu sectors needed for this meta-disk type\n",
1345 (unsigned long long) min_md_device_sectors);
1349 /* Make sure the new disk is big enough
1350 * (we may currently be R_PRIMARY with no local disk...) */
1351 if (drbd_get_max_capacity(nbc) <
1352 drbd_get_capacity(mdev->this_bdev)) {
1353 retcode = ERR_DISK_TO_SMALL;
1357 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1359 if (nbc->known_size > max_possible_sectors) {
1360 dev_warn(DEV, "==> truncating very big lower level device "
1361 "to currently maximum possible %llu sectors <==\n",
1362 (unsigned long long) max_possible_sectors);
1363 if ((int)nbc->dc.meta_dev_idx >= 0)
1364 dev_warn(DEV, "==>> using internal or flexible "
1365 "meta data may help <<==\n");
1368 drbd_suspend_io(mdev);
1369 /* also wait for the last barrier ack. */
1370 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1371 /* and for any other previously queued work */
1372 drbd_flush_workqueue(mdev);
1374 rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1375 retcode = rv; /* FIXME: Type mismatch. */
1376 drbd_resume_io(mdev);
1377 if (rv < SS_SUCCESS)
1380 if (!get_ldev_if_state(mdev, D_ATTACHING))
1381 goto force_diskless;
1383 drbd_md_set_sector_offsets(mdev, nbc);
1385 if (!mdev->bitmap) {
1386 if (drbd_bm_init(mdev)) {
1387 retcode = ERR_NOMEM;
1388 goto force_diskless_dec;
1392 retcode = drbd_md_read(mdev, nbc);
1393 if (retcode != NO_ERROR)
1394 goto force_diskless_dec;
1396 if (mdev->state.conn < C_CONNECTED &&
1397 mdev->state.role == R_PRIMARY &&
1398 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1399 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1400 (unsigned long long)mdev->ed_uuid);
1401 retcode = ERR_DATA_NOT_CURRENT;
1402 goto force_diskless_dec;
1405 /* Since we are diskless, fix the activity log first... */
1406 if (drbd_check_al_size(mdev, &nbc->dc)) {
1407 retcode = ERR_NOMEM;
1408 goto force_diskless_dec;
1411 /* Prevent shrinking of consistent devices ! */
1412 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1413 drbd_new_dev_size(mdev, nbc, 0) < nbc->md.la_size_sect) {
1414 dev_warn(DEV, "refusing to truncate a consistent device\n");
1415 retcode = ERR_DISK_TO_SMALL;
1416 goto force_diskless_dec;
1419 if (!drbd_al_read_log(mdev, nbc)) {
1420 retcode = ERR_IO_MD_DISK;
1421 goto force_diskless_dec;
1424 /* Reset the "barriers don't work" bits here, then force meta data to
1425 * be written, to ensure we determine if barriers are supported. */
1426 if (nbc->dc.no_md_flush)
1427 set_bit(MD_NO_FUA, &mdev->flags);
1429 clear_bit(MD_NO_FUA, &mdev->flags);
1431 /* Point of no return reached.
1432 * Devices and memory are no longer released by error cleanup below.
1433 * now mdev takes over responsibility, and the state engine should
1434 * clean it up somewhere. */
1435 D_ASSERT(mdev->ldev == NULL);
1437 mdev->resync = resync_lru;
1441 mdev->write_ordering = WO_bdev_flush;
1442 drbd_bump_write_ordering(mdev, WO_bdev_flush);
1444 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1445 set_bit(CRASHED_PRIMARY, &mdev->flags);
1447 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1449 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1450 !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod)) {
1451 set_bit(CRASHED_PRIMARY, &mdev->flags);
1460 drbd_reconsider_max_bio_size(mdev);
1462 /* If I am currently not R_PRIMARY,
1463 * but meta data primary indicator is set,
1464 * I just now recover from a hard crash,
1465 * and have been R_PRIMARY before that crash.
1467 * Now, if I had no connection before that crash
1468 * (have been degraded R_PRIMARY), chances are that
1469 * I won't find my peer now either.
1471 * In that case, and _only_ in that case,
1472 * we use the degr-wfc-timeout instead of the default,
1473 * so we can automatically recover from a crash of a
1474 * degraded but active "cluster" after a certain timeout.
1476 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1477 if (mdev->state.role != R_PRIMARY &&
1478 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1479 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1480 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1482 dd = drbd_determine_dev_size(mdev, 0);
1483 if (dd == dev_size_error) {
1484 retcode = ERR_NOMEM_BITMAP;
1485 goto force_diskless_dec;
1486 } else if (dd == grew)
1487 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1489 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1490 dev_info(DEV, "Assuming that all blocks are out of sync "
1491 "(aka FullSync)\n");
1492 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1493 "set_n_write from attaching", BM_LOCKED_MASK)) {
1494 retcode = ERR_IO_MD_DISK;
1495 goto force_diskless_dec;
1498 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1499 "read from attaching", BM_LOCKED_MASK)) {
1500 retcode = ERR_IO_MD_DISK;
1501 goto force_diskless_dec;
1505 if (cp_discovered) {
1506 drbd_al_apply_to_bm(mdev);
1507 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1508 "crashed primary apply AL", BM_LOCKED_MASK)) {
1509 retcode = ERR_IO_MD_DISK;
1510 goto force_diskless_dec;
1514 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1515 drbd_suspend_al(mdev); /* IO is still suspended here... */
1517 spin_lock_irq(&mdev->tconn->req_lock);
1518 os = drbd_read_state(mdev);
1520 /* If MDF_CONSISTENT is not set go into inconsistent state,
1521 otherwise investigate MDF_WasUpToDate...
1522 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1523 otherwise into D_CONSISTENT state.
1525 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1526 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1527 ns.disk = D_CONSISTENT;
1529 ns.disk = D_OUTDATED;
1531 ns.disk = D_INCONSISTENT;
1534 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1535 ns.pdsk = D_OUTDATED;
1537 if ( ns.disk == D_CONSISTENT &&
1538 (ns.pdsk == D_OUTDATED || mdev->ldev->dc.fencing == FP_DONT_CARE))
1539 ns.disk = D_UP_TO_DATE;
1541 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1542 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1543 this point, because drbd_request_state() modifies these
1546 /* In case we are C_CONNECTED postpone any decision on the new disk
1547 state after the negotiation phase. */
1548 if (mdev->state.conn == C_CONNECTED) {
1549 mdev->new_state_tmp.i = ns.i;
1551 ns.disk = D_NEGOTIATING;
1553 /* We expect to receive up-to-date UUIDs soon.
1554 To avoid a race in receive_state, free p_uuid while
1555 holding req_lock. I.e. atomic with the state change */
1556 kfree(mdev->p_uuid);
1557 mdev->p_uuid = NULL;
1560 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1561 spin_unlock_irq(&mdev->tconn->req_lock);
1563 if (rv < SS_SUCCESS)
1564 goto force_diskless_dec;
1566 if (mdev->state.role == R_PRIMARY)
1567 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1569 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1571 drbd_md_mark_dirty(mdev);
1574 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1576 conn_reconfig_done(mdev->tconn);
1577 drbd_adm_finish(info, retcode);
1583 drbd_force_state(mdev, NS(disk, D_FAILED));
1586 conn_reconfig_done(mdev->tconn);
1588 if (nbc->backing_bdev)
1589 blkdev_put(nbc->backing_bdev,
1590 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1592 blkdev_put(nbc->md_bdev,
1593 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1596 lc_destroy(resync_lru);
1599 drbd_adm_finish(info, retcode);
1603 static int adm_detach(struct drbd_conf *mdev)
1605 enum drbd_state_rv retcode;
1606 drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1607 retcode = drbd_request_state(mdev, NS(disk, D_DISKLESS));
1608 wait_event(mdev->misc_wait,
1609 mdev->state.disk != D_DISKLESS ||
1610 !atomic_read(&mdev->local_cnt));
1611 drbd_resume_io(mdev);
1615 /* Detaching the disk is a process in multiple stages. First we need to lock
1616 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1617 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1618 * internal references as well.
1619 * Only then we have finally detached. */
1620 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1622 enum drbd_ret_code retcode;
1624 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1625 if (!adm_ctx.reply_skb)
1627 if (retcode != NO_ERROR)
1630 retcode = adm_detach(adm_ctx.mdev);
1632 drbd_adm_finish(info, retcode);
1636 static bool conn_resync_running(struct drbd_tconn *tconn)
1638 struct drbd_conf *mdev;
1643 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1644 if (mdev->state.conn == C_SYNC_SOURCE ||
1645 mdev->state.conn == C_SYNC_TARGET ||
1646 mdev->state.conn == C_PAUSED_SYNC_S ||
1647 mdev->state.conn == C_PAUSED_SYNC_T) {
1657 static bool conn_ov_running(struct drbd_tconn *tconn)
1659 struct drbd_conf *mdev;
1664 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1665 if (mdev->state.conn == C_VERIFY_S ||
1666 mdev->state.conn == C_VERIFY_T) {
1676 static enum drbd_ret_code
1677 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1679 struct drbd_conf *mdev;
1682 if (old_conf && tconn->agreed_pro_version < 100 &&
1683 tconn->cstate == C_WF_REPORT_PARAMS &&
1684 new_conf->wire_protocol != old_conf->wire_protocol)
1685 return ERR_NEED_APV_100;
1687 if (new_conf->two_primaries &&
1688 (new_conf->wire_protocol != DRBD_PROT_C))
1689 return ERR_NOT_PROTO_C;
1691 idr_for_each_entry(&tconn->volumes, mdev, i) {
1692 if (get_ldev(mdev)) {
1693 enum drbd_fencing_p fp = mdev->ldev->dc.fencing;
1695 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1696 return ERR_STONITH_AND_PROT_A;
1698 if (mdev->state.role == R_PRIMARY && new_conf->want_lose)
1702 if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1703 return ERR_CONG_NOT_PROTO_A;
1708 static enum drbd_ret_code
1709 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1711 static enum drbd_ret_code rv;
1712 struct drbd_conf *mdev;
1716 rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1719 /* tconn->volumes protected by genl_lock() here */
1720 idr_for_each_entry(&tconn->volumes, mdev, i) {
1721 if (!mdev->bitmap) {
1722 if(drbd_bm_init(mdev))
1730 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1732 enum drbd_ret_code retcode;
1733 struct drbd_tconn *tconn;
1734 struct net_conf *old_conf, *new_conf = NULL;
1736 int ovr; /* online verify running */
1737 int rsr; /* re-sync running */
1738 struct crypto_hash *verify_tfm = NULL;
1739 struct crypto_hash *csums_tfm = NULL;
1742 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1743 if (!adm_ctx.reply_skb)
1745 if (retcode != NO_ERROR)
1748 tconn = adm_ctx.tconn;
1750 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1752 retcode = ERR_NOMEM;
1756 conn_reconfig_start(tconn);
1758 mutex_lock(&tconn->net_conf_update);
1759 old_conf = tconn->net_conf;
1762 drbd_msg_put_info("net conf missing, try connect");
1763 retcode = ERR_INVALID_REQUEST;
1767 *new_conf = *old_conf;
1769 err = net_conf_from_attrs_for_change(new_conf, info);
1771 retcode = ERR_MANDATORY_TAG;
1772 drbd_msg_put_info(from_attrs_err_to_txt(err));
1776 retcode = check_net_options(tconn, new_conf);
1777 if (retcode != NO_ERROR)
1780 /* re-sync running */
1781 rsr = conn_resync_running(tconn);
1782 if (rsr && old_conf && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1783 retcode = ERR_CSUMS_RESYNC_RUNNING;
1787 if (!rsr && new_conf->csums_alg[0]) {
1788 csums_tfm = crypto_alloc_hash(new_conf->csums_alg, 0, CRYPTO_ALG_ASYNC);
1789 if (IS_ERR(csums_tfm)) {
1791 retcode = ERR_CSUMS_ALG;
1795 if (!drbd_crypto_is_hash(crypto_hash_tfm(csums_tfm))) {
1796 retcode = ERR_CSUMS_ALG_ND;
1801 /* online verify running */
1802 ovr = conn_ov_running(tconn);
1804 if (strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1805 retcode = ERR_VERIFY_RUNNING;
1810 if (!ovr && new_conf->verify_alg[0]) {
1811 verify_tfm = crypto_alloc_hash(new_conf->verify_alg, 0, CRYPTO_ALG_ASYNC);
1812 if (IS_ERR(verify_tfm)) {
1814 retcode = ERR_VERIFY_ALG;
1818 if (!drbd_crypto_is_hash(crypto_hash_tfm(verify_tfm))) {
1819 retcode = ERR_VERIFY_ALG_ND;
1824 rcu_assign_pointer(tconn->net_conf, new_conf);
1827 crypto_free_hash(tconn->csums_tfm);
1828 tconn->csums_tfm = csums_tfm;
1832 crypto_free_hash(tconn->verify_tfm);
1833 tconn->verify_tfm = verify_tfm;
1837 mutex_unlock(&tconn->net_conf_update);
1841 if (tconn->cstate >= C_WF_REPORT_PARAMS)
1842 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1847 mutex_unlock(&tconn->net_conf_update);
1848 crypto_free_hash(csums_tfm);
1849 crypto_free_hash(verify_tfm);
1852 conn_reconfig_done(tconn);
1854 drbd_adm_finish(info, retcode);
1858 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1860 char hmac_name[CRYPTO_MAX_ALG_NAME];
1861 struct drbd_conf *mdev;
1862 struct net_conf *old_conf, *new_conf = NULL;
1863 struct crypto_hash *tfm = NULL;
1864 struct crypto_hash *integrity_w_tfm = NULL;
1865 struct crypto_hash *integrity_r_tfm = NULL;
1866 void *int_dig_in = NULL;
1867 void *int_dig_vv = NULL;
1868 struct drbd_tconn *oconn;
1869 struct drbd_tconn *tconn;
1870 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1871 enum drbd_ret_code retcode;
1875 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1876 if (!adm_ctx.reply_skb)
1878 if (retcode != NO_ERROR)
1881 tconn = adm_ctx.tconn;
1882 conn_reconfig_start(tconn);
1884 if (tconn->cstate > C_STANDALONE) {
1885 retcode = ERR_NET_CONFIGURED;
1889 /* allocation not in the IO path, cqueue thread context */
1890 new_conf = kmalloc(sizeof(struct net_conf), GFP_KERNEL);
1892 retcode = ERR_NOMEM;
1896 *new_conf = (struct net_conf) {
1897 {}, 0, /* my_addr */
1898 {}, 0, /* peer_addr */
1899 {}, 0, /* shared_secret */
1900 {}, 0, /* cram_hmac_alg */
1901 {}, 0, /* integrity_alg */
1902 {}, 0, /* verify_alg */
1903 {}, 0, /* csums_alg */
1904 DRBD_PROTOCOL_DEF, /* wire_protocol */
1905 DRBD_CONNECT_INT_DEF, /* try_connect_int */
1906 DRBD_TIMEOUT_DEF, /* timeout */
1907 DRBD_PING_INT_DEF, /* ping_int */
1908 DRBD_PING_TIMEO_DEF, /* ping_timeo */
1909 DRBD_SNDBUF_SIZE_DEF, /* sndbuf_size */
1910 DRBD_RCVBUF_SIZE_DEF, /* rcvbuf_size */
1911 DRBD_KO_COUNT_DEF, /* ko_count */
1912 DRBD_MAX_BUFFERS_DEF, /* max_buffers */
1913 DRBD_MAX_EPOCH_SIZE_DEF, /* max_epoch_size */
1914 DRBD_UNPLUG_WATERMARK_DEF, /* unplug_watermark */
1915 DRBD_AFTER_SB_0P_DEF, /* after_sb_0p */
1916 DRBD_AFTER_SB_1P_DEF, /* after_sb_1p */
1917 DRBD_AFTER_SB_2P_DEF, /* after_sb_2p */
1918 DRBD_RR_CONFLICT_DEF, /* rr_conflict */
1919 DRBD_ON_CONGESTION_DEF, /* on_congestion */
1920 DRBD_CONG_FILL_DEF, /* cong_fill */
1921 DRBD_CONG_EXTENTS_DEF, /* cong_extents */
1922 0, /* two_primaries */
1925 0, /* always_asbp */
1930 err = net_conf_from_attrs(new_conf, info);
1932 retcode = ERR_MANDATORY_TAG;
1933 drbd_msg_put_info(from_attrs_err_to_txt(err));
1937 retcode = check_net_options(tconn, new_conf);
1938 if (retcode != NO_ERROR)
1943 new_my_addr = (struct sockaddr *)&new_conf->my_addr;
1944 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
1946 /* No need to take drbd_cfg_rwsem here. All reconfiguration is
1947 * strictly serialized on genl_lock(). We are protected against
1948 * concurrent reconfiguration/addition/deletion */
1949 list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
1950 struct net_conf *nc;
1955 nc = rcu_dereference(oconn->net_conf);
1957 taken_addr = (struct sockaddr *)&nc->my_addr;
1958 if (new_conf->my_addr_len == nc->my_addr_len &&
1959 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
1960 retcode = ERR_LOCAL_ADDR;
1962 taken_addr = (struct sockaddr *)&nc->peer_addr;
1963 if (new_conf->peer_addr_len == nc->peer_addr_len &&
1964 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
1965 retcode = ERR_PEER_ADDR;
1968 if (retcode != NO_ERROR)
1972 if (new_conf->cram_hmac_alg[0] != 0) {
1973 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1974 new_conf->cram_hmac_alg);
1975 tfm = crypto_alloc_hash(hmac_name, 0, CRYPTO_ALG_ASYNC);
1978 retcode = ERR_AUTH_ALG;
1982 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
1983 retcode = ERR_AUTH_ALG_ND;
1988 if (new_conf->integrity_alg[0]) {
1989 integrity_w_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
1990 if (IS_ERR(integrity_w_tfm)) {
1991 integrity_w_tfm = NULL;
1992 retcode=ERR_INTEGRITY_ALG;
1996 if (!drbd_crypto_is_hash(crypto_hash_tfm(integrity_w_tfm))) {
1997 retcode=ERR_INTEGRITY_ALG_ND;
2001 integrity_r_tfm = crypto_alloc_hash(new_conf->integrity_alg, 0, CRYPTO_ALG_ASYNC);
2002 if (IS_ERR(integrity_r_tfm)) {
2003 integrity_r_tfm = NULL;
2004 retcode=ERR_INTEGRITY_ALG;
2009 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2011 /* allocation not in the IO path, cqueue thread context */
2012 if (integrity_w_tfm) {
2013 i = crypto_hash_digestsize(integrity_w_tfm);
2014 int_dig_in = kmalloc(i, GFP_KERNEL);
2016 retcode = ERR_NOMEM;
2019 int_dig_vv = kmalloc(i, GFP_KERNEL);
2021 retcode = ERR_NOMEM;
2026 conn_flush_workqueue(tconn);
2028 mutex_lock(&tconn->net_conf_update);
2029 old_conf = tconn->net_conf;
2031 retcode = ERR_NET_CONFIGURED;
2032 mutex_unlock(&tconn->net_conf_update);
2035 rcu_assign_pointer(tconn->net_conf, new_conf);
2037 conn_free_crypto(tconn);
2038 tconn->cram_hmac_tfm = tfm;
2039 tconn->integrity_w_tfm = integrity_w_tfm;
2040 tconn->integrity_r_tfm = integrity_r_tfm;
2041 tconn->int_dig_in = int_dig_in;
2042 tconn->int_dig_vv = int_dig_vv;
2044 mutex_unlock(&tconn->net_conf_update);
2047 idr_for_each_entry(&tconn->volumes, mdev, i) {
2053 retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2055 conn_reconfig_done(tconn);
2056 drbd_adm_finish(info, retcode);
2062 crypto_free_hash(tfm);
2063 crypto_free_hash(integrity_w_tfm);
2064 crypto_free_hash(integrity_r_tfm);
2067 conn_reconfig_done(tconn);
2069 drbd_adm_finish(info, retcode);
2073 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2075 enum drbd_state_rv rv;
2077 spin_lock_irq(&tconn->req_lock);
2078 if (tconn->cstate >= C_WF_CONNECTION)
2079 _conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
2080 spin_unlock_irq(&tconn->req_lock);
2084 rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING), 0);
2087 case SS_NOTHING_TO_DO:
2088 case SS_ALREADY_STANDALONE:
2090 case SS_PRIMARY_NOP:
2091 /* Our state checking code wants to see the peer outdated. */
2092 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2093 pdsk, D_OUTDATED), CS_VERBOSE);
2095 case SS_CW_FAILED_BY_PEER:
2096 /* The peer probably wants to see us outdated. */
2097 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2098 disk, D_OUTDATED), 0);
2099 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2100 conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD);
2105 /* no special handling necessary */
2111 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2113 struct disconnect_parms parms;
2114 struct drbd_tconn *tconn;
2115 enum drbd_state_rv rv;
2116 enum drbd_ret_code retcode;
2119 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2120 if (!adm_ctx.reply_skb)
2122 if (retcode != NO_ERROR)
2125 tconn = adm_ctx.tconn;
2126 memset(&parms, 0, sizeof(parms));
2127 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2128 err = disconnect_parms_from_attrs(&parms, info);
2130 retcode = ERR_MANDATORY_TAG;
2131 drbd_msg_put_info(from_attrs_err_to_txt(err));
2136 rv = conn_try_disconnect(tconn, parms.force_disconnect);
2137 if (rv < SS_SUCCESS)
2140 if (wait_event_interruptible(tconn->ping_wait,
2141 tconn->cstate != C_DISCONNECTING)) {
2142 /* Do not test for mdev->state.conn == C_STANDALONE, since
2143 someone else might connect us in the mean time! */
2150 drbd_adm_finish(info, retcode);
2154 void resync_after_online_grow(struct drbd_conf *mdev)
2156 int iass; /* I am sync source */
2158 dev_info(DEV, "Resync of new storage after online grow\n");
2159 if (mdev->state.role != mdev->state.peer)
2160 iass = (mdev->state.role == R_PRIMARY);
2162 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2165 drbd_start_resync(mdev, C_SYNC_SOURCE);
2167 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2170 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2172 struct resize_parms rs;
2173 struct drbd_conf *mdev;
2174 enum drbd_ret_code retcode;
2175 enum determine_dev_size dd;
2176 enum dds_flags ddsf;
2179 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2180 if (!adm_ctx.reply_skb)
2182 if (retcode != NO_ERROR)
2185 memset(&rs, 0, sizeof(struct resize_parms));
2186 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2187 err = resize_parms_from_attrs(&rs, info);
2189 retcode = ERR_MANDATORY_TAG;
2190 drbd_msg_put_info(from_attrs_err_to_txt(err));
2195 mdev = adm_ctx.mdev;
2196 if (mdev->state.conn > C_CONNECTED) {
2197 retcode = ERR_RESIZE_RESYNC;
2201 if (mdev->state.role == R_SECONDARY &&
2202 mdev->state.peer == R_SECONDARY) {
2203 retcode = ERR_NO_PRIMARY;
2207 if (!get_ldev(mdev)) {
2208 retcode = ERR_NO_DISK;
2212 if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2213 retcode = ERR_NEED_APV_93;
2217 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2218 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2220 mdev->ldev->dc.disk_size = (sector_t)rs.resize_size;
2221 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2222 dd = drbd_determine_dev_size(mdev, ddsf);
2225 if (dd == dev_size_error) {
2226 retcode = ERR_NOMEM_BITMAP;
2230 if (mdev->state.conn == C_CONNECTED) {
2232 set_bit(RESIZE_PENDING, &mdev->flags);
2234 drbd_send_uuids(mdev);
2235 drbd_send_sizes(mdev, 1, ddsf);
2239 drbd_adm_finish(info, retcode);
2243 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2245 enum drbd_ret_code retcode;
2246 cpumask_var_t new_cpu_mask;
2247 struct drbd_tconn *tconn;
2248 int *rs_plan_s = NULL;
2252 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2253 if (!adm_ctx.reply_skb)
2255 if (retcode != NO_ERROR)
2257 tconn = adm_ctx.tconn;
2259 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
2260 retcode = ERR_NOMEM;
2261 drbd_msg_put_info("unable to allocate cpumask");
2265 if (((struct drbd_genlmsghdr*)info->userhdr)->flags
2266 & DRBD_GENL_F_SET_DEFAULTS) {
2267 memset(&sc, 0, sizeof(struct res_opts));
2268 sc.on_no_data = DRBD_ON_NO_DATA_DEF;
2270 sc = tconn->res_opts;
2272 err = res_opts_from_attrs(&sc, info);
2274 retcode = ERR_MANDATORY_TAG;
2275 drbd_msg_put_info(from_attrs_err_to_txt(err));
2279 /* silently ignore cpu mask on UP kernel */
2280 if (nr_cpu_ids > 1 && sc.cpu_mask[0] != 0) {
2281 err = __bitmap_parse(sc.cpu_mask, 32, 0,
2282 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2284 conn_warn(tconn, "__bitmap_parse() failed with %d\n", err);
2285 retcode = ERR_CPU_MASK_PARSE;
2291 tconn->res_opts = sc;
2293 if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2294 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2295 drbd_calc_cpu_mask(tconn);
2296 tconn->receiver.reset_cpu_mask = 1;
2297 tconn->asender.reset_cpu_mask = 1;
2298 tconn->worker.reset_cpu_mask = 1;
2303 free_cpumask_var(new_cpu_mask);
2305 drbd_adm_finish(info, retcode);
2309 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2311 struct drbd_conf *mdev;
2312 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2314 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2315 if (!adm_ctx.reply_skb)
2317 if (retcode != NO_ERROR)
2320 mdev = adm_ctx.mdev;
2322 /* If there is still bitmap IO pending, probably because of a previous
2323 * resync just being finished, wait for it before requesting a new resync. */
2324 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2326 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2328 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2329 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2331 while (retcode == SS_NEED_CONNECTION) {
2332 spin_lock_irq(&mdev->tconn->req_lock);
2333 if (mdev->state.conn < C_CONNECTED)
2334 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2335 spin_unlock_irq(&mdev->tconn->req_lock);
2337 if (retcode != SS_NEED_CONNECTION)
2340 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2344 drbd_adm_finish(info, retcode);
2348 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2352 rv = drbd_bmio_set_n_write(mdev);
2353 drbd_suspend_al(mdev);
2357 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2358 union drbd_state mask, union drbd_state val)
2360 enum drbd_ret_code retcode;
2362 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2363 if (!adm_ctx.reply_skb)
2365 if (retcode != NO_ERROR)
2368 retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2370 drbd_adm_finish(info, retcode);
2374 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2376 return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2379 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2381 enum drbd_ret_code retcode;
2383 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2384 if (!adm_ctx.reply_skb)
2386 if (retcode != NO_ERROR)
2389 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2390 retcode = ERR_PAUSE_IS_SET;
2392 drbd_adm_finish(info, retcode);
2396 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2398 union drbd_dev_state s;
2399 enum drbd_ret_code retcode;
2401 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2402 if (!adm_ctx.reply_skb)
2404 if (retcode != NO_ERROR)
2407 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2408 s = adm_ctx.mdev->state;
2409 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2410 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2411 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2413 retcode = ERR_PAUSE_IS_CLEAR;
2418 drbd_adm_finish(info, retcode);
2422 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2424 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2427 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2429 struct drbd_conf *mdev;
2430 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2432 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2433 if (!adm_ctx.reply_skb)
2435 if (retcode != NO_ERROR)
2438 mdev = adm_ctx.mdev;
2439 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2440 drbd_uuid_new_current(mdev);
2441 clear_bit(NEW_CUR_UUID, &mdev->flags);
2443 drbd_suspend_io(mdev);
2444 retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2445 if (retcode == SS_SUCCESS) {
2446 if (mdev->state.conn < C_CONNECTED)
2447 tl_clear(mdev->tconn);
2448 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2449 tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2451 drbd_resume_io(mdev);
2454 drbd_adm_finish(info, retcode);
2458 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2460 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2463 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *conn_name, unsigned vnr)
2466 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2468 goto nla_put_failure;
2469 if (vnr != VOLUME_UNSPECIFIED)
2470 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2471 NLA_PUT_STRING(skb, T_ctx_conn_name, conn_name);
2472 nla_nest_end(skb, nla);
2477 nla_nest_cancel(skb, nla);
2481 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2482 const struct sib_info *sib)
2484 struct state_info *si = NULL; /* for sizeof(si->member); */
2485 struct net_conf *nc;
2489 int exclude_sensitive;
2491 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2492 * to. So we better exclude_sensitive information.
2494 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2495 * in the context of the requesting user process. Exclude sensitive
2496 * information, unless current has superuser.
2498 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2499 * relies on the current implementation of netlink_dump(), which
2500 * executes the dump callback successively from netlink_recvmsg(),
2501 * always in the context of the receiving process */
2502 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2504 got_ldev = get_ldev(mdev);
2506 /* We need to add connection name and volume number information still.
2507 * Minor number is in drbd_genlmsghdr. */
2508 if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2509 goto nla_put_failure;
2511 if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2512 goto nla_put_failure;
2515 if (disk_conf_to_skb(skb, &mdev->ldev->dc, exclude_sensitive))
2516 goto nla_put_failure;
2519 nc = rcu_dereference(mdev->tconn->net_conf);
2521 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2524 goto nla_put_failure;
2526 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2528 goto nla_put_failure;
2529 NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2530 NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2531 NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2532 NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2535 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2536 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2537 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2538 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2539 if (C_SYNC_SOURCE <= mdev->state.conn &&
2540 C_PAUSED_SYNC_T >= mdev->state.conn) {
2541 NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2542 NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2547 switch(sib->sib_reason) {
2548 case SIB_SYNC_PROGRESS:
2549 case SIB_GET_STATUS_REPLY:
2551 case SIB_STATE_CHANGE:
2552 NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2553 NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2555 case SIB_HELPER_POST:
2557 T_helper_exit_code, sib->helper_exit_code);
2559 case SIB_HELPER_PRE:
2560 NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2564 nla_nest_end(skb, nla);
2574 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2576 enum drbd_ret_code retcode;
2579 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2580 if (!adm_ctx.reply_skb)
2582 if (retcode != NO_ERROR)
2585 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2587 nlmsg_free(adm_ctx.reply_skb);
2591 drbd_adm_finish(info, retcode);
2595 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2597 struct drbd_conf *mdev;
2598 struct drbd_genlmsghdr *dh;
2599 struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2600 struct drbd_tconn *tconn = NULL;
2601 struct drbd_tconn *tmp;
2602 unsigned volume = cb->args[1];
2604 /* Open coded, deferred, iteration:
2605 * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2606 * idr_for_each_entry(&tconn->volumes, mdev, i) {
2610 * where tconn is cb->args[0];
2611 * and i is cb->args[1];
2613 * cb->args[2] indicates if we shall loop over all resources,
2614 * or just dump all volumes of a single resource.
2616 * This may miss entries inserted after this dump started,
2617 * or entries deleted before they are reached.
2619 * We need to make sure the mdev won't disappear while
2620 * we are looking at it, and revalidate our iterators
2621 * on each iteration.
2624 /* synchronize with conn_create()/conn_destroy() */
2625 down_read(&drbd_cfg_rwsem);
2626 /* revalidate iterator position */
2627 list_for_each_entry(tmp, &drbd_tconns, all_tconn) {
2629 /* first iteration */
2641 mdev = idr_get_next(&tconn->volumes, &volume);
2643 /* No more volumes to dump on this tconn.
2644 * Advance tconn iterator. */
2645 pos = list_entry(tconn->all_tconn.next,
2646 struct drbd_tconn, all_tconn);
2647 /* Did we dump any volume on this tconn yet? */
2649 /* If we reached the end of the list,
2650 * or only a single resource dump was requested,
2652 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2660 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2661 cb->nlh->nlmsg_seq, &drbd_genl_family,
2662 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2667 /* this is a tconn without a single volume */
2669 dh->ret_code = NO_ERROR;
2670 if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2671 genlmsg_cancel(skb, dh);
2673 genlmsg_end(skb, dh);
2677 D_ASSERT(mdev->vnr == volume);
2678 D_ASSERT(mdev->tconn == tconn);
2680 dh->minor = mdev_to_minor(mdev);
2681 dh->ret_code = NO_ERROR;
2683 if (nla_put_status_info(skb, mdev, NULL)) {
2684 genlmsg_cancel(skb, dh);
2687 genlmsg_end(skb, dh);
2691 up_read(&drbd_cfg_rwsem);
2692 /* where to start the next iteration */
2693 cb->args[0] = (long)pos;
2694 cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2696 /* No more tconns/volumes/minors found results in an empty skb.
2697 * Which will terminate the dump. */
2702 * Request status of all resources, or of all volumes within a single resource.
2704 * This is a dump, as the answer may not fit in a single reply skb otherwise.
2705 * Which means we cannot use the family->attrbuf or other such members, because
2706 * dump is NOT protected by the genl_lock(). During dump, we only have access
2707 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2709 * Once things are setup properly, we call into get_one_status().
2711 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2713 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2715 const char *conn_name;
2716 struct drbd_tconn *tconn;
2718 /* Is this a followup call? */
2720 /* ... of a single resource dump,
2721 * and the resource iterator has been advanced already? */
2722 if (cb->args[2] && cb->args[2] != cb->args[0])
2723 return 0; /* DONE. */
2727 /* First call (from netlink_dump_start). We need to figure out
2728 * which resource(s) the user wants us to dump. */
2729 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2730 nlmsg_attrlen(cb->nlh, hdrlen),
2731 DRBD_NLA_CFG_CONTEXT);
2733 /* No explicit context given. Dump all. */
2736 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
2737 /* context given, but no name present? */
2740 conn_name = nla_data(nla);
2741 tconn = conn_get_by_name(conn_name);
2746 kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2748 /* prime iterators, and set "filter" mode mark:
2749 * only dump this tconn. */
2750 cb->args[0] = (long)tconn;
2751 /* cb->args[1] = 0; passed in this way. */
2752 cb->args[2] = (long)tconn;
2755 return get_one_status(skb, cb);
2758 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2760 enum drbd_ret_code retcode;
2761 struct timeout_parms tp;
2764 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2765 if (!adm_ctx.reply_skb)
2767 if (retcode != NO_ERROR)
2771 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2772 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2775 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2777 nlmsg_free(adm_ctx.reply_skb);
2781 drbd_adm_finish(info, retcode);
2785 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2787 struct drbd_conf *mdev;
2788 enum drbd_ret_code retcode;
2790 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2791 if (!adm_ctx.reply_skb)
2793 if (retcode != NO_ERROR)
2796 mdev = adm_ctx.mdev;
2797 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2798 /* resume from last known position, if possible */
2799 struct start_ov_parms parms =
2800 { .ov_start_sector = mdev->ov_start_sector };
2801 int err = start_ov_parms_from_attrs(&parms, info);
2803 retcode = ERR_MANDATORY_TAG;
2804 drbd_msg_put_info(from_attrs_err_to_txt(err));
2807 /* w_make_ov_request expects position to be aligned */
2808 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2810 /* If there is still bitmap IO pending, e.g. previous resync or verify
2811 * just being finished, wait for it before requesting a new resync. */
2812 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2813 retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2815 drbd_adm_finish(info, retcode);
2820 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2822 struct drbd_conf *mdev;
2823 enum drbd_ret_code retcode;
2824 int skip_initial_sync = 0;
2826 struct new_c_uuid_parms args;
2828 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2829 if (!adm_ctx.reply_skb)
2831 if (retcode != NO_ERROR)
2834 mdev = adm_ctx.mdev;
2835 memset(&args, 0, sizeof(args));
2836 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2837 err = new_c_uuid_parms_from_attrs(&args, info);
2839 retcode = ERR_MANDATORY_TAG;
2840 drbd_msg_put_info(from_attrs_err_to_txt(err));
2845 mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2847 if (!get_ldev(mdev)) {
2848 retcode = ERR_NO_DISK;
2852 /* this is "skip initial sync", assume to be clean */
2853 if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2854 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2855 dev_info(DEV, "Preparing to skip initial sync\n");
2856 skip_initial_sync = 1;
2857 } else if (mdev->state.conn != C_STANDALONE) {
2858 retcode = ERR_CONNECTED;
2862 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2863 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2865 if (args.clear_bm) {
2866 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2867 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2869 dev_err(DEV, "Writing bitmap failed with %d\n",err);
2870 retcode = ERR_IO_MD_DISK;
2872 if (skip_initial_sync) {
2873 drbd_send_uuids_skip_initial_sync(mdev);
2874 _drbd_uuid_set(mdev, UI_BITMAP, 0);
2875 drbd_print_uuids(mdev, "cleared bitmap UUID");
2876 spin_lock_irq(&mdev->tconn->req_lock);
2877 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2879 spin_unlock_irq(&mdev->tconn->req_lock);
2887 mutex_unlock(mdev->state_mutex);
2889 drbd_adm_finish(info, retcode);
2893 static enum drbd_ret_code
2894 drbd_check_conn_name(const char *name)
2896 if (!name || !name[0]) {
2897 drbd_msg_put_info("connection name missing");
2898 return ERR_MANDATORY_TAG;
2900 /* if we want to use these in sysfs/configfs/debugfs some day,
2901 * we must not allow slashes */
2902 if (strchr(name, '/')) {
2903 drbd_msg_put_info("invalid connection name");
2904 return ERR_INVALID_REQUEST;
2909 int drbd_adm_create_connection(struct sk_buff *skb, struct genl_info *info)
2911 enum drbd_ret_code retcode;
2913 retcode = drbd_adm_prepare(skb, info, 0);
2914 if (!adm_ctx.reply_skb)
2916 if (retcode != NO_ERROR)
2919 retcode = drbd_check_conn_name(adm_ctx.conn_name);
2920 if (retcode != NO_ERROR)
2923 if (adm_ctx.tconn) {
2924 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
2925 retcode = ERR_INVALID_REQUEST;
2926 drbd_msg_put_info("connection exists");
2928 /* else: still NO_ERROR */
2932 if (!conn_create(adm_ctx.conn_name))
2933 retcode = ERR_NOMEM;
2935 drbd_adm_finish(info, retcode);
2939 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
2941 struct drbd_genlmsghdr *dh = info->userhdr;
2942 enum drbd_ret_code retcode;
2944 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2945 if (!adm_ctx.reply_skb)
2947 if (retcode != NO_ERROR)
2950 /* FIXME drop minor_count parameter, limit to MINORMASK */
2951 if (dh->minor >= minor_count) {
2952 drbd_msg_put_info("requested minor out of range");
2953 retcode = ERR_INVALID_REQUEST;
2956 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
2957 drbd_msg_put_info("requested volume id out of range");
2958 retcode = ERR_INVALID_REQUEST;
2962 /* drbd_adm_prepare made sure already
2963 * that mdev->tconn and mdev->vnr match the request. */
2965 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
2966 retcode = ERR_MINOR_EXISTS;
2967 /* else: still NO_ERROR */
2971 down_write(&drbd_cfg_rwsem);
2972 retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
2973 up_write(&drbd_cfg_rwsem);
2975 drbd_adm_finish(info, retcode);
2979 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
2981 if (mdev->state.disk == D_DISKLESS &&
2982 /* no need to be mdev->state.conn == C_STANDALONE &&
2983 * we may want to delete a minor from a live replication group.
2985 mdev->state.role == R_SECONDARY) {
2986 drbd_delete_device(mdev);
2989 return ERR_MINOR_CONFIGURED;
2992 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
2994 enum drbd_ret_code retcode;
2996 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2997 if (!adm_ctx.reply_skb)
2999 if (retcode != NO_ERROR)
3002 down_write(&drbd_cfg_rwsem);
3003 retcode = adm_delete_minor(adm_ctx.mdev);
3004 up_write(&drbd_cfg_rwsem);
3006 drbd_adm_finish(info, retcode);
3010 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3012 enum drbd_ret_code retcode;
3013 enum drbd_state_rv rv;
3014 struct drbd_conf *mdev;
3017 retcode = drbd_adm_prepare(skb, info, 0);
3018 if (!adm_ctx.reply_skb)
3020 if (retcode != NO_ERROR)
3023 if (!adm_ctx.tconn) {
3024 retcode = ERR_CONN_NOT_KNOWN;
3028 down_read(&drbd_cfg_rwsem);
3030 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3031 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3032 if (retcode < SS_SUCCESS) {
3033 drbd_msg_put_info("failed to demote");
3039 rv = conn_try_disconnect(adm_ctx.tconn, 0);
3040 if (rv < SS_SUCCESS) {
3041 retcode = rv; /* enum type mismatch! */
3042 drbd_msg_put_info("failed to disconnect");
3047 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3048 rv = adm_detach(mdev);
3049 if (rv < SS_SUCCESS) {
3050 retcode = rv; /* enum type mismatch! */
3051 drbd_msg_put_info("failed to detach");
3055 up_read(&drbd_cfg_rwsem);
3057 /* delete volumes */
3058 down_write(&drbd_cfg_rwsem);
3059 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3060 retcode = adm_delete_minor(mdev);
3061 if (retcode != NO_ERROR) {
3062 /* "can not happen" */
3063 drbd_msg_put_info("failed to delete volume");
3064 up_write(&drbd_cfg_rwsem);
3069 /* stop all threads */
3070 conn_reconfig_done(adm_ctx.tconn);
3072 /* delete connection */
3073 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3074 list_del(&adm_ctx.tconn->all_tconn);
3075 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3079 /* "can not happen" */
3080 retcode = ERR_CONN_IN_USE;
3081 drbd_msg_put_info("failed to delete connection");
3084 up_write(&drbd_cfg_rwsem);
3087 up_read(&drbd_cfg_rwsem);
3089 drbd_adm_finish(info, retcode);
3093 int drbd_adm_delete_connection(struct sk_buff *skb, struct genl_info *info)
3095 enum drbd_ret_code retcode;
3097 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3098 if (!adm_ctx.reply_skb)
3100 if (retcode != NO_ERROR)
3103 down_write(&drbd_cfg_rwsem);
3104 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3105 list_del(&adm_ctx.tconn->all_tconn);
3106 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3110 retcode = ERR_CONN_IN_USE;
3112 up_write(&drbd_cfg_rwsem);
3115 drbd_adm_finish(info, retcode);
3119 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3121 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3122 struct sk_buff *msg;
3123 struct drbd_genlmsghdr *d_out;
3127 seq = atomic_inc_return(&drbd_genl_seq);
3128 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3133 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3134 if (!d_out) /* cannot happen, but anyways. */
3135 goto nla_put_failure;
3136 d_out->minor = mdev_to_minor(mdev);
3137 d_out->ret_code = 0;
3139 if (nla_put_status_info(msg, mdev, sib))
3140 goto nla_put_failure;
3141 genlmsg_end(msg, d_out);
3142 err = drbd_genl_multicast_events(msg, 0);
3143 /* msg has been consumed or freed in netlink_broadcast() */
3144 if (err && err != -ESRCH)
3152 dev_err(DEV, "Error %d while broadcasting event. "
3153 "Event seq:%u sib_reason:%u\n",
3154 err, seq, sib->sib_reason);