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_new_resource(struct sk_buff *skb, struct genl_info *info);
51 int drbd_adm_del_resource(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>
79 #include <linux/genl_magic_func.h>
81 /* used blkdev_get_by_path, to claim our meta data device(s) */
82 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
84 /* Configuration is strictly serialized, because generic netlink message
85 * processing is strictly serialized by the genl_lock().
86 * Which means we can use one static global drbd_config_context struct.
88 static struct drbd_config_context {
89 /* assigned from drbd_genlmsghdr */
91 /* assigned from request attributes, if present */
93 #define VOLUME_UNSPECIFIED (-1U)
94 /* pointer into the request skb,
95 * limited lifetime! */
99 struct sk_buff *reply_skb;
100 /* pointer into reply buffer */
101 struct drbd_genlmsghdr *reply_dh;
102 /* resolved from attributes, if possible */
103 struct drbd_conf *mdev;
104 struct drbd_tconn *tconn;
107 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
109 genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
110 if (genlmsg_reply(skb, info))
111 printk(KERN_ERR "drbd: error sending genl reply\n");
114 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
115 * reason it could fail was no space in skb, and there are 4k available. */
116 int drbd_msg_put_info(const char *info)
118 struct sk_buff *skb = adm_ctx.reply_skb;
122 if (!info || !info[0])
125 nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
129 err = nla_put_string(skb, T_info_text, info);
131 nla_nest_cancel(skb, nla);
134 nla_nest_end(skb, nla);
138 /* This would be a good candidate for a "pre_doit" hook,
139 * and per-family private info->pointers.
140 * But we need to stay compatible with older kernels.
141 * If it returns successfully, adm_ctx members are valid.
143 #define DRBD_ADM_NEED_MINOR 1
144 #define DRBD_ADM_NEED_CONN 2
145 static int drbd_adm_prepare(struct sk_buff *skb, struct genl_info *info,
148 struct drbd_genlmsghdr *d_in = info->userhdr;
149 const u8 cmd = info->genlhdr->cmd;
152 memset(&adm_ctx, 0, sizeof(adm_ctx));
154 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
155 if (cmd != DRBD_ADM_GET_STATUS
156 && security_netlink_recv(skb, CAP_SYS_ADMIN))
159 adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
160 if (!adm_ctx.reply_skb) {
165 adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
166 info, &drbd_genl_family, 0, cmd);
167 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
169 if (!adm_ctx.reply_dh) {
174 adm_ctx.reply_dh->minor = d_in->minor;
175 adm_ctx.reply_dh->ret_code = NO_ERROR;
177 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
179 /* parse and validate only */
180 err = drbd_cfg_context_from_attrs(NULL, info);
184 /* It was present, and valid,
185 * copy it over to the reply skb. */
186 err = nla_put_nohdr(adm_ctx.reply_skb,
187 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
188 info->attrs[DRBD_NLA_CFG_CONTEXT]);
192 /* and assign stuff to the global adm_ctx */
193 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
194 adm_ctx.volume = nla ? nla_get_u32(nla) : VOLUME_UNSPECIFIED;
195 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
197 adm_ctx.resource_name = nla_data(nla);
199 adm_ctx.volume = VOLUME_UNSPECIFIED;
201 adm_ctx.minor = d_in->minor;
202 adm_ctx.mdev = minor_to_mdev(d_in->minor);
203 adm_ctx.tconn = conn_get_by_name(adm_ctx.resource_name);
205 if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
206 drbd_msg_put_info("unknown minor");
207 return ERR_MINOR_INVALID;
209 if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_CONN)) {
210 drbd_msg_put_info("unknown connection");
211 return ERR_INVALID_REQUEST;
214 /* some more paranoia, if the request was over-determined */
215 if (adm_ctx.mdev && adm_ctx.tconn &&
216 adm_ctx.mdev->tconn != adm_ctx.tconn) {
217 pr_warning("request: minor=%u, conn=%s; but that minor belongs to connection %s\n",
218 adm_ctx.minor, adm_ctx.resource_name,
219 adm_ctx.mdev->tconn->name);
220 drbd_msg_put_info("minor exists in different connection");
221 return ERR_INVALID_REQUEST;
224 adm_ctx.volume != VOLUME_UNSPECIFIED &&
225 adm_ctx.volume != adm_ctx.mdev->vnr) {
226 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
227 adm_ctx.minor, adm_ctx.volume,
228 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
229 drbd_msg_put_info("minor exists as different volume");
230 return ERR_INVALID_REQUEST;
236 nlmsg_free(adm_ctx.reply_skb);
237 adm_ctx.reply_skb = NULL;
241 static int drbd_adm_finish(struct genl_info *info, int retcode)
244 const char *resource_name = NULL;
247 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
248 adm_ctx.tconn = NULL;
251 if (!adm_ctx.reply_skb)
254 adm_ctx.reply_dh->ret_code = retcode;
256 nla = info->attrs[DRBD_NLA_CFG_CONTEXT];
258 int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
259 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
260 if (nla && !IS_ERR(nla))
261 resource_name = nla_data(nla);
264 drbd_adm_send_reply(adm_ctx.reply_skb, info);
268 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
274 nc = rcu_dereference(tconn->net_conf);
276 switch (((struct sockaddr *)nc->peer_addr)->sa_family) {
279 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
280 &((struct sockaddr_in6 *)nc->peer_addr)->sin6_addr);
284 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
285 &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
289 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
290 &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
292 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
297 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
299 char *envp[] = { "HOME=/",
301 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
302 (char[20]) { }, /* address family */
303 (char[60]) { }, /* address */
306 char *argv[] = {usermode_helper, cmd, mb, NULL };
310 snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
311 setup_khelper_env(mdev->tconn, envp);
313 /* The helper may take some time.
314 * write out any unsynced meta data changes now */
317 dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
318 sib.sib_reason = SIB_HELPER_PRE;
319 sib.helper_name = cmd;
320 drbd_bcast_event(mdev, &sib);
321 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
323 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
324 usermode_helper, cmd, mb,
325 (ret >> 8) & 0xff, ret);
327 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
328 usermode_helper, cmd, mb,
329 (ret >> 8) & 0xff, ret);
330 sib.sib_reason = SIB_HELPER_POST;
331 sib.helper_exit_code = ret;
332 drbd_bcast_event(mdev, &sib);
334 if (ret < 0) /* Ignore any ERRNOs we got. */
340 static void conn_md_sync(struct drbd_tconn *tconn)
342 struct drbd_conf *mdev;
346 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
347 kref_get(&mdev->kref);
350 kref_put(&mdev->kref, &drbd_minor_destroy);
356 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
358 char *envp[] = { "HOME=/",
360 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
361 (char[20]) { }, /* address family */
362 (char[60]) { }, /* address */
364 char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
367 setup_khelper_env(tconn, envp);
370 conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
371 /* TODO: conn_bcast_event() ?? */
373 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
375 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
376 usermode_helper, cmd, tconn->name,
377 (ret >> 8) & 0xff, ret);
379 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
380 usermode_helper, cmd, tconn->name,
381 (ret >> 8) & 0xff, ret);
382 /* TODO: conn_bcast_event() ?? */
384 if (ret < 0) /* Ignore any ERRNOs we got. */
390 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
392 enum drbd_fencing_p fp = FP_NOT_AVAIL;
393 struct drbd_conf *mdev;
397 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
398 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
399 fp = max_t(enum drbd_fencing_p, fp,
400 rcu_dereference(mdev->ldev->disk_conf)->fencing);
409 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
411 union drbd_state mask = { };
412 union drbd_state val = { };
413 enum drbd_fencing_p fp;
417 if (tconn->cstate >= C_WF_REPORT_PARAMS) {
418 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
422 fp = highest_fencing_policy(tconn);
425 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
432 r = conn_khelper(tconn, "fence-peer");
434 switch ((r>>8) & 0xff) {
435 case 3: /* peer is inconsistent */
436 ex_to_string = "peer is inconsistent or worse";
438 val.pdsk = D_INCONSISTENT;
440 case 4: /* peer got outdated, or was already outdated */
441 ex_to_string = "peer was fenced";
443 val.pdsk = D_OUTDATED;
445 case 5: /* peer was down */
446 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
447 /* we will(have) create(d) a new UUID anyways... */
448 ex_to_string = "peer is unreachable, assumed to be dead";
450 val.pdsk = D_OUTDATED;
452 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
455 case 6: /* Peer is primary, voluntarily outdate myself.
456 * This is useful when an unconnected R_SECONDARY is asked to
457 * become R_PRIMARY, but finds the other peer being active. */
458 ex_to_string = "peer is active";
459 conn_warn(tconn, "Peer is primary, outdating myself.\n");
461 val.disk = D_OUTDATED;
464 if (fp != FP_STONITH)
465 conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
466 ex_to_string = "peer was stonithed";
468 val.pdsk = D_OUTDATED;
471 /* The script is broken ... */
472 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
473 return false; /* Eventually leave IO frozen */
476 conn_info(tconn, "fence-peer helper returned %d (%s)\n",
477 (r>>8) & 0xff, ex_to_string);
482 conn_request_state(tconn, mask, val, CS_VERBOSE);
483 here, because we might were able to re-establish the connection in the
485 spin_lock_irq(&tconn->req_lock);
486 if (tconn->cstate < C_WF_REPORT_PARAMS)
487 _conn_request_state(tconn, mask, val, CS_VERBOSE);
488 spin_unlock_irq(&tconn->req_lock);
490 return conn_highest_pdsk(tconn) <= D_OUTDATED;
493 static int _try_outdate_peer_async(void *data)
495 struct drbd_tconn *tconn = (struct drbd_tconn *)data;
497 conn_try_outdate_peer(tconn);
499 kref_put(&tconn->kref, &conn_destroy);
503 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
505 struct task_struct *opa;
507 kref_get(&tconn->kref);
508 opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
510 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
511 kref_put(&tconn->kref, &conn_destroy);
516 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
518 const int max_tries = 4;
519 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
523 union drbd_state mask, val;
525 if (new_role == R_PRIMARY)
526 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
528 mutex_lock(mdev->state_mutex);
530 mask.i = 0; mask.role = R_MASK;
531 val.i = 0; val.role = new_role;
533 while (try++ < max_tries) {
534 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
536 /* in case we first succeeded to outdate,
537 * but now suddenly could establish a connection */
538 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
544 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
545 (mdev->state.disk < D_UP_TO_DATE &&
546 mdev->state.disk >= D_INCONSISTENT)) {
548 val.disk = D_UP_TO_DATE;
553 if (rv == SS_NO_UP_TO_DATE_DISK &&
554 mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
555 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
557 if (conn_try_outdate_peer(mdev->tconn)) {
558 val.disk = D_UP_TO_DATE;
564 if (rv == SS_NOTHING_TO_DO)
566 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
567 if (!conn_try_outdate_peer(mdev->tconn) && force) {
568 dev_warn(DEV, "Forced into split brain situation!\n");
570 val.pdsk = D_OUTDATED;
575 if (rv == SS_TWO_PRIMARIES) {
576 /* Maybe the peer is detected as dead very soon...
577 retry at most once more in this case. */
580 nc = rcu_dereference(mdev->tconn->net_conf);
581 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
583 schedule_timeout_interruptible(timeo);
588 if (rv < SS_SUCCESS) {
589 rv = _drbd_request_state(mdev, mask, val,
590 CS_VERBOSE + CS_WAIT_COMPLETE);
601 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
603 /* Wait until nothing is on the fly :) */
604 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
606 if (new_role == R_SECONDARY) {
607 set_disk_ro(mdev->vdisk, true);
608 if (get_ldev(mdev)) {
609 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
613 mutex_lock(&mdev->tconn->conf_update);
614 nc = mdev->tconn->net_conf;
616 nc->discard_my_data = 0; /* without copy; single bit op is atomic */
617 mutex_unlock(&mdev->tconn->conf_update);
619 set_disk_ro(mdev->vdisk, false);
620 if (get_ldev(mdev)) {
621 if (((mdev->state.conn < C_CONNECTED ||
622 mdev->state.pdsk <= D_FAILED)
623 && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
624 drbd_uuid_new_current(mdev);
626 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
631 /* writeout of activity log covered areas of the bitmap
632 * to stable storage done in after state change already */
634 if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
635 /* if this was forced, we should consider sync */
637 drbd_send_uuids(mdev);
638 drbd_send_state(mdev);
643 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
645 mutex_unlock(mdev->state_mutex);
649 static const char *from_attrs_err_to_txt(int err)
651 return err == -ENOMSG ? "required attribute missing" :
652 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
653 err == -EEXIST ? "can not change invariant setting" :
654 "invalid attribute value";
657 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
659 struct set_role_parms parms;
661 enum drbd_ret_code retcode;
663 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
664 if (!adm_ctx.reply_skb)
666 if (retcode != NO_ERROR)
669 memset(&parms, 0, sizeof(parms));
670 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
671 err = set_role_parms_from_attrs(&parms, info);
673 retcode = ERR_MANDATORY_TAG;
674 drbd_msg_put_info(from_attrs_err_to_txt(err));
679 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
680 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
682 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
684 drbd_adm_finish(info, retcode);
688 /* initializes the md.*_offset members, so we are able to find
689 * the on disk meta data */
690 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
691 struct drbd_backing_dev *bdev)
693 sector_t md_size_sect = 0;
697 meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx;
699 switch (meta_dev_idx) {
701 /* v07 style fixed size indexed meta data */
702 bdev->md.md_size_sect = MD_RESERVED_SECT;
703 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
704 bdev->md.al_offset = MD_AL_OFFSET;
705 bdev->md.bm_offset = MD_BM_OFFSET;
707 case DRBD_MD_INDEX_FLEX_EXT:
708 /* just occupy the full device; unit: sectors */
709 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
710 bdev->md.md_offset = 0;
711 bdev->md.al_offset = MD_AL_OFFSET;
712 bdev->md.bm_offset = MD_BM_OFFSET;
714 case DRBD_MD_INDEX_INTERNAL:
715 case DRBD_MD_INDEX_FLEX_INT:
716 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
717 /* al size is still fixed */
718 bdev->md.al_offset = -MD_AL_SECTORS;
719 /* we need (slightly less than) ~ this much bitmap sectors: */
720 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
721 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
722 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
723 md_size_sect = ALIGN(md_size_sect, 8);
725 /* plus the "drbd meta data super block",
726 * and the activity log; */
727 md_size_sect += MD_BM_OFFSET;
729 bdev->md.md_size_sect = md_size_sect;
730 /* bitmap offset is adjusted by 'super' block size */
731 bdev->md.bm_offset = -md_size_sect + MD_AL_OFFSET;
737 /* input size is expected to be in KB */
738 char *ppsize(char *buf, unsigned long long size)
740 /* Needs 9 bytes at max including trailing NUL:
741 * -1ULL ==> "16384 EB" */
742 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
744 while (size >= 10000 && base < sizeof(units)-1) {
746 size = (size >> 10) + !!(size & (1<<9));
749 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
754 /* there is still a theoretical deadlock when called from receiver
755 * on an D_INCONSISTENT R_PRIMARY:
756 * remote READ does inc_ap_bio, receiver would need to receive answer
757 * packet from remote to dec_ap_bio again.
758 * receiver receive_sizes(), comes here,
759 * waits for ap_bio_cnt == 0. -> deadlock.
760 * but this cannot happen, actually, because:
761 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
762 * (not connected, or bad/no disk on peer):
763 * see drbd_fail_request_early, ap_bio_cnt is zero.
764 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
765 * peer may not initiate a resize.
767 /* Note these are not to be confused with
768 * drbd_adm_suspend_io/drbd_adm_resume_io,
769 * which are (sub) state changes triggered by admin (drbdsetup),
770 * and can be long lived.
771 * This changes an mdev->flag, is triggered by drbd internals,
772 * and should be short-lived. */
773 void drbd_suspend_io(struct drbd_conf *mdev)
775 set_bit(SUSPEND_IO, &mdev->flags);
776 if (drbd_suspended(mdev))
778 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
781 void drbd_resume_io(struct drbd_conf *mdev)
783 clear_bit(SUSPEND_IO, &mdev->flags);
784 wake_up(&mdev->misc_wait);
788 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
789 * @mdev: DRBD device.
791 * Returns 0 on success, negative return values indicate errors.
792 * You should call drbd_md_sync() after calling this function.
794 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
796 sector_t prev_first_sect, prev_size; /* previous meta location */
797 sector_t la_size, u_size;
801 int md_moved, la_size_changed;
802 enum determine_dev_size rv = unchanged;
805 * application request passes inc_ap_bio,
806 * but then cannot get an AL-reference.
807 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
810 * Suspend IO right here.
811 * still lock the act_log to not trigger ASSERTs there.
813 drbd_suspend_io(mdev);
815 /* no wait necessary anymore, actually we could assert that */
816 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
818 prev_first_sect = drbd_md_first_sector(mdev->ldev);
819 prev_size = mdev->ldev->md.md_size_sect;
820 la_size = mdev->ldev->md.la_size_sect;
822 /* TODO: should only be some assert here, not (re)init... */
823 drbd_md_set_sector_offsets(mdev, mdev->ldev);
826 u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
828 size = drbd_new_dev_size(mdev, mdev->ldev, u_size, flags & DDSF_FORCED);
830 if (drbd_get_capacity(mdev->this_bdev) != size ||
831 drbd_bm_capacity(mdev) != size) {
833 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
835 /* currently there is only one error: ENOMEM! */
836 size = drbd_bm_capacity(mdev)>>1;
838 dev_err(DEV, "OUT OF MEMORY! "
839 "Could not allocate bitmap!\n");
841 dev_err(DEV, "BM resizing failed. "
842 "Leaving size unchanged at size = %lu KB\n",
843 (unsigned long)size);
847 /* racy, see comments above. */
848 drbd_set_my_capacity(mdev, size);
849 mdev->ldev->md.la_size_sect = size;
850 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
851 (unsigned long long)size>>1);
853 if (rv == dev_size_error)
856 la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
858 md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
859 || prev_size != mdev->ldev->md.md_size_sect;
861 if (la_size_changed || md_moved) {
864 drbd_al_shrink(mdev); /* All extents inactive. */
865 dev_info(DEV, "Writing the whole bitmap, %s\n",
866 la_size_changed && md_moved ? "size changed and md moved" :
867 la_size_changed ? "size changed" : "md moved");
868 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
869 err = drbd_bitmap_io(mdev, &drbd_bm_write,
870 "size changed", BM_LOCKED_MASK);
875 drbd_md_mark_dirty(mdev);
883 lc_unlock(mdev->act_log);
884 wake_up(&mdev->al_wait);
885 drbd_resume_io(mdev);
891 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
892 sector_t u_size, int assume_peer_has_space)
894 sector_t p_size = mdev->p_size; /* partner's disk size. */
895 sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
896 sector_t m_size; /* my size */
899 m_size = drbd_get_max_capacity(bdev);
901 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
902 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
906 if (p_size && m_size) {
907 size = min_t(sector_t, p_size, m_size);
911 if (m_size && m_size < size)
913 if (p_size && p_size < size)
924 dev_err(DEV, "Both nodes diskless!\n");
928 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
929 (unsigned long)u_size>>1, (unsigned long)size>>1);
938 * drbd_check_al_size() - Ensures that the AL is of the right size
939 * @mdev: DRBD device.
941 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
942 * failed, and 0 on success. You should call drbd_md_sync() after you called
945 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
947 struct lru_cache *n, *t;
948 struct lc_element *e;
953 mdev->act_log->nr_elements == dc->al_extents)
958 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
959 dc->al_extents, sizeof(struct lc_element), 0);
962 dev_err(DEV, "Cannot allocate act_log lru!\n");
965 spin_lock_irq(&mdev->al_lock);
967 for (i = 0; i < t->nr_elements; i++) {
968 e = lc_element_by_index(t, i);
970 dev_err(DEV, "refcnt(%d)==%d\n",
971 e->lc_number, e->refcnt);
977 spin_unlock_irq(&mdev->al_lock);
979 dev_err(DEV, "Activity log still in use!\n");
986 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
990 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
992 struct request_queue * const q = mdev->rq_queue;
993 int max_hw_sectors = max_bio_size >> 9;
994 int max_segments = 0;
996 if (get_ldev_if_state(mdev, D_ATTACHING)) {
997 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
999 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1001 max_segments = rcu_dereference(mdev->ldev->disk_conf)->max_bio_bvecs;
1006 blk_queue_logical_block_size(q, 512);
1007 blk_queue_max_hw_sectors(q, max_hw_sectors);
1008 /* This is the workaround for "bio would need to, but cannot, be split" */
1009 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1010 blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
1012 if (get_ldev_if_state(mdev, D_ATTACHING)) {
1013 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1015 blk_queue_stack_limits(q, b);
1017 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1018 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1019 q->backing_dev_info.ra_pages,
1020 b->backing_dev_info.ra_pages);
1021 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1027 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1029 int now, new, local, peer;
1031 now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1032 local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1033 peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1035 if (get_ldev_if_state(mdev, D_ATTACHING)) {
1036 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1037 mdev->local_max_bio_size = local;
1041 /* We may ignore peer limits if the peer is modern enough.
1042 Because new from 8.3.8 onwards the peer can use multiple
1043 BIOs for a single peer_request */
1044 if (mdev->state.conn >= C_CONNECTED) {
1045 if (mdev->tconn->agreed_pro_version < 94)
1046 peer = mdev->peer_max_bio_size;
1047 else if (mdev->tconn->agreed_pro_version == 94)
1048 peer = DRBD_MAX_SIZE_H80_PACKET;
1049 else /* drbd 8.3.8 onwards */
1050 peer = DRBD_MAX_BIO_SIZE;
1053 new = min_t(int, local, peer);
1055 if (mdev->state.role == R_PRIMARY && new < now)
1056 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
1059 dev_info(DEV, "max BIO size = %u\n", new);
1061 drbd_setup_queue_param(mdev, new);
1064 /* Starts the worker thread */
1065 static void conn_reconfig_start(struct drbd_tconn *tconn)
1067 drbd_thread_start(&tconn->worker);
1068 conn_flush_workqueue(tconn);
1071 /* if still unconfigured, stops worker again. */
1072 static void conn_reconfig_done(struct drbd_tconn *tconn)
1075 spin_lock_irq(&tconn->req_lock);
1076 stop_threads = conn_all_vols_unconf(tconn);
1077 spin_unlock_irq(&tconn->req_lock);
1079 /* asender is implicitly stopped by receiver
1080 * in conn_disconnect() */
1081 drbd_thread_stop(&tconn->receiver);
1082 drbd_thread_stop(&tconn->worker);
1086 /* Make sure IO is suspended before calling this function(). */
1087 static void drbd_suspend_al(struct drbd_conf *mdev)
1091 if (!lc_try_lock(mdev->act_log)) {
1092 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1096 drbd_al_shrink(mdev);
1097 spin_lock_irq(&mdev->tconn->req_lock);
1098 if (mdev->state.conn < C_CONNECTED)
1099 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1100 spin_unlock_irq(&mdev->tconn->req_lock);
1101 lc_unlock(mdev->act_log);
1104 dev_info(DEV, "Suspended AL updates\n");
1108 static bool should_set_defaults(struct genl_info *info)
1110 unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1111 return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1114 static void enforce_disk_conf_limits(struct disk_conf *dc)
1116 if (dc->al_extents < DRBD_AL_EXTENTS_MIN)
1117 dc->al_extents = DRBD_AL_EXTENTS_MIN;
1118 if (dc->al_extents > DRBD_AL_EXTENTS_MAX)
1119 dc->al_extents = DRBD_AL_EXTENTS_MAX;
1121 if (dc->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1122 dc->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1125 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1127 enum drbd_ret_code retcode;
1128 struct drbd_conf *mdev;
1129 struct disk_conf *new_disk_conf, *old_disk_conf;
1130 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1133 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1134 if (!adm_ctx.reply_skb)
1136 if (retcode != NO_ERROR)
1139 mdev = adm_ctx.mdev;
1141 /* we also need a disk
1142 * to change the options on */
1143 if (!get_ldev(mdev)) {
1144 retcode = ERR_NO_DISK;
1148 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1149 if (!new_disk_conf) {
1150 retcode = ERR_NOMEM;
1154 mutex_lock(&mdev->tconn->conf_update);
1155 old_disk_conf = mdev->ldev->disk_conf;
1156 *new_disk_conf = *old_disk_conf;
1157 if (should_set_defaults(info))
1158 set_disk_conf_defaults(new_disk_conf);
1160 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1161 if (err && err != -ENOMSG) {
1162 retcode = ERR_MANDATORY_TAG;
1163 drbd_msg_put_info(from_attrs_err_to_txt(err));
1166 if (!expect(new_disk_conf->resync_rate >= 1))
1167 new_disk_conf->resync_rate = 1;
1169 enforce_disk_conf_limits(new_disk_conf);
1171 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1172 if (fifo_size != mdev->rs_plan_s->size) {
1173 new_plan = fifo_alloc(fifo_size);
1175 dev_err(DEV, "kmalloc of fifo_buffer failed");
1176 retcode = ERR_NOMEM;
1181 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1182 drbd_al_shrink(mdev);
1183 err = drbd_check_al_size(mdev, new_disk_conf);
1184 lc_unlock(mdev->act_log);
1185 wake_up(&mdev->al_wait);
1188 retcode = ERR_NOMEM;
1192 write_lock_irq(&global_state_lock);
1193 retcode = drbd_resync_after_valid(mdev, new_disk_conf->resync_after);
1194 if (retcode == NO_ERROR) {
1195 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
1196 drbd_resync_after_changed(mdev);
1198 write_unlock_irq(&global_state_lock);
1200 if (retcode != NO_ERROR)
1204 old_plan = mdev->rs_plan_s;
1205 rcu_assign_pointer(mdev->rs_plan_s, new_plan);
1208 mutex_unlock(&mdev->tconn->conf_update);
1211 if (mdev->state.conn >= C_CONNECTED)
1212 drbd_send_sync_param(mdev);
1215 kfree(old_disk_conf);
1220 mutex_unlock(&mdev->tconn->conf_update);
1222 kfree(new_disk_conf);
1227 drbd_adm_finish(info, retcode);
1231 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1233 struct drbd_conf *mdev;
1235 enum drbd_ret_code retcode;
1236 enum determine_dev_size dd;
1237 sector_t max_possible_sectors;
1238 sector_t min_md_device_sectors;
1239 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1240 struct disk_conf *new_disk_conf = NULL;
1241 struct block_device *bdev;
1242 struct lru_cache *resync_lru = NULL;
1243 struct fifo_buffer *new_plan = NULL;
1244 union drbd_state ns, os;
1245 enum drbd_state_rv rv;
1246 struct net_conf *nc;
1247 int cp_discovered = 0;
1249 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1250 if (!adm_ctx.reply_skb)
1252 if (retcode != NO_ERROR)
1255 mdev = adm_ctx.mdev;
1256 conn_reconfig_start(mdev->tconn);
1258 /* if you want to reconfigure, please tear down first */
1259 if (mdev->state.disk > D_DISKLESS) {
1260 retcode = ERR_DISK_CONFIGURED;
1263 /* It may just now have detached because of IO error. Make sure
1264 * drbd_ldev_destroy is done already, we may end up here very fast,
1265 * e.g. if someone calls attach from the on-io-error handler,
1266 * to realize a "hot spare" feature (not that I'd recommend that) */
1267 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1269 /* allocation not in the IO path, drbdsetup context */
1270 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1272 retcode = ERR_NOMEM;
1275 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1276 if (!new_disk_conf) {
1277 retcode = ERR_NOMEM;
1280 nbc->disk_conf = new_disk_conf;
1282 set_disk_conf_defaults(new_disk_conf);
1283 err = disk_conf_from_attrs(new_disk_conf, info);
1285 retcode = ERR_MANDATORY_TAG;
1286 drbd_msg_put_info(from_attrs_err_to_txt(err));
1290 enforce_disk_conf_limits(new_disk_conf);
1292 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1294 retcode = ERR_NOMEM;
1298 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1299 retcode = ERR_MD_IDX_INVALID;
1304 nc = rcu_dereference(mdev->tconn->net_conf);
1306 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1308 retcode = ERR_STONITH_AND_PROT_A;
1314 bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1315 FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1317 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1319 retcode = ERR_OPEN_DISK;
1322 nbc->backing_bdev = bdev;
1325 * meta_dev_idx >= 0: external fixed size, possibly multiple
1326 * drbd sharing one meta device. TODO in that case, paranoia
1327 * check that [md_bdev, meta_dev_idx] is not yet used by some
1328 * other drbd minor! (if you use drbd.conf + drbdadm, that
1329 * should check it for you already; but if you don't, or
1330 * someone fooled it, we need to double check here)
1332 bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1333 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1334 (new_disk_conf->meta_dev_idx < 0) ?
1335 (void *)mdev : (void *)drbd_m_holder);
1337 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1339 retcode = ERR_OPEN_MD_DISK;
1342 nbc->md_bdev = bdev;
1344 if ((nbc->backing_bdev == nbc->md_bdev) !=
1345 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1346 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1347 retcode = ERR_MD_IDX_INVALID;
1351 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1352 1, 61, sizeof(struct bm_extent),
1353 offsetof(struct bm_extent, lce));
1355 retcode = ERR_NOMEM;
1359 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1360 drbd_md_set_sector_offsets(mdev, nbc);
1362 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1363 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1364 (unsigned long long) drbd_get_max_capacity(nbc),
1365 (unsigned long long) new_disk_conf->disk_size);
1366 retcode = ERR_DISK_TOO_SMALL;
1370 if (new_disk_conf->meta_dev_idx < 0) {
1371 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1372 /* at least one MB, otherwise it does not make sense */
1373 min_md_device_sectors = (2<<10);
1375 max_possible_sectors = DRBD_MAX_SECTORS;
1376 min_md_device_sectors = MD_RESERVED_SECT * (new_disk_conf->meta_dev_idx + 1);
1379 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1380 retcode = ERR_MD_DISK_TOO_SMALL;
1381 dev_warn(DEV, "refusing attach: md-device too small, "
1382 "at least %llu sectors needed for this meta-disk type\n",
1383 (unsigned long long) min_md_device_sectors);
1387 /* Make sure the new disk is big enough
1388 * (we may currently be R_PRIMARY with no local disk...) */
1389 if (drbd_get_max_capacity(nbc) <
1390 drbd_get_capacity(mdev->this_bdev)) {
1391 retcode = ERR_DISK_TOO_SMALL;
1395 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1397 if (nbc->known_size > max_possible_sectors) {
1398 dev_warn(DEV, "==> truncating very big lower level device "
1399 "to currently maximum possible %llu sectors <==\n",
1400 (unsigned long long) max_possible_sectors);
1401 if (new_disk_conf->meta_dev_idx >= 0)
1402 dev_warn(DEV, "==>> using internal or flexible "
1403 "meta data may help <<==\n");
1406 drbd_suspend_io(mdev);
1407 /* also wait for the last barrier ack. */
1408 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1409 /* and for any other previously queued work */
1410 drbd_flush_workqueue(mdev);
1412 rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1413 retcode = rv; /* FIXME: Type mismatch. */
1414 drbd_resume_io(mdev);
1415 if (rv < SS_SUCCESS)
1418 if (!get_ldev_if_state(mdev, D_ATTACHING))
1419 goto force_diskless;
1421 drbd_md_set_sector_offsets(mdev, nbc);
1423 if (!mdev->bitmap) {
1424 if (drbd_bm_init(mdev)) {
1425 retcode = ERR_NOMEM;
1426 goto force_diskless_dec;
1430 retcode = drbd_md_read(mdev, nbc);
1431 if (retcode != NO_ERROR)
1432 goto force_diskless_dec;
1434 if (mdev->state.conn < C_CONNECTED &&
1435 mdev->state.role == R_PRIMARY &&
1436 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1437 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1438 (unsigned long long)mdev->ed_uuid);
1439 retcode = ERR_DATA_NOT_CURRENT;
1440 goto force_diskless_dec;
1443 /* Since we are diskless, fix the activity log first... */
1444 if (drbd_check_al_size(mdev, new_disk_conf)) {
1445 retcode = ERR_NOMEM;
1446 goto force_diskless_dec;
1449 /* Prevent shrinking of consistent devices ! */
1450 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1451 drbd_new_dev_size(mdev, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1452 dev_warn(DEV, "refusing to truncate a consistent device\n");
1453 retcode = ERR_DISK_TOO_SMALL;
1454 goto force_diskless_dec;
1457 if (!drbd_al_read_log(mdev, nbc)) {
1458 retcode = ERR_IO_MD_DISK;
1459 goto force_diskless_dec;
1462 /* Reset the "barriers don't work" bits here, then force meta data to
1463 * be written, to ensure we determine if barriers are supported. */
1464 if (new_disk_conf->md_flushes)
1465 clear_bit(MD_NO_FUA, &mdev->flags);
1467 set_bit(MD_NO_FUA, &mdev->flags);
1469 /* Point of no return reached.
1470 * Devices and memory are no longer released by error cleanup below.
1471 * now mdev takes over responsibility, and the state engine should
1472 * clean it up somewhere. */
1473 D_ASSERT(mdev->ldev == NULL);
1475 mdev->resync = resync_lru;
1476 mdev->rs_plan_s = new_plan;
1479 new_disk_conf = NULL;
1482 mdev->write_ordering = WO_bdev_flush;
1483 drbd_bump_write_ordering(mdev, WO_bdev_flush);
1485 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1486 set_bit(CRASHED_PRIMARY, &mdev->flags);
1488 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1490 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1491 !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod)) {
1492 set_bit(CRASHED_PRIMARY, &mdev->flags);
1501 drbd_reconsider_max_bio_size(mdev);
1503 /* If I am currently not R_PRIMARY,
1504 * but meta data primary indicator is set,
1505 * I just now recover from a hard crash,
1506 * and have been R_PRIMARY before that crash.
1508 * Now, if I had no connection before that crash
1509 * (have been degraded R_PRIMARY), chances are that
1510 * I won't find my peer now either.
1512 * In that case, and _only_ in that case,
1513 * we use the degr-wfc-timeout instead of the default,
1514 * so we can automatically recover from a crash of a
1515 * degraded but active "cluster" after a certain timeout.
1517 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1518 if (mdev->state.role != R_PRIMARY &&
1519 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1520 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1521 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1523 dd = drbd_determine_dev_size(mdev, 0);
1524 if (dd == dev_size_error) {
1525 retcode = ERR_NOMEM_BITMAP;
1526 goto force_diskless_dec;
1527 } else if (dd == grew)
1528 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1530 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1531 dev_info(DEV, "Assuming that all blocks are out of sync "
1532 "(aka FullSync)\n");
1533 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1534 "set_n_write from attaching", BM_LOCKED_MASK)) {
1535 retcode = ERR_IO_MD_DISK;
1536 goto force_diskless_dec;
1539 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1540 "read from attaching", BM_LOCKED_MASK)) {
1541 retcode = ERR_IO_MD_DISK;
1542 goto force_diskless_dec;
1546 if (cp_discovered) {
1547 drbd_al_apply_to_bm(mdev);
1548 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1549 "crashed primary apply AL", BM_LOCKED_MASK)) {
1550 retcode = ERR_IO_MD_DISK;
1551 goto force_diskless_dec;
1555 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1556 drbd_suspend_al(mdev); /* IO is still suspended here... */
1558 spin_lock_irq(&mdev->tconn->req_lock);
1559 os = drbd_read_state(mdev);
1561 /* If MDF_CONSISTENT is not set go into inconsistent state,
1562 otherwise investigate MDF_WasUpToDate...
1563 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1564 otherwise into D_CONSISTENT state.
1566 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1567 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1568 ns.disk = D_CONSISTENT;
1570 ns.disk = D_OUTDATED;
1572 ns.disk = D_INCONSISTENT;
1575 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1576 ns.pdsk = D_OUTDATED;
1579 if (ns.disk == D_CONSISTENT &&
1580 (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1581 ns.disk = D_UP_TO_DATE;
1584 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1585 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1586 this point, because drbd_request_state() modifies these
1589 /* In case we are C_CONNECTED postpone any decision on the new disk
1590 state after the negotiation phase. */
1591 if (mdev->state.conn == C_CONNECTED) {
1592 mdev->new_state_tmp.i = ns.i;
1594 ns.disk = D_NEGOTIATING;
1596 /* We expect to receive up-to-date UUIDs soon.
1597 To avoid a race in receive_state, free p_uuid while
1598 holding req_lock. I.e. atomic with the state change */
1599 kfree(mdev->p_uuid);
1600 mdev->p_uuid = NULL;
1603 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1604 spin_unlock_irq(&mdev->tconn->req_lock);
1606 if (rv < SS_SUCCESS)
1607 goto force_diskless_dec;
1609 if (mdev->state.role == R_PRIMARY)
1610 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1612 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1614 drbd_md_mark_dirty(mdev);
1617 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1619 conn_reconfig_done(mdev->tconn);
1620 drbd_adm_finish(info, retcode);
1626 drbd_force_state(mdev, NS(disk, D_FAILED));
1629 conn_reconfig_done(mdev->tconn);
1631 if (nbc->backing_bdev)
1632 blkdev_put(nbc->backing_bdev,
1633 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1635 blkdev_put(nbc->md_bdev,
1636 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1639 kfree(new_disk_conf);
1640 lc_destroy(resync_lru);
1644 drbd_adm_finish(info, retcode);
1648 static int adm_detach(struct drbd_conf *mdev)
1650 enum drbd_state_rv retcode;
1652 drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1653 retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1654 /* D_FAILED will transition to DISKLESS. */
1655 ret = wait_event_interruptible(mdev->misc_wait,
1656 mdev->state.disk != D_FAILED);
1657 drbd_resume_io(mdev);
1658 if ((int)retcode == (int)SS_IS_DISKLESS)
1659 retcode = SS_NOTHING_TO_DO;
1665 /* Detaching the disk is a process in multiple stages. First we need to lock
1666 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1667 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1668 * internal references as well.
1669 * Only then we have finally detached. */
1670 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1672 enum drbd_ret_code retcode;
1674 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1675 if (!adm_ctx.reply_skb)
1677 if (retcode != NO_ERROR)
1680 retcode = adm_detach(adm_ctx.mdev);
1682 drbd_adm_finish(info, retcode);
1686 static bool conn_resync_running(struct drbd_tconn *tconn)
1688 struct drbd_conf *mdev;
1693 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1694 if (mdev->state.conn == C_SYNC_SOURCE ||
1695 mdev->state.conn == C_SYNC_TARGET ||
1696 mdev->state.conn == C_PAUSED_SYNC_S ||
1697 mdev->state.conn == C_PAUSED_SYNC_T) {
1707 static bool conn_ov_running(struct drbd_tconn *tconn)
1709 struct drbd_conf *mdev;
1714 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1715 if (mdev->state.conn == C_VERIFY_S ||
1716 mdev->state.conn == C_VERIFY_T) {
1726 static enum drbd_ret_code
1727 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1729 struct drbd_conf *mdev;
1732 if (old_conf && tconn->cstate == C_WF_REPORT_PARAMS && tconn->agreed_pro_version < 100) {
1733 if (new_conf->wire_protocol != old_conf->wire_protocol)
1734 return ERR_NEED_APV_100;
1736 if (new_conf->two_primaries != old_conf->two_primaries)
1737 return ERR_NEED_APV_100;
1739 if (!new_conf->integrity_alg != !old_conf->integrity_alg)
1740 return ERR_NEED_APV_100;
1742 if (strcmp(new_conf->integrity_alg, old_conf->integrity_alg))
1743 return ERR_NEED_APV_100;
1746 if (!new_conf->two_primaries &&
1747 conn_highest_role(tconn) == R_PRIMARY &&
1748 conn_highest_peer(tconn) == R_PRIMARY)
1749 return ERR_NEED_ALLOW_TWO_PRI;
1751 if (new_conf->two_primaries &&
1752 (new_conf->wire_protocol != DRBD_PROT_C))
1753 return ERR_NOT_PROTO_C;
1755 idr_for_each_entry(&tconn->volumes, mdev, i) {
1756 if (get_ldev(mdev)) {
1757 enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1759 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1760 return ERR_STONITH_AND_PROT_A;
1762 if (mdev->state.role == R_PRIMARY && new_conf->discard_my_data)
1766 if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1767 return ERR_CONG_NOT_PROTO_A;
1772 static enum drbd_ret_code
1773 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1775 static enum drbd_ret_code rv;
1776 struct drbd_conf *mdev;
1780 rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1783 /* tconn->volumes protected by genl_lock() here */
1784 idr_for_each_entry(&tconn->volumes, mdev, i) {
1785 if (!mdev->bitmap) {
1786 if(drbd_bm_init(mdev))
1795 struct crypto_hash *verify_tfm;
1796 struct crypto_hash *csums_tfm;
1797 struct crypto_hash *cram_hmac_tfm;
1798 struct crypto_hash *integrity_tfm;
1804 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1809 *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1818 static enum drbd_ret_code
1819 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1821 char hmac_name[CRYPTO_MAX_ALG_NAME];
1822 enum drbd_ret_code rv;
1825 rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1829 rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1833 rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1837 if (new_conf->cram_hmac_alg[0] != 0) {
1838 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1839 new_conf->cram_hmac_alg);
1841 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1844 if (crypto->integrity_tfm) {
1845 hash_size = crypto_hash_digestsize(crypto->integrity_tfm);
1846 crypto->int_dig_in = kmalloc(hash_size, GFP_KERNEL);
1847 if (!crypto->int_dig_in)
1849 crypto->int_dig_vv = kmalloc(hash_size, GFP_KERNEL);
1850 if (!crypto->int_dig_vv)
1857 static void free_crypto(struct crypto *crypto)
1859 kfree(crypto->int_dig_in);
1860 kfree(crypto->int_dig_vv);
1861 crypto_free_hash(crypto->cram_hmac_tfm);
1862 crypto_free_hash(crypto->integrity_tfm);
1863 crypto_free_hash(crypto->csums_tfm);
1864 crypto_free_hash(crypto->verify_tfm);
1867 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1869 enum drbd_ret_code retcode;
1870 struct drbd_tconn *tconn;
1871 struct net_conf *old_conf, *new_conf = NULL;
1873 int ovr; /* online verify running */
1874 int rsr; /* re-sync running */
1875 struct crypto crypto = { };
1877 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1878 if (!adm_ctx.reply_skb)
1880 if (retcode != NO_ERROR)
1883 tconn = adm_ctx.tconn;
1885 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1887 retcode = ERR_NOMEM;
1891 conn_reconfig_start(tconn);
1893 mutex_lock(&tconn->data.mutex);
1894 mutex_lock(&tconn->conf_update);
1895 old_conf = tconn->net_conf;
1898 drbd_msg_put_info("net conf missing, try connect");
1899 retcode = ERR_INVALID_REQUEST;
1903 *new_conf = *old_conf;
1904 if (should_set_defaults(info))
1905 set_net_conf_defaults(new_conf);
1907 err = net_conf_from_attrs_for_change(new_conf, info);
1908 if (err && err != -ENOMSG) {
1909 retcode = ERR_MANDATORY_TAG;
1910 drbd_msg_put_info(from_attrs_err_to_txt(err));
1914 retcode = check_net_options(tconn, new_conf);
1915 if (retcode != NO_ERROR)
1918 /* re-sync running */
1919 rsr = conn_resync_running(tconn);
1920 if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1921 retcode = ERR_CSUMS_RESYNC_RUNNING;
1925 /* online verify running */
1926 ovr = conn_ov_running(tconn);
1927 if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1928 retcode = ERR_VERIFY_RUNNING;
1932 retcode = alloc_crypto(&crypto, new_conf);
1933 if (retcode != NO_ERROR)
1936 rcu_assign_pointer(tconn->net_conf, new_conf);
1939 crypto_free_hash(tconn->csums_tfm);
1940 tconn->csums_tfm = crypto.csums_tfm;
1941 crypto.csums_tfm = NULL;
1944 crypto_free_hash(tconn->verify_tfm);
1945 tconn->verify_tfm = crypto.verify_tfm;
1946 crypto.verify_tfm = NULL;
1949 kfree(tconn->int_dig_in);
1950 tconn->int_dig_in = crypto.int_dig_in;
1951 kfree(tconn->int_dig_vv);
1952 tconn->int_dig_vv = crypto.int_dig_vv;
1953 crypto_free_hash(tconn->integrity_tfm);
1954 tconn->integrity_tfm = crypto.integrity_tfm;
1955 if (tconn->cstate >= C_WF_REPORT_PARAMS && tconn->agreed_pro_version >= 100)
1956 /* Do this without trying to take tconn->data.mutex again. */
1957 __drbd_send_protocol(tconn, P_PROTOCOL_UPDATE);
1959 crypto_free_hash(tconn->cram_hmac_tfm);
1960 tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
1962 mutex_unlock(&tconn->conf_update);
1963 mutex_unlock(&tconn->data.mutex);
1967 if (tconn->cstate >= C_WF_REPORT_PARAMS)
1968 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1973 mutex_unlock(&tconn->conf_update);
1974 mutex_unlock(&tconn->data.mutex);
1975 free_crypto(&crypto);
1978 conn_reconfig_done(tconn);
1980 drbd_adm_finish(info, retcode);
1984 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1986 struct drbd_conf *mdev;
1987 struct net_conf *old_conf, *new_conf = NULL;
1988 struct crypto crypto = { };
1989 struct drbd_tconn *oconn;
1990 struct drbd_tconn *tconn;
1991 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1992 enum drbd_ret_code retcode;
1996 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1997 if (!adm_ctx.reply_skb)
1999 if (retcode != NO_ERROR)
2002 tconn = adm_ctx.tconn;
2003 conn_reconfig_start(tconn);
2005 if (tconn->cstate > C_STANDALONE) {
2006 retcode = ERR_NET_CONFIGURED;
2010 /* allocation not in the IO path, cqueue thread context */
2011 new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
2013 retcode = ERR_NOMEM;
2017 set_net_conf_defaults(new_conf);
2019 err = net_conf_from_attrs(new_conf, info);
2021 retcode = ERR_MANDATORY_TAG;
2022 drbd_msg_put_info(from_attrs_err_to_txt(err));
2026 retcode = check_net_options(tconn, new_conf);
2027 if (retcode != NO_ERROR)
2032 new_my_addr = (struct sockaddr *)&new_conf->my_addr;
2033 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
2035 /* No need for _rcu here. All reconfiguration is
2036 * strictly serialized on genl_lock(). We are protected against
2037 * concurrent reconfiguration/addition/deletion */
2038 list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
2039 struct net_conf *nc;
2044 nc = rcu_dereference(oconn->net_conf);
2046 taken_addr = (struct sockaddr *)&nc->my_addr;
2047 if (new_conf->my_addr_len == nc->my_addr_len &&
2048 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
2049 retcode = ERR_LOCAL_ADDR;
2051 taken_addr = (struct sockaddr *)&nc->peer_addr;
2052 if (new_conf->peer_addr_len == nc->peer_addr_len &&
2053 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
2054 retcode = ERR_PEER_ADDR;
2057 if (retcode != NO_ERROR)
2061 retcode = alloc_crypto(&crypto, new_conf);
2062 if (retcode != NO_ERROR)
2065 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2067 conn_flush_workqueue(tconn);
2069 mutex_lock(&tconn->conf_update);
2070 old_conf = tconn->net_conf;
2072 retcode = ERR_NET_CONFIGURED;
2073 mutex_unlock(&tconn->conf_update);
2076 rcu_assign_pointer(tconn->net_conf, new_conf);
2078 conn_free_crypto(tconn);
2079 tconn->int_dig_in = crypto.int_dig_in;
2080 tconn->int_dig_vv = crypto.int_dig_vv;
2081 tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2082 tconn->integrity_tfm = crypto.integrity_tfm;
2083 tconn->csums_tfm = crypto.csums_tfm;
2084 tconn->verify_tfm = crypto.verify_tfm;
2086 mutex_unlock(&tconn->conf_update);
2089 idr_for_each_entry(&tconn->volumes, mdev, i) {
2095 retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2097 conn_reconfig_done(tconn);
2098 drbd_adm_finish(info, retcode);
2102 free_crypto(&crypto);
2105 conn_reconfig_done(tconn);
2107 drbd_adm_finish(info, retcode);
2111 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2113 enum drbd_state_rv rv;
2115 rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2116 force ? CS_HARD : 0);
2119 case SS_NOTHING_TO_DO:
2121 case SS_ALREADY_STANDALONE:
2123 case SS_PRIMARY_NOP:
2124 /* Our state checking code wants to see the peer outdated. */
2125 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2126 pdsk, D_OUTDATED), CS_VERBOSE);
2128 case SS_CW_FAILED_BY_PEER:
2129 /* The peer probably wants to see us outdated. */
2130 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2131 disk, D_OUTDATED), 0);
2132 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2133 rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2138 /* no special handling necessary */
2141 if (rv >= SS_SUCCESS) {
2142 enum drbd_state_rv rv2;
2143 /* No one else can reconfigure the network while I am here.
2144 * The state handling only uses drbd_thread_stop_nowait(),
2145 * we want to really wait here until the receiver is no more.
2147 drbd_thread_stop(&adm_ctx.tconn->receiver);
2149 /* Race breaker. This additional state change request may be
2150 * necessary, if this was a forced disconnect during a receiver
2151 * restart. We may have "killed" the receiver thread just
2152 * after drbdd_init() returned. Typically, we should be
2153 * C_STANDALONE already, now, and this becomes a no-op.
2155 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2156 CS_VERBOSE | CS_HARD);
2157 if (rv2 < SS_SUCCESS)
2159 "unexpected rv2=%d in conn_try_disconnect()\n",
2165 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2167 struct disconnect_parms parms;
2168 struct drbd_tconn *tconn;
2169 enum drbd_state_rv rv;
2170 enum drbd_ret_code retcode;
2173 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2174 if (!adm_ctx.reply_skb)
2176 if (retcode != NO_ERROR)
2179 tconn = adm_ctx.tconn;
2180 memset(&parms, 0, sizeof(parms));
2181 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2182 err = disconnect_parms_from_attrs(&parms, info);
2184 retcode = ERR_MANDATORY_TAG;
2185 drbd_msg_put_info(from_attrs_err_to_txt(err));
2190 rv = conn_try_disconnect(tconn, parms.force_disconnect);
2191 if (rv < SS_SUCCESS)
2192 retcode = rv; /* FIXME: Type mismatch. */
2196 drbd_adm_finish(info, retcode);
2200 void resync_after_online_grow(struct drbd_conf *mdev)
2202 int iass; /* I am sync source */
2204 dev_info(DEV, "Resync of new storage after online grow\n");
2205 if (mdev->state.role != mdev->state.peer)
2206 iass = (mdev->state.role == R_PRIMARY);
2208 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2211 drbd_start_resync(mdev, C_SYNC_SOURCE);
2213 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2216 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2218 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2219 struct resize_parms rs;
2220 struct drbd_conf *mdev;
2221 enum drbd_ret_code retcode;
2222 enum determine_dev_size dd;
2223 enum dds_flags ddsf;
2227 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2228 if (!adm_ctx.reply_skb)
2230 if (retcode != NO_ERROR)
2233 memset(&rs, 0, sizeof(struct resize_parms));
2234 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2235 err = resize_parms_from_attrs(&rs, info);
2237 retcode = ERR_MANDATORY_TAG;
2238 drbd_msg_put_info(from_attrs_err_to_txt(err));
2243 mdev = adm_ctx.mdev;
2244 if (mdev->state.conn > C_CONNECTED) {
2245 retcode = ERR_RESIZE_RESYNC;
2249 if (mdev->state.role == R_SECONDARY &&
2250 mdev->state.peer == R_SECONDARY) {
2251 retcode = ERR_NO_PRIMARY;
2255 if (!get_ldev(mdev)) {
2256 retcode = ERR_NO_DISK;
2260 if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2261 retcode = ERR_NEED_APV_93;
2266 u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2268 if (u_size != (sector_t)rs.resize_size) {
2269 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2270 if (!new_disk_conf) {
2271 retcode = ERR_NOMEM;
2276 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2277 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2279 if (new_disk_conf) {
2280 mutex_lock(&mdev->tconn->conf_update);
2281 old_disk_conf = mdev->ldev->disk_conf;
2282 *new_disk_conf = *old_disk_conf;
2283 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2284 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2285 mutex_unlock(&mdev->tconn->conf_update);
2287 kfree(old_disk_conf);
2290 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2291 dd = drbd_determine_dev_size(mdev, ddsf);
2294 if (dd == dev_size_error) {
2295 retcode = ERR_NOMEM_BITMAP;
2299 if (mdev->state.conn == C_CONNECTED) {
2301 set_bit(RESIZE_PENDING, &mdev->flags);
2303 drbd_send_uuids(mdev);
2304 drbd_send_sizes(mdev, 1, ddsf);
2308 drbd_adm_finish(info, retcode);
2312 void drbd_set_res_opts_defaults(struct res_opts *r)
2314 return set_res_opts_defaults(r);
2317 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2319 enum drbd_ret_code retcode;
2320 cpumask_var_t new_cpu_mask;
2321 struct drbd_tconn *tconn;
2322 struct res_opts res_opts;
2325 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2326 if (!adm_ctx.reply_skb)
2328 if (retcode != NO_ERROR)
2330 tconn = adm_ctx.tconn;
2332 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
2333 retcode = ERR_NOMEM;
2334 drbd_msg_put_info("unable to allocate cpumask");
2338 res_opts = tconn->res_opts;
2339 if (should_set_defaults(info))
2340 set_res_opts_defaults(&res_opts);
2342 err = res_opts_from_attrs(&res_opts, info);
2343 if (err && err != -ENOMSG) {
2344 retcode = ERR_MANDATORY_TAG;
2345 drbd_msg_put_info(from_attrs_err_to_txt(err));
2349 /* silently ignore cpu mask on UP kernel */
2350 if (nr_cpu_ids > 1 && res_opts.cpu_mask[0] != 0) {
2351 err = __bitmap_parse(res_opts.cpu_mask, 32, 0,
2352 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2354 conn_warn(tconn, "__bitmap_parse() failed with %d\n", err);
2355 retcode = ERR_CPU_MASK_PARSE;
2361 tconn->res_opts = res_opts;
2363 if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2364 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2365 drbd_calc_cpu_mask(tconn);
2366 tconn->receiver.reset_cpu_mask = 1;
2367 tconn->asender.reset_cpu_mask = 1;
2368 tconn->worker.reset_cpu_mask = 1;
2372 free_cpumask_var(new_cpu_mask);
2374 drbd_adm_finish(info, retcode);
2378 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2380 struct drbd_conf *mdev;
2381 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2383 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2384 if (!adm_ctx.reply_skb)
2386 if (retcode != NO_ERROR)
2389 mdev = adm_ctx.mdev;
2391 /* If there is still bitmap IO pending, probably because of a previous
2392 * resync just being finished, wait for it before requesting a new resync. */
2393 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2395 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2397 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2398 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2400 while (retcode == SS_NEED_CONNECTION) {
2401 spin_lock_irq(&mdev->tconn->req_lock);
2402 if (mdev->state.conn < C_CONNECTED)
2403 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2404 spin_unlock_irq(&mdev->tconn->req_lock);
2406 if (retcode != SS_NEED_CONNECTION)
2409 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2413 drbd_adm_finish(info, retcode);
2417 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2421 rv = drbd_bmio_set_n_write(mdev);
2422 drbd_suspend_al(mdev);
2426 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2427 union drbd_state mask, union drbd_state val)
2429 enum drbd_ret_code retcode;
2431 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2432 if (!adm_ctx.reply_skb)
2434 if (retcode != NO_ERROR)
2437 retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2439 drbd_adm_finish(info, retcode);
2443 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2445 return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2448 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2450 enum drbd_ret_code retcode;
2452 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2453 if (!adm_ctx.reply_skb)
2455 if (retcode != NO_ERROR)
2458 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2459 retcode = ERR_PAUSE_IS_SET;
2461 drbd_adm_finish(info, retcode);
2465 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2467 union drbd_dev_state s;
2468 enum drbd_ret_code retcode;
2470 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2471 if (!adm_ctx.reply_skb)
2473 if (retcode != NO_ERROR)
2476 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2477 s = adm_ctx.mdev->state;
2478 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2479 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2480 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2482 retcode = ERR_PAUSE_IS_CLEAR;
2487 drbd_adm_finish(info, retcode);
2491 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2493 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2496 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2498 struct drbd_conf *mdev;
2499 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2501 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2502 if (!adm_ctx.reply_skb)
2504 if (retcode != NO_ERROR)
2507 mdev = adm_ctx.mdev;
2508 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2509 drbd_uuid_new_current(mdev);
2510 clear_bit(NEW_CUR_UUID, &mdev->flags);
2512 drbd_suspend_io(mdev);
2513 retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2514 if (retcode == SS_SUCCESS) {
2515 if (mdev->state.conn < C_CONNECTED)
2516 tl_clear(mdev->tconn);
2517 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2518 tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2520 drbd_resume_io(mdev);
2523 drbd_adm_finish(info, retcode);
2527 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2529 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2532 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *resource_name, unsigned vnr)
2535 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2537 goto nla_put_failure;
2538 if (vnr != VOLUME_UNSPECIFIED)
2539 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2540 NLA_PUT_STRING(skb, T_ctx_resource_name, resource_name);
2541 nla_nest_end(skb, nla);
2546 nla_nest_cancel(skb, nla);
2550 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2551 const struct sib_info *sib)
2553 struct state_info *si = NULL; /* for sizeof(si->member); */
2554 struct net_conf *nc;
2558 int exclude_sensitive;
2560 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2561 * to. So we better exclude_sensitive information.
2563 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2564 * in the context of the requesting user process. Exclude sensitive
2565 * information, unless current has superuser.
2567 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2568 * relies on the current implementation of netlink_dump(), which
2569 * executes the dump callback successively from netlink_recvmsg(),
2570 * always in the context of the receiving process */
2571 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2573 got_ldev = get_ldev(mdev);
2575 /* We need to add connection name and volume number information still.
2576 * Minor number is in drbd_genlmsghdr. */
2577 if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2578 goto nla_put_failure;
2580 if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2581 goto nla_put_failure;
2585 if (disk_conf_to_skb(skb, rcu_dereference(mdev->ldev->disk_conf), exclude_sensitive))
2586 goto nla_put_failure;
2588 nc = rcu_dereference(mdev->tconn->net_conf);
2590 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2593 goto nla_put_failure;
2595 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2597 goto nla_put_failure;
2598 NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2599 NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2600 NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2601 NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2604 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2605 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2606 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2607 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2608 if (C_SYNC_SOURCE <= mdev->state.conn &&
2609 C_PAUSED_SYNC_T >= mdev->state.conn) {
2610 NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2611 NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2616 switch(sib->sib_reason) {
2617 case SIB_SYNC_PROGRESS:
2618 case SIB_GET_STATUS_REPLY:
2620 case SIB_STATE_CHANGE:
2621 NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2622 NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2624 case SIB_HELPER_POST:
2626 T_helper_exit_code, sib->helper_exit_code);
2628 case SIB_HELPER_PRE:
2629 NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2633 nla_nest_end(skb, nla);
2643 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2645 enum drbd_ret_code retcode;
2648 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2649 if (!adm_ctx.reply_skb)
2651 if (retcode != NO_ERROR)
2654 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2656 nlmsg_free(adm_ctx.reply_skb);
2660 drbd_adm_finish(info, retcode);
2664 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2666 struct drbd_conf *mdev;
2667 struct drbd_genlmsghdr *dh;
2668 struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2669 struct drbd_tconn *tconn = NULL;
2670 struct drbd_tconn *tmp;
2671 unsigned volume = cb->args[1];
2673 /* Open coded, deferred, iteration:
2674 * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2675 * idr_for_each_entry(&tconn->volumes, mdev, i) {
2679 * where tconn is cb->args[0];
2680 * and i is cb->args[1];
2682 * cb->args[2] indicates if we shall loop over all resources,
2683 * or just dump all volumes of a single resource.
2685 * This may miss entries inserted after this dump started,
2686 * or entries deleted before they are reached.
2688 * We need to make sure the mdev won't disappear while
2689 * we are looking at it, and revalidate our iterators
2690 * on each iteration.
2693 /* synchronize with conn_create()/conn_destroy() */
2695 /* revalidate iterator position */
2696 list_for_each_entry_rcu(tmp, &drbd_tconns, all_tconn) {
2698 /* first iteration */
2710 mdev = idr_get_next(&tconn->volumes, &volume);
2712 /* No more volumes to dump on this tconn.
2713 * Advance tconn iterator. */
2714 pos = list_entry_rcu(tconn->all_tconn.next,
2715 struct drbd_tconn, all_tconn);
2716 /* Did we dump any volume on this tconn yet? */
2718 /* If we reached the end of the list,
2719 * or only a single resource dump was requested,
2721 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2729 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2730 cb->nlh->nlmsg_seq, &drbd_genl_family,
2731 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2736 /* this is a tconn without a single volume */
2738 dh->ret_code = NO_ERROR;
2739 if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2740 genlmsg_cancel(skb, dh);
2742 genlmsg_end(skb, dh);
2746 D_ASSERT(mdev->vnr == volume);
2747 D_ASSERT(mdev->tconn == tconn);
2749 dh->minor = mdev_to_minor(mdev);
2750 dh->ret_code = NO_ERROR;
2752 if (nla_put_status_info(skb, mdev, NULL)) {
2753 genlmsg_cancel(skb, dh);
2756 genlmsg_end(skb, dh);
2761 /* where to start the next iteration */
2762 cb->args[0] = (long)pos;
2763 cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2765 /* No more tconns/volumes/minors found results in an empty skb.
2766 * Which will terminate the dump. */
2771 * Request status of all resources, or of all volumes within a single resource.
2773 * This is a dump, as the answer may not fit in a single reply skb otherwise.
2774 * Which means we cannot use the family->attrbuf or other such members, because
2775 * dump is NOT protected by the genl_lock(). During dump, we only have access
2776 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2778 * Once things are setup properly, we call into get_one_status().
2780 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2782 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2784 const char *resource_name;
2785 struct drbd_tconn *tconn;
2788 /* Is this a followup call? */
2790 /* ... of a single resource dump,
2791 * and the resource iterator has been advanced already? */
2792 if (cb->args[2] && cb->args[2] != cb->args[0])
2793 return 0; /* DONE. */
2797 /* First call (from netlink_dump_start). We need to figure out
2798 * which resource(s) the user wants us to dump. */
2799 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2800 nlmsg_attrlen(cb->nlh, hdrlen),
2801 DRBD_NLA_CFG_CONTEXT);
2803 /* No explicit context given. Dump all. */
2806 maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
2807 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
2809 return PTR_ERR(nla);
2810 /* context given, but no name present? */
2813 resource_name = nla_data(nla);
2814 tconn = conn_get_by_name(resource_name);
2819 kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2821 /* prime iterators, and set "filter" mode mark:
2822 * only dump this tconn. */
2823 cb->args[0] = (long)tconn;
2824 /* cb->args[1] = 0; passed in this way. */
2825 cb->args[2] = (long)tconn;
2828 return get_one_status(skb, cb);
2831 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2833 enum drbd_ret_code retcode;
2834 struct timeout_parms tp;
2837 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2838 if (!adm_ctx.reply_skb)
2840 if (retcode != NO_ERROR)
2844 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2845 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2848 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2850 nlmsg_free(adm_ctx.reply_skb);
2854 drbd_adm_finish(info, retcode);
2858 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2860 struct drbd_conf *mdev;
2861 enum drbd_ret_code retcode;
2863 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2864 if (!adm_ctx.reply_skb)
2866 if (retcode != NO_ERROR)
2869 mdev = adm_ctx.mdev;
2870 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2871 /* resume from last known position, if possible */
2872 struct start_ov_parms parms =
2873 { .ov_start_sector = mdev->ov_start_sector };
2874 int err = start_ov_parms_from_attrs(&parms, info);
2876 retcode = ERR_MANDATORY_TAG;
2877 drbd_msg_put_info(from_attrs_err_to_txt(err));
2880 /* w_make_ov_request expects position to be aligned */
2881 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2883 /* If there is still bitmap IO pending, e.g. previous resync or verify
2884 * just being finished, wait for it before requesting a new resync. */
2885 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2886 retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2888 drbd_adm_finish(info, retcode);
2893 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2895 struct drbd_conf *mdev;
2896 enum drbd_ret_code retcode;
2897 int skip_initial_sync = 0;
2899 struct new_c_uuid_parms args;
2901 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2902 if (!adm_ctx.reply_skb)
2904 if (retcode != NO_ERROR)
2907 mdev = adm_ctx.mdev;
2908 memset(&args, 0, sizeof(args));
2909 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2910 err = new_c_uuid_parms_from_attrs(&args, info);
2912 retcode = ERR_MANDATORY_TAG;
2913 drbd_msg_put_info(from_attrs_err_to_txt(err));
2918 mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2920 if (!get_ldev(mdev)) {
2921 retcode = ERR_NO_DISK;
2925 /* this is "skip initial sync", assume to be clean */
2926 if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2927 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2928 dev_info(DEV, "Preparing to skip initial sync\n");
2929 skip_initial_sync = 1;
2930 } else if (mdev->state.conn != C_STANDALONE) {
2931 retcode = ERR_CONNECTED;
2935 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2936 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2938 if (args.clear_bm) {
2939 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2940 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2942 dev_err(DEV, "Writing bitmap failed with %d\n",err);
2943 retcode = ERR_IO_MD_DISK;
2945 if (skip_initial_sync) {
2946 drbd_send_uuids_skip_initial_sync(mdev);
2947 _drbd_uuid_set(mdev, UI_BITMAP, 0);
2948 drbd_print_uuids(mdev, "cleared bitmap UUID");
2949 spin_lock_irq(&mdev->tconn->req_lock);
2950 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2952 spin_unlock_irq(&mdev->tconn->req_lock);
2960 mutex_unlock(mdev->state_mutex);
2962 drbd_adm_finish(info, retcode);
2966 static enum drbd_ret_code
2967 drbd_check_resource_name(const char *name)
2969 if (!name || !name[0]) {
2970 drbd_msg_put_info("resource name missing");
2971 return ERR_MANDATORY_TAG;
2973 /* if we want to use these in sysfs/configfs/debugfs some day,
2974 * we must not allow slashes */
2975 if (strchr(name, '/')) {
2976 drbd_msg_put_info("invalid resource name");
2977 return ERR_INVALID_REQUEST;
2982 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
2984 enum drbd_ret_code retcode;
2986 retcode = drbd_adm_prepare(skb, info, 0);
2987 if (!adm_ctx.reply_skb)
2989 if (retcode != NO_ERROR)
2992 retcode = drbd_check_resource_name(adm_ctx.resource_name);
2993 if (retcode != NO_ERROR)
2996 if (adm_ctx.tconn) {
2997 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
2998 retcode = ERR_INVALID_REQUEST;
2999 drbd_msg_put_info("resource exists");
3001 /* else: still NO_ERROR */
3005 if (!conn_create(adm_ctx.resource_name))
3006 retcode = ERR_NOMEM;
3008 drbd_adm_finish(info, retcode);
3012 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
3014 struct drbd_genlmsghdr *dh = info->userhdr;
3015 enum drbd_ret_code retcode;
3017 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3018 if (!adm_ctx.reply_skb)
3020 if (retcode != NO_ERROR)
3023 /* FIXME drop minor_count parameter, limit to MINORMASK */
3024 if (dh->minor >= minor_count) {
3025 drbd_msg_put_info("requested minor out of range");
3026 retcode = ERR_INVALID_REQUEST;
3029 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3030 drbd_msg_put_info("requested volume id out of range");
3031 retcode = ERR_INVALID_REQUEST;
3035 /* drbd_adm_prepare made sure already
3036 * that mdev->tconn and mdev->vnr match the request. */
3038 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3039 retcode = ERR_MINOR_EXISTS;
3040 /* else: still NO_ERROR */
3044 retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3046 drbd_adm_finish(info, retcode);
3050 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3052 if (mdev->state.disk == D_DISKLESS &&
3053 /* no need to be mdev->state.conn == C_STANDALONE &&
3054 * we may want to delete a minor from a live replication group.
3056 mdev->state.role == R_SECONDARY) {
3057 idr_remove(&mdev->tconn->volumes, mdev->vnr);
3058 idr_remove(&minors, mdev_to_minor(mdev));
3059 del_gendisk(mdev->vdisk);
3061 kref_put(&mdev->kref, &drbd_minor_destroy);
3064 return ERR_MINOR_CONFIGURED;
3067 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3069 enum drbd_ret_code retcode;
3071 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3072 if (!adm_ctx.reply_skb)
3074 if (retcode != NO_ERROR)
3077 retcode = adm_delete_minor(adm_ctx.mdev);
3079 drbd_adm_finish(info, retcode);
3083 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3085 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3086 struct drbd_conf *mdev;
3089 retcode = drbd_adm_prepare(skb, info, 0);
3090 if (!adm_ctx.reply_skb)
3092 if (retcode != NO_ERROR)
3095 if (!adm_ctx.tconn) {
3096 retcode = ERR_RES_NOT_KNOWN;
3101 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3102 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3103 if (retcode < SS_SUCCESS) {
3104 drbd_msg_put_info("failed to demote");
3109 retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3110 if (retcode < SS_SUCCESS) {
3111 drbd_msg_put_info("failed to disconnect");
3116 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3117 retcode = adm_detach(mdev);
3118 if (retcode < SS_SUCCESS) {
3119 drbd_msg_put_info("failed to detach");
3124 /* If we reach this, all volumes (of this tconn) are Secondary,
3125 * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3126 * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3127 drbd_thread_stop(&adm_ctx.tconn->worker);
3129 /* Now, nothing can fail anymore */
3131 /* delete volumes */
3132 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3133 retcode = adm_delete_minor(mdev);
3134 if (retcode != NO_ERROR) {
3135 /* "can not happen" */
3136 drbd_msg_put_info("failed to delete volume");
3141 /* delete connection */
3142 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3143 list_del_rcu(&adm_ctx.tconn->all_tconn);
3145 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3149 /* "can not happen" */
3150 retcode = ERR_RES_IN_USE;
3151 drbd_msg_put_info("failed to delete connection");
3155 drbd_adm_finish(info, retcode);
3159 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
3161 enum drbd_ret_code retcode;
3163 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3164 if (!adm_ctx.reply_skb)
3166 if (retcode != NO_ERROR)
3169 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3170 list_del_rcu(&adm_ctx.tconn->all_tconn);
3172 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3176 retcode = ERR_RES_IN_USE;
3179 if (retcode == NO_ERROR)
3180 drbd_thread_stop(&adm_ctx.tconn->worker);
3182 drbd_adm_finish(info, retcode);
3186 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3188 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3189 struct sk_buff *msg;
3190 struct drbd_genlmsghdr *d_out;
3194 seq = atomic_inc_return(&drbd_genl_seq);
3195 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3200 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3201 if (!d_out) /* cannot happen, but anyways. */
3202 goto nla_put_failure;
3203 d_out->minor = mdev_to_minor(mdev);
3204 d_out->ret_code = NO_ERROR;
3206 if (nla_put_status_info(msg, mdev, sib))
3207 goto nla_put_failure;
3208 genlmsg_end(msg, d_out);
3209 err = drbd_genl_multicast_events(msg, 0);
3210 /* msg has been consumed or freed in netlink_broadcast() */
3211 if (err && err != -ESRCH)
3219 dev_err(DEV, "Error %d while broadcasting event. "
3220 "Event seq:%u sib_reason:%u\n",
3221 err, seq, sib->sib_reason);