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>
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) {
164 adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
165 info, &drbd_genl_family, 0, cmd);
166 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
168 if (!adm_ctx.reply_dh) {
173 adm_ctx.reply_dh->minor = d_in->minor;
174 adm_ctx.reply_dh->ret_code = NO_ERROR;
176 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
178 /* parse and validate only */
179 err = drbd_cfg_context_from_attrs(NULL, info);
183 /* It was present, and valid,
184 * copy it over to the reply skb. */
185 err = nla_put_nohdr(adm_ctx.reply_skb,
186 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
187 info->attrs[DRBD_NLA_CFG_CONTEXT]);
191 /* and assign stuff to the global adm_ctx */
192 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
193 adm_ctx.volume = nla ? nla_get_u32(nla) : VOLUME_UNSPECIFIED;
194 nla = nested_attr_tb[__nla_type(T_ctx_conn_name)];
196 adm_ctx.conn_name = nla_data(nla);
198 adm_ctx.volume = VOLUME_UNSPECIFIED;
200 adm_ctx.minor = d_in->minor;
201 adm_ctx.mdev = minor_to_mdev(d_in->minor);
202 adm_ctx.tconn = conn_get_by_name(adm_ctx.conn_name);
204 if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
205 drbd_msg_put_info("unknown minor");
206 return ERR_MINOR_INVALID;
208 if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_CONN)) {
209 drbd_msg_put_info("unknown connection");
210 return ERR_INVALID_REQUEST;
213 /* some more paranoia, if the request was over-determined */
214 if (adm_ctx.mdev && adm_ctx.tconn &&
215 adm_ctx.mdev->tconn != adm_ctx.tconn) {
216 pr_warning("request: minor=%u, conn=%s; but that minor belongs to connection %s\n",
217 adm_ctx.minor, adm_ctx.conn_name, adm_ctx.mdev->tconn->name);
218 drbd_msg_put_info("minor exists in different connection");
219 return ERR_INVALID_REQUEST;
222 adm_ctx.volume != VOLUME_UNSPECIFIED &&
223 adm_ctx.volume != adm_ctx.mdev->vnr) {
224 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
225 adm_ctx.minor, adm_ctx.volume,
226 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
227 drbd_msg_put_info("minor exists as different volume");
228 return ERR_INVALID_REQUEST;
234 nlmsg_free(adm_ctx.reply_skb);
235 adm_ctx.reply_skb = NULL;
239 static int drbd_adm_finish(struct genl_info *info, int retcode)
242 const char *conn_name = NULL;
245 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
246 adm_ctx.tconn = NULL;
249 if (!adm_ctx.reply_skb)
252 adm_ctx.reply_dh->ret_code = retcode;
254 nla = info->attrs[DRBD_NLA_CFG_CONTEXT];
256 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
258 conn_name = nla_data(nla);
261 drbd_adm_send_reply(adm_ctx.reply_skb, info);
265 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
271 nc = rcu_dereference(tconn->net_conf);
273 switch (((struct sockaddr *)nc->peer_addr)->sa_family) {
276 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
277 &((struct sockaddr_in6 *)nc->peer_addr)->sin6_addr);
281 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
282 &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
286 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
287 &((struct sockaddr_in *)nc->peer_addr)->sin_addr);
289 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
294 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
296 char *envp[] = { "HOME=/",
298 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
299 (char[20]) { }, /* address family */
300 (char[60]) { }, /* address */
303 char *argv[] = {usermode_helper, cmd, mb, NULL };
307 snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
308 setup_khelper_env(mdev->tconn, envp);
310 /* The helper may take some time.
311 * write out any unsynced meta data changes now */
314 dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
315 sib.sib_reason = SIB_HELPER_PRE;
316 sib.helper_name = cmd;
317 drbd_bcast_event(mdev, &sib);
318 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
320 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
321 usermode_helper, cmd, mb,
322 (ret >> 8) & 0xff, ret);
324 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
325 usermode_helper, cmd, mb,
326 (ret >> 8) & 0xff, ret);
327 sib.sib_reason = SIB_HELPER_POST;
328 sib.helper_exit_code = ret;
329 drbd_bcast_event(mdev, &sib);
331 if (ret < 0) /* Ignore any ERRNOs we got. */
337 static void conn_md_sync(struct drbd_tconn *tconn)
339 struct drbd_conf *mdev;
343 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
344 kref_get(&mdev->kref);
347 kref_put(&mdev->kref, &drbd_minor_destroy);
353 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
355 char *envp[] = { "HOME=/",
357 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
358 (char[20]) { }, /* address family */
359 (char[60]) { }, /* address */
361 char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
364 setup_khelper_env(tconn, envp);
367 conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
368 /* TODO: conn_bcast_event() ?? */
370 ret = call_usermodehelper(usermode_helper, argv, envp, 1);
372 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
373 usermode_helper, cmd, tconn->name,
374 (ret >> 8) & 0xff, ret);
376 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
377 usermode_helper, cmd, tconn->name,
378 (ret >> 8) & 0xff, ret);
379 /* TODO: conn_bcast_event() ?? */
381 if (ret < 0) /* Ignore any ERRNOs we got. */
387 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
389 enum drbd_fencing_p fp = FP_NOT_AVAIL;
390 struct drbd_conf *mdev;
394 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
395 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
396 fp = max_t(enum drbd_fencing_p, fp,
397 rcu_dereference(mdev->ldev->disk_conf)->fencing);
406 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
408 union drbd_state mask = { };
409 union drbd_state val = { };
410 enum drbd_fencing_p fp;
414 if (tconn->cstate >= C_WF_REPORT_PARAMS) {
415 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
419 fp = highest_fencing_policy(tconn);
422 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
429 r = conn_khelper(tconn, "fence-peer");
431 switch ((r>>8) & 0xff) {
432 case 3: /* peer is inconsistent */
433 ex_to_string = "peer is inconsistent or worse";
435 val.pdsk = D_INCONSISTENT;
437 case 4: /* peer got outdated, or was already outdated */
438 ex_to_string = "peer was fenced";
440 val.pdsk = D_OUTDATED;
442 case 5: /* peer was down */
443 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
444 /* we will(have) create(d) a new UUID anyways... */
445 ex_to_string = "peer is unreachable, assumed to be dead";
447 val.pdsk = D_OUTDATED;
449 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
452 case 6: /* Peer is primary, voluntarily outdate myself.
453 * This is useful when an unconnected R_SECONDARY is asked to
454 * become R_PRIMARY, but finds the other peer being active. */
455 ex_to_string = "peer is active";
456 conn_warn(tconn, "Peer is primary, outdating myself.\n");
458 val.disk = D_OUTDATED;
461 if (fp != FP_STONITH)
462 conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
463 ex_to_string = "peer was stonithed";
465 val.pdsk = D_OUTDATED;
468 /* The script is broken ... */
469 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
470 return false; /* Eventually leave IO frozen */
473 conn_info(tconn, "fence-peer helper returned %d (%s)\n",
474 (r>>8) & 0xff, ex_to_string);
479 conn_request_state(tconn, mask, val, CS_VERBOSE);
480 here, because we might were able to re-establish the connection in the
482 spin_lock_irq(&tconn->req_lock);
483 if (tconn->cstate < C_WF_REPORT_PARAMS)
484 _conn_request_state(tconn, mask, val, CS_VERBOSE);
485 spin_unlock_irq(&tconn->req_lock);
487 return conn_highest_pdsk(tconn) <= D_OUTDATED;
490 static int _try_outdate_peer_async(void *data)
492 struct drbd_tconn *tconn = (struct drbd_tconn *)data;
494 conn_try_outdate_peer(tconn);
496 kref_put(&tconn->kref, &conn_destroy);
500 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
502 struct task_struct *opa;
504 kref_get(&tconn->kref);
505 opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
507 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
508 kref_put(&tconn->kref, &conn_destroy);
513 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
515 const int max_tries = 4;
516 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
520 union drbd_state mask, val;
522 if (new_role == R_PRIMARY)
523 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
525 mutex_lock(mdev->state_mutex);
527 mask.i = 0; mask.role = R_MASK;
528 val.i = 0; val.role = new_role;
530 while (try++ < max_tries) {
531 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
533 /* in case we first succeeded to outdate,
534 * but now suddenly could establish a connection */
535 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
541 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
542 (mdev->state.disk < D_UP_TO_DATE &&
543 mdev->state.disk >= D_INCONSISTENT)) {
545 val.disk = D_UP_TO_DATE;
550 if (rv == SS_NO_UP_TO_DATE_DISK &&
551 mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
552 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
554 if (conn_try_outdate_peer(mdev->tconn)) {
555 val.disk = D_UP_TO_DATE;
561 if (rv == SS_NOTHING_TO_DO)
563 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
564 if (!conn_try_outdate_peer(mdev->tconn) && force) {
565 dev_warn(DEV, "Forced into split brain situation!\n");
567 val.pdsk = D_OUTDATED;
572 if (rv == SS_TWO_PRIMARIES) {
573 /* Maybe the peer is detected as dead very soon...
574 retry at most once more in this case. */
577 nc = rcu_dereference(mdev->tconn->net_conf);
578 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
580 schedule_timeout_interruptible(timeo);
585 if (rv < SS_SUCCESS) {
586 rv = _drbd_request_state(mdev, mask, val,
587 CS_VERBOSE + CS_WAIT_COMPLETE);
598 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
600 /* Wait until nothing is on the fly :) */
601 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
603 if (new_role == R_SECONDARY) {
604 set_disk_ro(mdev->vdisk, true);
605 if (get_ldev(mdev)) {
606 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
610 mutex_lock(&mdev->tconn->conf_update);
611 nc = mdev->tconn->net_conf;
613 nc->discard_my_data = 0; /* without copy; single bit op is atomic */
614 mutex_unlock(&mdev->tconn->conf_update);
616 set_disk_ro(mdev->vdisk, false);
617 if (get_ldev(mdev)) {
618 if (((mdev->state.conn < C_CONNECTED ||
619 mdev->state.pdsk <= D_FAILED)
620 && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
621 drbd_uuid_new_current(mdev);
623 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
628 /* writeout of activity log covered areas of the bitmap
629 * to stable storage done in after state change already */
631 if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
632 /* if this was forced, we should consider sync */
634 drbd_send_uuids(mdev);
635 drbd_send_state(mdev);
640 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
642 mutex_unlock(mdev->state_mutex);
646 static const char *from_attrs_err_to_txt(int err)
648 return err == -ENOMSG ? "required attribute missing" :
649 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
650 err == -EEXIST ? "can not change invariant setting" :
651 "invalid attribute value";
654 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
656 struct set_role_parms parms;
658 enum drbd_ret_code retcode;
660 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
661 if (!adm_ctx.reply_skb)
663 if (retcode != NO_ERROR)
666 memset(&parms, 0, sizeof(parms));
667 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
668 err = set_role_parms_from_attrs(&parms, info);
670 retcode = ERR_MANDATORY_TAG;
671 drbd_msg_put_info(from_attrs_err_to_txt(err));
676 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
677 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
679 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
681 drbd_adm_finish(info, retcode);
685 /* initializes the md.*_offset members, so we are able to find
686 * the on disk meta data */
687 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
688 struct drbd_backing_dev *bdev)
690 sector_t md_size_sect = 0;
694 meta_dev_idx = rcu_dereference(bdev->disk_conf)->meta_dev_idx;
696 switch (meta_dev_idx) {
698 /* v07 style fixed size indexed meta data */
699 bdev->md.md_size_sect = MD_RESERVED_SECT;
700 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
701 bdev->md.al_offset = MD_AL_OFFSET;
702 bdev->md.bm_offset = MD_BM_OFFSET;
704 case DRBD_MD_INDEX_FLEX_EXT:
705 /* just occupy the full device; unit: sectors */
706 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
707 bdev->md.md_offset = 0;
708 bdev->md.al_offset = MD_AL_OFFSET;
709 bdev->md.bm_offset = MD_BM_OFFSET;
711 case DRBD_MD_INDEX_INTERNAL:
712 case DRBD_MD_INDEX_FLEX_INT:
713 bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
714 /* al size is still fixed */
715 bdev->md.al_offset = -MD_AL_SECTORS;
716 /* we need (slightly less than) ~ this much bitmap sectors: */
717 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
718 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
719 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
720 md_size_sect = ALIGN(md_size_sect, 8);
722 /* plus the "drbd meta data super block",
723 * and the activity log; */
724 md_size_sect += MD_BM_OFFSET;
726 bdev->md.md_size_sect = md_size_sect;
727 /* bitmap offset is adjusted by 'super' block size */
728 bdev->md.bm_offset = -md_size_sect + MD_AL_OFFSET;
734 /* input size is expected to be in KB */
735 char *ppsize(char *buf, unsigned long long size)
737 /* Needs 9 bytes at max including trailing NUL:
738 * -1ULL ==> "16384 EB" */
739 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
741 while (size >= 10000 && base < sizeof(units)-1) {
743 size = (size >> 10) + !!(size & (1<<9));
746 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
751 /* there is still a theoretical deadlock when called from receiver
752 * on an D_INCONSISTENT R_PRIMARY:
753 * remote READ does inc_ap_bio, receiver would need to receive answer
754 * packet from remote to dec_ap_bio again.
755 * receiver receive_sizes(), comes here,
756 * waits for ap_bio_cnt == 0. -> deadlock.
757 * but this cannot happen, actually, because:
758 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
759 * (not connected, or bad/no disk on peer):
760 * see drbd_fail_request_early, ap_bio_cnt is zero.
761 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
762 * peer may not initiate a resize.
764 /* Note these are not to be confused with
765 * drbd_adm_suspend_io/drbd_adm_resume_io,
766 * which are (sub) state changes triggered by admin (drbdsetup),
767 * and can be long lived.
768 * This changes an mdev->flag, is triggered by drbd internals,
769 * and should be short-lived. */
770 void drbd_suspend_io(struct drbd_conf *mdev)
772 set_bit(SUSPEND_IO, &mdev->flags);
773 if (drbd_suspended(mdev))
775 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
778 void drbd_resume_io(struct drbd_conf *mdev)
780 clear_bit(SUSPEND_IO, &mdev->flags);
781 wake_up(&mdev->misc_wait);
785 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
786 * @mdev: DRBD device.
788 * Returns 0 on success, negative return values indicate errors.
789 * You should call drbd_md_sync() after calling this function.
791 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
793 sector_t prev_first_sect, prev_size; /* previous meta location */
794 sector_t la_size, u_size;
798 int md_moved, la_size_changed;
799 enum determine_dev_size rv = unchanged;
802 * application request passes inc_ap_bio,
803 * but then cannot get an AL-reference.
804 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
807 * Suspend IO right here.
808 * still lock the act_log to not trigger ASSERTs there.
810 drbd_suspend_io(mdev);
812 /* no wait necessary anymore, actually we could assert that */
813 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
815 prev_first_sect = drbd_md_first_sector(mdev->ldev);
816 prev_size = mdev->ldev->md.md_size_sect;
817 la_size = mdev->ldev->md.la_size_sect;
819 /* TODO: should only be some assert here, not (re)init... */
820 drbd_md_set_sector_offsets(mdev, mdev->ldev);
823 u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
825 size = drbd_new_dev_size(mdev, mdev->ldev, u_size, flags & DDSF_FORCED);
827 if (drbd_get_capacity(mdev->this_bdev) != size ||
828 drbd_bm_capacity(mdev) != size) {
830 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
832 /* currently there is only one error: ENOMEM! */
833 size = drbd_bm_capacity(mdev)>>1;
835 dev_err(DEV, "OUT OF MEMORY! "
836 "Could not allocate bitmap!\n");
838 dev_err(DEV, "BM resizing failed. "
839 "Leaving size unchanged at size = %lu KB\n",
840 (unsigned long)size);
844 /* racy, see comments above. */
845 drbd_set_my_capacity(mdev, size);
846 mdev->ldev->md.la_size_sect = size;
847 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
848 (unsigned long long)size>>1);
850 if (rv == dev_size_error)
853 la_size_changed = (la_size != mdev->ldev->md.la_size_sect);
855 md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
856 || prev_size != mdev->ldev->md.md_size_sect;
858 if (la_size_changed || md_moved) {
861 drbd_al_shrink(mdev); /* All extents inactive. */
862 dev_info(DEV, "Writing the whole bitmap, %s\n",
863 la_size_changed && md_moved ? "size changed and md moved" :
864 la_size_changed ? "size changed" : "md moved");
865 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
866 err = drbd_bitmap_io(mdev, &drbd_bm_write,
867 "size changed", BM_LOCKED_MASK);
872 drbd_md_mark_dirty(mdev);
880 lc_unlock(mdev->act_log);
881 wake_up(&mdev->al_wait);
882 drbd_resume_io(mdev);
888 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
889 sector_t u_size, int assume_peer_has_space)
891 sector_t p_size = mdev->p_size; /* partner's disk size. */
892 sector_t la_size = bdev->md.la_size_sect; /* last agreed size. */
893 sector_t m_size; /* my size */
896 m_size = drbd_get_max_capacity(bdev);
898 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
899 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
903 if (p_size && m_size) {
904 size = min_t(sector_t, p_size, m_size);
908 if (m_size && m_size < size)
910 if (p_size && p_size < size)
921 dev_err(DEV, "Both nodes diskless!\n");
925 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
926 (unsigned long)u_size>>1, (unsigned long)size>>1);
935 * drbd_check_al_size() - Ensures that the AL is of the right size
936 * @mdev: DRBD device.
938 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
939 * failed, and 0 on success. You should call drbd_md_sync() after you called
942 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
944 struct lru_cache *n, *t;
945 struct lc_element *e;
950 mdev->act_log->nr_elements == dc->al_extents)
955 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
956 dc->al_extents, sizeof(struct lc_element), 0);
959 dev_err(DEV, "Cannot allocate act_log lru!\n");
962 spin_lock_irq(&mdev->al_lock);
964 for (i = 0; i < t->nr_elements; i++) {
965 e = lc_element_by_index(t, i);
967 dev_err(DEV, "refcnt(%d)==%d\n",
968 e->lc_number, e->refcnt);
974 spin_unlock_irq(&mdev->al_lock);
976 dev_err(DEV, "Activity log still in use!\n");
983 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
987 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
989 struct request_queue * const q = mdev->rq_queue;
990 int max_hw_sectors = max_bio_size >> 9;
991 int max_segments = 0;
993 if (get_ldev_if_state(mdev, D_ATTACHING)) {
994 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
996 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
998 max_segments = rcu_dereference(mdev->ldev->disk_conf)->max_bio_bvecs;
1003 blk_queue_logical_block_size(q, 512);
1004 blk_queue_max_hw_sectors(q, max_hw_sectors);
1005 /* This is the workaround for "bio would need to, but cannot, be split" */
1006 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1007 blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
1009 if (get_ldev_if_state(mdev, D_ATTACHING)) {
1010 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1012 blk_queue_stack_limits(q, b);
1014 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1015 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1016 q->backing_dev_info.ra_pages,
1017 b->backing_dev_info.ra_pages);
1018 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1024 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1026 int now, new, local, peer;
1028 now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1029 local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1030 peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1032 if (get_ldev_if_state(mdev, D_ATTACHING)) {
1033 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1034 mdev->local_max_bio_size = local;
1038 /* We may ignore peer limits if the peer is modern enough.
1039 Because new from 8.3.8 onwards the peer can use multiple
1040 BIOs for a single peer_request */
1041 if (mdev->state.conn >= C_CONNECTED) {
1042 if (mdev->tconn->agreed_pro_version < 94)
1043 peer = mdev->peer_max_bio_size;
1044 else if (mdev->tconn->agreed_pro_version == 94)
1045 peer = DRBD_MAX_SIZE_H80_PACKET;
1046 else /* drbd 8.3.8 onwards */
1047 peer = DRBD_MAX_BIO_SIZE;
1050 new = min_t(int, local, peer);
1052 if (mdev->state.role == R_PRIMARY && new < now)
1053 dev_err(DEV, "ASSERT FAILED new < now; (%d < %d)\n", new, now);
1056 dev_info(DEV, "max BIO size = %u\n", new);
1058 drbd_setup_queue_param(mdev, new);
1061 /* Starts the worker thread */
1062 static void conn_reconfig_start(struct drbd_tconn *tconn)
1064 drbd_thread_start(&tconn->worker);
1065 conn_flush_workqueue(tconn);
1068 /* if still unconfigured, stops worker again. */
1069 static void conn_reconfig_done(struct drbd_tconn *tconn)
1072 spin_lock_irq(&tconn->req_lock);
1073 stop_threads = conn_all_vols_unconf(tconn);
1074 spin_unlock_irq(&tconn->req_lock);
1076 /* asender is implicitly stopped by receiver
1077 * in conn_disconnect() */
1078 drbd_thread_stop(&tconn->receiver);
1079 drbd_thread_stop(&tconn->worker);
1083 /* Make sure IO is suspended before calling this function(). */
1084 static void drbd_suspend_al(struct drbd_conf *mdev)
1088 if (!lc_try_lock(mdev->act_log)) {
1089 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1093 drbd_al_shrink(mdev);
1094 spin_lock_irq(&mdev->tconn->req_lock);
1095 if (mdev->state.conn < C_CONNECTED)
1096 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1097 spin_unlock_irq(&mdev->tconn->req_lock);
1098 lc_unlock(mdev->act_log);
1101 dev_info(DEV, "Suspended AL updates\n");
1105 static bool should_set_defaults(struct genl_info *info)
1107 unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1108 return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1111 static void enforce_disk_conf_limits(struct disk_conf *dc)
1113 if (dc->al_extents < DRBD_AL_EXTENTS_MIN)
1114 dc->al_extents = DRBD_AL_EXTENTS_MIN;
1115 if (dc->al_extents > DRBD_AL_EXTENTS_MAX)
1116 dc->al_extents = DRBD_AL_EXTENTS_MAX;
1118 if (dc->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1119 dc->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1122 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1124 enum drbd_ret_code retcode;
1125 struct drbd_conf *mdev;
1126 struct disk_conf *new_disk_conf, *old_disk_conf;
1127 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1130 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1131 if (!adm_ctx.reply_skb)
1133 if (retcode != NO_ERROR)
1136 mdev = adm_ctx.mdev;
1138 /* we also need a disk
1139 * to change the options on */
1140 if (!get_ldev(mdev)) {
1141 retcode = ERR_NO_DISK;
1145 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1146 if (!new_disk_conf) {
1147 retcode = ERR_NOMEM;
1151 mutex_lock(&mdev->tconn->conf_update);
1152 old_disk_conf = mdev->ldev->disk_conf;
1153 *new_disk_conf = *old_disk_conf;
1154 if (should_set_defaults(info))
1155 set_disk_conf_defaults(new_disk_conf);
1157 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1158 if (err && err != -ENOMSG) {
1159 retcode = ERR_MANDATORY_TAG;
1160 drbd_msg_put_info(from_attrs_err_to_txt(err));
1163 if (!expect(new_disk_conf->resync_rate >= 1))
1164 new_disk_conf->resync_rate = 1;
1166 enforce_disk_conf_limits(new_disk_conf);
1168 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1169 if (fifo_size != mdev->rs_plan_s->size) {
1170 new_plan = fifo_alloc(fifo_size);
1172 dev_err(DEV, "kmalloc of fifo_buffer failed");
1173 retcode = ERR_NOMEM;
1178 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1179 drbd_al_shrink(mdev);
1180 err = drbd_check_al_size(mdev, new_disk_conf);
1181 lc_unlock(mdev->act_log);
1182 wake_up(&mdev->al_wait);
1185 retcode = ERR_NOMEM;
1189 write_lock_irq(&global_state_lock);
1190 retcode = drbd_resync_after_valid(mdev, new_disk_conf->resync_after);
1191 if (retcode == NO_ERROR) {
1192 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
1193 drbd_resync_after_changed(mdev);
1195 write_unlock_irq(&global_state_lock);
1197 if (retcode != NO_ERROR)
1201 old_plan = mdev->rs_plan_s;
1202 rcu_assign_pointer(mdev->rs_plan_s, new_plan);
1205 mutex_unlock(&mdev->tconn->conf_update);
1208 if (mdev->state.conn >= C_CONNECTED)
1209 drbd_send_sync_param(mdev);
1212 kfree(old_disk_conf);
1217 mutex_unlock(&mdev->tconn->conf_update);
1219 kfree(new_disk_conf);
1224 drbd_adm_finish(info, retcode);
1228 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1230 struct drbd_conf *mdev;
1232 enum drbd_ret_code retcode;
1233 enum determine_dev_size dd;
1234 sector_t max_possible_sectors;
1235 sector_t min_md_device_sectors;
1236 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1237 struct disk_conf *new_disk_conf = NULL;
1238 struct block_device *bdev;
1239 struct lru_cache *resync_lru = NULL;
1240 struct fifo_buffer *new_plan = NULL;
1241 union drbd_state ns, os;
1242 enum drbd_state_rv rv;
1243 struct net_conf *nc;
1244 int cp_discovered = 0;
1246 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1247 if (!adm_ctx.reply_skb)
1249 if (retcode != NO_ERROR)
1252 mdev = adm_ctx.mdev;
1253 conn_reconfig_start(mdev->tconn);
1255 /* if you want to reconfigure, please tear down first */
1256 if (mdev->state.disk > D_DISKLESS) {
1257 retcode = ERR_DISK_CONFIGURED;
1260 /* It may just now have detached because of IO error. Make sure
1261 * drbd_ldev_destroy is done already, we may end up here very fast,
1262 * e.g. if someone calls attach from the on-io-error handler,
1263 * to realize a "hot spare" feature (not that I'd recommend that) */
1264 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1266 /* allocation not in the IO path, drbdsetup context */
1267 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1269 retcode = ERR_NOMEM;
1272 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1273 if (!new_disk_conf) {
1274 retcode = ERR_NOMEM;
1277 nbc->disk_conf = new_disk_conf;
1279 set_disk_conf_defaults(new_disk_conf);
1280 err = disk_conf_from_attrs(new_disk_conf, info);
1282 retcode = ERR_MANDATORY_TAG;
1283 drbd_msg_put_info(from_attrs_err_to_txt(err));
1287 enforce_disk_conf_limits(new_disk_conf);
1289 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1291 retcode = ERR_NOMEM;
1295 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1296 retcode = ERR_MD_IDX_INVALID;
1301 nc = rcu_dereference(mdev->tconn->net_conf);
1303 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1305 retcode = ERR_STONITH_AND_PROT_A;
1311 bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1312 FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1314 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1316 retcode = ERR_OPEN_DISK;
1319 nbc->backing_bdev = bdev;
1322 * meta_dev_idx >= 0: external fixed size, possibly multiple
1323 * drbd sharing one meta device. TODO in that case, paranoia
1324 * check that [md_bdev, meta_dev_idx] is not yet used by some
1325 * other drbd minor! (if you use drbd.conf + drbdadm, that
1326 * should check it for you already; but if you don't, or
1327 * someone fooled it, we need to double check here)
1329 bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1330 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1331 (new_disk_conf->meta_dev_idx < 0) ?
1332 (void *)mdev : (void *)drbd_m_holder);
1334 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1336 retcode = ERR_OPEN_MD_DISK;
1339 nbc->md_bdev = bdev;
1341 if ((nbc->backing_bdev == nbc->md_bdev) !=
1342 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1343 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1344 retcode = ERR_MD_IDX_INVALID;
1348 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1349 1, 61, sizeof(struct bm_extent),
1350 offsetof(struct bm_extent, lce));
1352 retcode = ERR_NOMEM;
1356 /* RT - for drbd_get_max_capacity() DRBD_MD_INDEX_FLEX_INT */
1357 drbd_md_set_sector_offsets(mdev, nbc);
1359 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1360 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1361 (unsigned long long) drbd_get_max_capacity(nbc),
1362 (unsigned long long) new_disk_conf->disk_size);
1363 retcode = ERR_DISK_TOO_SMALL;
1367 if (new_disk_conf->meta_dev_idx < 0) {
1368 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1369 /* at least one MB, otherwise it does not make sense */
1370 min_md_device_sectors = (2<<10);
1372 max_possible_sectors = DRBD_MAX_SECTORS;
1373 min_md_device_sectors = MD_RESERVED_SECT * (new_disk_conf->meta_dev_idx + 1);
1376 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1377 retcode = ERR_MD_DISK_TOO_SMALL;
1378 dev_warn(DEV, "refusing attach: md-device too small, "
1379 "at least %llu sectors needed for this meta-disk type\n",
1380 (unsigned long long) min_md_device_sectors);
1384 /* Make sure the new disk is big enough
1385 * (we may currently be R_PRIMARY with no local disk...) */
1386 if (drbd_get_max_capacity(nbc) <
1387 drbd_get_capacity(mdev->this_bdev)) {
1388 retcode = ERR_DISK_TOO_SMALL;
1392 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1394 if (nbc->known_size > max_possible_sectors) {
1395 dev_warn(DEV, "==> truncating very big lower level device "
1396 "to currently maximum possible %llu sectors <==\n",
1397 (unsigned long long) max_possible_sectors);
1398 if (new_disk_conf->meta_dev_idx >= 0)
1399 dev_warn(DEV, "==>> using internal or flexible "
1400 "meta data may help <<==\n");
1403 drbd_suspend_io(mdev);
1404 /* also wait for the last barrier ack. */
1405 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1406 /* and for any other previously queued work */
1407 drbd_flush_workqueue(mdev);
1409 rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1410 retcode = rv; /* FIXME: Type mismatch. */
1411 drbd_resume_io(mdev);
1412 if (rv < SS_SUCCESS)
1415 if (!get_ldev_if_state(mdev, D_ATTACHING))
1416 goto force_diskless;
1418 drbd_md_set_sector_offsets(mdev, nbc);
1420 if (!mdev->bitmap) {
1421 if (drbd_bm_init(mdev)) {
1422 retcode = ERR_NOMEM;
1423 goto force_diskless_dec;
1427 retcode = drbd_md_read(mdev, nbc);
1428 if (retcode != NO_ERROR)
1429 goto force_diskless_dec;
1431 if (mdev->state.conn < C_CONNECTED &&
1432 mdev->state.role == R_PRIMARY &&
1433 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1434 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1435 (unsigned long long)mdev->ed_uuid);
1436 retcode = ERR_DATA_NOT_CURRENT;
1437 goto force_diskless_dec;
1440 /* Since we are diskless, fix the activity log first... */
1441 if (drbd_check_al_size(mdev, new_disk_conf)) {
1442 retcode = ERR_NOMEM;
1443 goto force_diskless_dec;
1446 /* Prevent shrinking of consistent devices ! */
1447 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1448 drbd_new_dev_size(mdev, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1449 dev_warn(DEV, "refusing to truncate a consistent device\n");
1450 retcode = ERR_DISK_TOO_SMALL;
1451 goto force_diskless_dec;
1454 if (!drbd_al_read_log(mdev, nbc)) {
1455 retcode = ERR_IO_MD_DISK;
1456 goto force_diskless_dec;
1459 /* Reset the "barriers don't work" bits here, then force meta data to
1460 * be written, to ensure we determine if barriers are supported. */
1461 if (new_disk_conf->md_flushes)
1462 clear_bit(MD_NO_FUA, &mdev->flags);
1464 set_bit(MD_NO_FUA, &mdev->flags);
1466 /* Point of no return reached.
1467 * Devices and memory are no longer released by error cleanup below.
1468 * now mdev takes over responsibility, and the state engine should
1469 * clean it up somewhere. */
1470 D_ASSERT(mdev->ldev == NULL);
1472 mdev->resync = resync_lru;
1473 mdev->rs_plan_s = new_plan;
1476 new_disk_conf = NULL;
1479 mdev->write_ordering = WO_bdev_flush;
1480 drbd_bump_write_ordering(mdev, WO_bdev_flush);
1482 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1483 set_bit(CRASHED_PRIMARY, &mdev->flags);
1485 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1487 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1488 !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod)) {
1489 set_bit(CRASHED_PRIMARY, &mdev->flags);
1498 drbd_reconsider_max_bio_size(mdev);
1500 /* If I am currently not R_PRIMARY,
1501 * but meta data primary indicator is set,
1502 * I just now recover from a hard crash,
1503 * and have been R_PRIMARY before that crash.
1505 * Now, if I had no connection before that crash
1506 * (have been degraded R_PRIMARY), chances are that
1507 * I won't find my peer now either.
1509 * In that case, and _only_ in that case,
1510 * we use the degr-wfc-timeout instead of the default,
1511 * so we can automatically recover from a crash of a
1512 * degraded but active "cluster" after a certain timeout.
1514 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1515 if (mdev->state.role != R_PRIMARY &&
1516 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1517 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1518 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1520 dd = drbd_determine_dev_size(mdev, 0);
1521 if (dd == dev_size_error) {
1522 retcode = ERR_NOMEM_BITMAP;
1523 goto force_diskless_dec;
1524 } else if (dd == grew)
1525 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1527 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1528 dev_info(DEV, "Assuming that all blocks are out of sync "
1529 "(aka FullSync)\n");
1530 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1531 "set_n_write from attaching", BM_LOCKED_MASK)) {
1532 retcode = ERR_IO_MD_DISK;
1533 goto force_diskless_dec;
1536 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1537 "read from attaching", BM_LOCKED_MASK)) {
1538 retcode = ERR_IO_MD_DISK;
1539 goto force_diskless_dec;
1543 if (cp_discovered) {
1544 drbd_al_apply_to_bm(mdev);
1545 if (drbd_bitmap_io(mdev, &drbd_bm_write,
1546 "crashed primary apply AL", BM_LOCKED_MASK)) {
1547 retcode = ERR_IO_MD_DISK;
1548 goto force_diskless_dec;
1552 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1553 drbd_suspend_al(mdev); /* IO is still suspended here... */
1555 spin_lock_irq(&mdev->tconn->req_lock);
1556 os = drbd_read_state(mdev);
1558 /* If MDF_CONSISTENT is not set go into inconsistent state,
1559 otherwise investigate MDF_WasUpToDate...
1560 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1561 otherwise into D_CONSISTENT state.
1563 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1564 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1565 ns.disk = D_CONSISTENT;
1567 ns.disk = D_OUTDATED;
1569 ns.disk = D_INCONSISTENT;
1572 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1573 ns.pdsk = D_OUTDATED;
1576 if (ns.disk == D_CONSISTENT &&
1577 (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1578 ns.disk = D_UP_TO_DATE;
1581 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1582 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1583 this point, because drbd_request_state() modifies these
1586 /* In case we are C_CONNECTED postpone any decision on the new disk
1587 state after the negotiation phase. */
1588 if (mdev->state.conn == C_CONNECTED) {
1589 mdev->new_state_tmp.i = ns.i;
1591 ns.disk = D_NEGOTIATING;
1593 /* We expect to receive up-to-date UUIDs soon.
1594 To avoid a race in receive_state, free p_uuid while
1595 holding req_lock. I.e. atomic with the state change */
1596 kfree(mdev->p_uuid);
1597 mdev->p_uuid = NULL;
1600 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1601 spin_unlock_irq(&mdev->tconn->req_lock);
1603 if (rv < SS_SUCCESS)
1604 goto force_diskless_dec;
1606 if (mdev->state.role == R_PRIMARY)
1607 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1609 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1611 drbd_md_mark_dirty(mdev);
1614 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1616 conn_reconfig_done(mdev->tconn);
1617 drbd_adm_finish(info, retcode);
1623 drbd_force_state(mdev, NS(disk, D_FAILED));
1626 conn_reconfig_done(mdev->tconn);
1628 if (nbc->backing_bdev)
1629 blkdev_put(nbc->backing_bdev,
1630 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1632 blkdev_put(nbc->md_bdev,
1633 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1636 kfree(new_disk_conf);
1637 lc_destroy(resync_lru);
1641 drbd_adm_finish(info, retcode);
1645 static int adm_detach(struct drbd_conf *mdev)
1647 enum drbd_state_rv retcode;
1649 drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1650 retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1651 /* D_FAILED will transition to DISKLESS. */
1652 ret = wait_event_interruptible(mdev->misc_wait,
1653 mdev->state.disk != D_FAILED);
1654 drbd_resume_io(mdev);
1655 if ((int)retcode == (int)SS_IS_DISKLESS)
1656 retcode = SS_NOTHING_TO_DO;
1662 /* Detaching the disk is a process in multiple stages. First we need to lock
1663 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1664 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1665 * internal references as well.
1666 * Only then we have finally detached. */
1667 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1669 enum drbd_ret_code retcode;
1671 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1672 if (!adm_ctx.reply_skb)
1674 if (retcode != NO_ERROR)
1677 retcode = adm_detach(adm_ctx.mdev);
1679 drbd_adm_finish(info, retcode);
1683 static bool conn_resync_running(struct drbd_tconn *tconn)
1685 struct drbd_conf *mdev;
1690 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1691 if (mdev->state.conn == C_SYNC_SOURCE ||
1692 mdev->state.conn == C_SYNC_TARGET ||
1693 mdev->state.conn == C_PAUSED_SYNC_S ||
1694 mdev->state.conn == C_PAUSED_SYNC_T) {
1704 static bool conn_ov_running(struct drbd_tconn *tconn)
1706 struct drbd_conf *mdev;
1711 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1712 if (mdev->state.conn == C_VERIFY_S ||
1713 mdev->state.conn == C_VERIFY_T) {
1723 static enum drbd_ret_code
1724 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1726 struct drbd_conf *mdev;
1729 if (old_conf && tconn->cstate == C_WF_REPORT_PARAMS && tconn->agreed_pro_version < 100) {
1730 if (new_conf->wire_protocol != old_conf->wire_protocol)
1731 return ERR_NEED_APV_100;
1733 if (new_conf->two_primaries != old_conf->two_primaries)
1734 return ERR_NEED_APV_100;
1736 if (!new_conf->integrity_alg != !old_conf->integrity_alg)
1737 return ERR_NEED_APV_100;
1739 if (strcmp(new_conf->integrity_alg, old_conf->integrity_alg))
1740 return ERR_NEED_APV_100;
1743 if (!new_conf->two_primaries &&
1744 conn_highest_role(tconn) == R_PRIMARY &&
1745 conn_highest_peer(tconn) == R_PRIMARY)
1746 return ERR_NEED_ALLOW_TWO_PRI;
1748 if (new_conf->two_primaries &&
1749 (new_conf->wire_protocol != DRBD_PROT_C))
1750 return ERR_NOT_PROTO_C;
1752 idr_for_each_entry(&tconn->volumes, mdev, i) {
1753 if (get_ldev(mdev)) {
1754 enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1756 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1757 return ERR_STONITH_AND_PROT_A;
1759 if (mdev->state.role == R_PRIMARY && new_conf->discard_my_data)
1763 if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1764 return ERR_CONG_NOT_PROTO_A;
1769 static enum drbd_ret_code
1770 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1772 static enum drbd_ret_code rv;
1773 struct drbd_conf *mdev;
1777 rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1780 /* tconn->volumes protected by genl_lock() here */
1781 idr_for_each_entry(&tconn->volumes, mdev, i) {
1782 if (!mdev->bitmap) {
1783 if(drbd_bm_init(mdev))
1792 struct crypto_hash *verify_tfm;
1793 struct crypto_hash *csums_tfm;
1794 struct crypto_hash *cram_hmac_tfm;
1795 struct crypto_hash *integrity_tfm;
1801 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1806 *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1815 static enum drbd_ret_code
1816 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1818 char hmac_name[CRYPTO_MAX_ALG_NAME];
1819 enum drbd_ret_code rv;
1822 rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1826 rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1830 rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1834 if (new_conf->cram_hmac_alg[0] != 0) {
1835 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1836 new_conf->cram_hmac_alg);
1838 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1841 if (crypto->integrity_tfm) {
1842 hash_size = crypto_hash_digestsize(crypto->integrity_tfm);
1843 crypto->int_dig_in = kmalloc(hash_size, GFP_KERNEL);
1844 if (!crypto->int_dig_in)
1846 crypto->int_dig_vv = kmalloc(hash_size, GFP_KERNEL);
1847 if (!crypto->int_dig_vv)
1854 static void free_crypto(struct crypto *crypto)
1856 kfree(crypto->int_dig_in);
1857 kfree(crypto->int_dig_vv);
1858 crypto_free_hash(crypto->cram_hmac_tfm);
1859 crypto_free_hash(crypto->integrity_tfm);
1860 crypto_free_hash(crypto->csums_tfm);
1861 crypto_free_hash(crypto->verify_tfm);
1864 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1866 enum drbd_ret_code retcode;
1867 struct drbd_tconn *tconn;
1868 struct net_conf *old_conf, *new_conf = NULL;
1870 int ovr; /* online verify running */
1871 int rsr; /* re-sync running */
1872 struct crypto crypto = { };
1874 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1875 if (!adm_ctx.reply_skb)
1877 if (retcode != NO_ERROR)
1880 tconn = adm_ctx.tconn;
1882 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1884 retcode = ERR_NOMEM;
1888 conn_reconfig_start(tconn);
1890 mutex_lock(&tconn->data.mutex);
1891 mutex_lock(&tconn->conf_update);
1892 old_conf = tconn->net_conf;
1895 drbd_msg_put_info("net conf missing, try connect");
1896 retcode = ERR_INVALID_REQUEST;
1900 *new_conf = *old_conf;
1901 if (should_set_defaults(info))
1902 set_net_conf_defaults(new_conf);
1904 err = net_conf_from_attrs_for_change(new_conf, info);
1905 if (err && err != -ENOMSG) {
1906 retcode = ERR_MANDATORY_TAG;
1907 drbd_msg_put_info(from_attrs_err_to_txt(err));
1911 retcode = check_net_options(tconn, new_conf);
1912 if (retcode != NO_ERROR)
1915 /* re-sync running */
1916 rsr = conn_resync_running(tconn);
1917 if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1918 retcode = ERR_CSUMS_RESYNC_RUNNING;
1922 /* online verify running */
1923 ovr = conn_ov_running(tconn);
1924 if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1925 retcode = ERR_VERIFY_RUNNING;
1929 retcode = alloc_crypto(&crypto, new_conf);
1930 if (retcode != NO_ERROR)
1933 rcu_assign_pointer(tconn->net_conf, new_conf);
1936 crypto_free_hash(tconn->csums_tfm);
1937 tconn->csums_tfm = crypto.csums_tfm;
1938 crypto.csums_tfm = NULL;
1941 crypto_free_hash(tconn->verify_tfm);
1942 tconn->verify_tfm = crypto.verify_tfm;
1943 crypto.verify_tfm = NULL;
1946 kfree(tconn->int_dig_in);
1947 tconn->int_dig_in = crypto.int_dig_in;
1948 kfree(tconn->int_dig_vv);
1949 tconn->int_dig_vv = crypto.int_dig_vv;
1950 crypto_free_hash(tconn->integrity_tfm);
1951 tconn->integrity_tfm = crypto.integrity_tfm;
1952 if (tconn->cstate >= C_WF_REPORT_PARAMS && tconn->agreed_pro_version >= 100)
1953 /* Do this without trying to take tconn->data.mutex again. */
1954 __drbd_send_protocol(tconn, P_PROTOCOL_UPDATE);
1956 crypto_free_hash(tconn->cram_hmac_tfm);
1957 tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
1959 mutex_unlock(&tconn->conf_update);
1960 mutex_unlock(&tconn->data.mutex);
1964 if (tconn->cstate >= C_WF_REPORT_PARAMS)
1965 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
1970 mutex_unlock(&tconn->conf_update);
1971 mutex_unlock(&tconn->data.mutex);
1972 free_crypto(&crypto);
1975 conn_reconfig_done(tconn);
1977 drbd_adm_finish(info, retcode);
1981 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
1983 struct drbd_conf *mdev;
1984 struct net_conf *old_conf, *new_conf = NULL;
1985 struct crypto crypto = { };
1986 struct drbd_tconn *oconn;
1987 struct drbd_tconn *tconn;
1988 struct sockaddr *new_my_addr, *new_peer_addr, *taken_addr;
1989 enum drbd_ret_code retcode;
1993 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
1994 if (!adm_ctx.reply_skb)
1996 if (retcode != NO_ERROR)
1999 tconn = adm_ctx.tconn;
2000 conn_reconfig_start(tconn);
2002 if (tconn->cstate > C_STANDALONE) {
2003 retcode = ERR_NET_CONFIGURED;
2007 /* allocation not in the IO path, cqueue thread context */
2008 new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
2010 retcode = ERR_NOMEM;
2014 set_net_conf_defaults(new_conf);
2016 err = net_conf_from_attrs(new_conf, info);
2018 retcode = ERR_MANDATORY_TAG;
2019 drbd_msg_put_info(from_attrs_err_to_txt(err));
2023 retcode = check_net_options(tconn, new_conf);
2024 if (retcode != NO_ERROR)
2029 new_my_addr = (struct sockaddr *)&new_conf->my_addr;
2030 new_peer_addr = (struct sockaddr *)&new_conf->peer_addr;
2032 /* No need for _rcu here. All reconfiguration is
2033 * strictly serialized on genl_lock(). We are protected against
2034 * concurrent reconfiguration/addition/deletion */
2035 list_for_each_entry(oconn, &drbd_tconns, all_tconn) {
2036 struct net_conf *nc;
2041 nc = rcu_dereference(oconn->net_conf);
2043 taken_addr = (struct sockaddr *)&nc->my_addr;
2044 if (new_conf->my_addr_len == nc->my_addr_len &&
2045 !memcmp(new_my_addr, taken_addr, new_conf->my_addr_len))
2046 retcode = ERR_LOCAL_ADDR;
2048 taken_addr = (struct sockaddr *)&nc->peer_addr;
2049 if (new_conf->peer_addr_len == nc->peer_addr_len &&
2050 !memcmp(new_peer_addr, taken_addr, new_conf->peer_addr_len))
2051 retcode = ERR_PEER_ADDR;
2054 if (retcode != NO_ERROR)
2058 retcode = alloc_crypto(&crypto, new_conf);
2059 if (retcode != NO_ERROR)
2062 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2064 conn_flush_workqueue(tconn);
2066 mutex_lock(&tconn->conf_update);
2067 old_conf = tconn->net_conf;
2069 retcode = ERR_NET_CONFIGURED;
2070 mutex_unlock(&tconn->conf_update);
2073 rcu_assign_pointer(tconn->net_conf, new_conf);
2075 conn_free_crypto(tconn);
2076 tconn->int_dig_in = crypto.int_dig_in;
2077 tconn->int_dig_vv = crypto.int_dig_vv;
2078 tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2079 tconn->integrity_tfm = crypto.integrity_tfm;
2080 tconn->csums_tfm = crypto.csums_tfm;
2081 tconn->verify_tfm = crypto.verify_tfm;
2083 mutex_unlock(&tconn->conf_update);
2086 idr_for_each_entry(&tconn->volumes, mdev, i) {
2092 retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2094 conn_reconfig_done(tconn);
2095 drbd_adm_finish(info, retcode);
2099 free_crypto(&crypto);
2102 conn_reconfig_done(tconn);
2104 drbd_adm_finish(info, retcode);
2108 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2110 enum drbd_state_rv rv;
2112 rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2113 force ? CS_HARD : 0);
2116 case SS_NOTHING_TO_DO:
2118 case SS_ALREADY_STANDALONE:
2120 case SS_PRIMARY_NOP:
2121 /* Our state checking code wants to see the peer outdated. */
2122 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2123 pdsk, D_OUTDATED), CS_VERBOSE);
2125 case SS_CW_FAILED_BY_PEER:
2126 /* The peer probably wants to see us outdated. */
2127 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2128 disk, D_OUTDATED), 0);
2129 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2130 rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2135 /* no special handling necessary */
2138 if (rv >= SS_SUCCESS) {
2139 enum drbd_state_rv rv2;
2140 /* No one else can reconfigure the network while I am here.
2141 * The state handling only uses drbd_thread_stop_nowait(),
2142 * we want to really wait here until the receiver is no more.
2144 drbd_thread_stop(&adm_ctx.tconn->receiver);
2146 /* Race breaker. This additional state change request may be
2147 * necessary, if this was a forced disconnect during a receiver
2148 * restart. We may have "killed" the receiver thread just
2149 * after drbdd_init() returned. Typically, we should be
2150 * C_STANDALONE already, now, and this becomes a no-op.
2152 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2153 CS_VERBOSE | CS_HARD);
2154 if (rv2 < SS_SUCCESS)
2156 "unexpected rv2=%d in conn_try_disconnect()\n",
2162 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2164 struct disconnect_parms parms;
2165 struct drbd_tconn *tconn;
2166 enum drbd_state_rv rv;
2167 enum drbd_ret_code retcode;
2170 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2171 if (!adm_ctx.reply_skb)
2173 if (retcode != NO_ERROR)
2176 tconn = adm_ctx.tconn;
2177 memset(&parms, 0, sizeof(parms));
2178 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2179 err = disconnect_parms_from_attrs(&parms, info);
2181 retcode = ERR_MANDATORY_TAG;
2182 drbd_msg_put_info(from_attrs_err_to_txt(err));
2187 rv = conn_try_disconnect(tconn, parms.force_disconnect);
2188 if (rv < SS_SUCCESS)
2189 retcode = rv; /* FIXME: Type mismatch. */
2193 drbd_adm_finish(info, retcode);
2197 void resync_after_online_grow(struct drbd_conf *mdev)
2199 int iass; /* I am sync source */
2201 dev_info(DEV, "Resync of new storage after online grow\n");
2202 if (mdev->state.role != mdev->state.peer)
2203 iass = (mdev->state.role == R_PRIMARY);
2205 iass = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags);
2208 drbd_start_resync(mdev, C_SYNC_SOURCE);
2210 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2213 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2215 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2216 struct resize_parms rs;
2217 struct drbd_conf *mdev;
2218 enum drbd_ret_code retcode;
2219 enum determine_dev_size dd;
2220 enum dds_flags ddsf;
2224 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2225 if (!adm_ctx.reply_skb)
2227 if (retcode != NO_ERROR)
2230 memset(&rs, 0, sizeof(struct resize_parms));
2231 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2232 err = resize_parms_from_attrs(&rs, info);
2234 retcode = ERR_MANDATORY_TAG;
2235 drbd_msg_put_info(from_attrs_err_to_txt(err));
2240 mdev = adm_ctx.mdev;
2241 if (mdev->state.conn > C_CONNECTED) {
2242 retcode = ERR_RESIZE_RESYNC;
2246 if (mdev->state.role == R_SECONDARY &&
2247 mdev->state.peer == R_SECONDARY) {
2248 retcode = ERR_NO_PRIMARY;
2252 if (!get_ldev(mdev)) {
2253 retcode = ERR_NO_DISK;
2257 if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2258 retcode = ERR_NEED_APV_93;
2263 u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2265 if (u_size != (sector_t)rs.resize_size) {
2266 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2267 if (!new_disk_conf) {
2268 retcode = ERR_NOMEM;
2273 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2274 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2276 if (new_disk_conf) {
2277 mutex_lock(&mdev->tconn->conf_update);
2278 old_disk_conf = mdev->ldev->disk_conf;
2279 *new_disk_conf = *old_disk_conf;
2280 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2281 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2282 mutex_unlock(&mdev->tconn->conf_update);
2284 kfree(old_disk_conf);
2287 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2288 dd = drbd_determine_dev_size(mdev, ddsf);
2291 if (dd == dev_size_error) {
2292 retcode = ERR_NOMEM_BITMAP;
2296 if (mdev->state.conn == C_CONNECTED) {
2298 set_bit(RESIZE_PENDING, &mdev->flags);
2300 drbd_send_uuids(mdev);
2301 drbd_send_sizes(mdev, 1, ddsf);
2305 drbd_adm_finish(info, retcode);
2309 void drbd_set_res_opts_defaults(struct res_opts *r)
2311 return set_res_opts_defaults(r);
2314 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2316 enum drbd_ret_code retcode;
2317 cpumask_var_t new_cpu_mask;
2318 struct drbd_tconn *tconn;
2319 struct res_opts res_opts;
2322 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
2323 if (!adm_ctx.reply_skb)
2325 if (retcode != NO_ERROR)
2327 tconn = adm_ctx.tconn;
2329 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) {
2330 retcode = ERR_NOMEM;
2331 drbd_msg_put_info("unable to allocate cpumask");
2335 res_opts = tconn->res_opts;
2336 if (should_set_defaults(info))
2337 set_res_opts_defaults(&res_opts);
2339 err = res_opts_from_attrs(&res_opts, info);
2340 if (err && err != -ENOMSG) {
2341 retcode = ERR_MANDATORY_TAG;
2342 drbd_msg_put_info(from_attrs_err_to_txt(err));
2346 /* silently ignore cpu mask on UP kernel */
2347 if (nr_cpu_ids > 1 && res_opts.cpu_mask[0] != 0) {
2348 err = __bitmap_parse(res_opts.cpu_mask, 32, 0,
2349 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2351 conn_warn(tconn, "__bitmap_parse() failed with %d\n", err);
2352 retcode = ERR_CPU_MASK_PARSE;
2358 tconn->res_opts = res_opts;
2360 if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2361 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2362 drbd_calc_cpu_mask(tconn);
2363 tconn->receiver.reset_cpu_mask = 1;
2364 tconn->asender.reset_cpu_mask = 1;
2365 tconn->worker.reset_cpu_mask = 1;
2369 free_cpumask_var(new_cpu_mask);
2371 drbd_adm_finish(info, retcode);
2375 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2377 struct drbd_conf *mdev;
2378 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2380 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2381 if (!adm_ctx.reply_skb)
2383 if (retcode != NO_ERROR)
2386 mdev = adm_ctx.mdev;
2388 /* If there is still bitmap IO pending, probably because of a previous
2389 * resync just being finished, wait for it before requesting a new resync. */
2390 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2392 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2394 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2395 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2397 while (retcode == SS_NEED_CONNECTION) {
2398 spin_lock_irq(&mdev->tconn->req_lock);
2399 if (mdev->state.conn < C_CONNECTED)
2400 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2401 spin_unlock_irq(&mdev->tconn->req_lock);
2403 if (retcode != SS_NEED_CONNECTION)
2406 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2410 drbd_adm_finish(info, retcode);
2414 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2418 rv = drbd_bmio_set_n_write(mdev);
2419 drbd_suspend_al(mdev);
2423 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2424 union drbd_state mask, union drbd_state val)
2426 enum drbd_ret_code retcode;
2428 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2429 if (!adm_ctx.reply_skb)
2431 if (retcode != NO_ERROR)
2434 retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2436 drbd_adm_finish(info, retcode);
2440 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2442 return drbd_adm_simple_request_state(skb, info, NS(conn, C_STARTING_SYNC_S));
2445 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2447 enum drbd_ret_code retcode;
2449 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2450 if (!adm_ctx.reply_skb)
2452 if (retcode != NO_ERROR)
2455 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2456 retcode = ERR_PAUSE_IS_SET;
2458 drbd_adm_finish(info, retcode);
2462 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2464 union drbd_dev_state s;
2465 enum drbd_ret_code retcode;
2467 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2468 if (!adm_ctx.reply_skb)
2470 if (retcode != NO_ERROR)
2473 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2474 s = adm_ctx.mdev->state;
2475 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2476 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2477 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2479 retcode = ERR_PAUSE_IS_CLEAR;
2484 drbd_adm_finish(info, retcode);
2488 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2490 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2493 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2495 struct drbd_conf *mdev;
2496 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2498 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2499 if (!adm_ctx.reply_skb)
2501 if (retcode != NO_ERROR)
2504 mdev = adm_ctx.mdev;
2505 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2506 drbd_uuid_new_current(mdev);
2507 clear_bit(NEW_CUR_UUID, &mdev->flags);
2509 drbd_suspend_io(mdev);
2510 retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2511 if (retcode == SS_SUCCESS) {
2512 if (mdev->state.conn < C_CONNECTED)
2513 tl_clear(mdev->tconn);
2514 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2515 tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2517 drbd_resume_io(mdev);
2520 drbd_adm_finish(info, retcode);
2524 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2526 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2529 int nla_put_drbd_cfg_context(struct sk_buff *skb, const char *conn_name, unsigned vnr)
2532 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2534 goto nla_put_failure;
2535 if (vnr != VOLUME_UNSPECIFIED)
2536 NLA_PUT_U32(skb, T_ctx_volume, vnr);
2537 NLA_PUT_STRING(skb, T_ctx_conn_name, conn_name);
2538 nla_nest_end(skb, nla);
2543 nla_nest_cancel(skb, nla);
2547 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2548 const struct sib_info *sib)
2550 struct state_info *si = NULL; /* for sizeof(si->member); */
2551 struct net_conf *nc;
2555 int exclude_sensitive;
2557 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2558 * to. So we better exclude_sensitive information.
2560 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2561 * in the context of the requesting user process. Exclude sensitive
2562 * information, unless current has superuser.
2564 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2565 * relies on the current implementation of netlink_dump(), which
2566 * executes the dump callback successively from netlink_recvmsg(),
2567 * always in the context of the receiving process */
2568 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2570 got_ldev = get_ldev(mdev);
2572 /* We need to add connection name and volume number information still.
2573 * Minor number is in drbd_genlmsghdr. */
2574 if (nla_put_drbd_cfg_context(skb, mdev->tconn->name, mdev->vnr))
2575 goto nla_put_failure;
2577 if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2578 goto nla_put_failure;
2582 if (disk_conf_to_skb(skb, rcu_dereference(mdev->ldev->disk_conf), exclude_sensitive))
2583 goto nla_put_failure;
2585 nc = rcu_dereference(mdev->tconn->net_conf);
2587 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2590 goto nla_put_failure;
2592 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2594 goto nla_put_failure;
2595 NLA_PUT_U32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY);
2596 NLA_PUT_U32(skb, T_current_state, mdev->state.i);
2597 NLA_PUT_U64(skb, T_ed_uuid, mdev->ed_uuid);
2598 NLA_PUT_U64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev));
2601 NLA_PUT_U32(skb, T_disk_flags, mdev->ldev->md.flags);
2602 NLA_PUT(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2603 NLA_PUT_U64(skb, T_bits_total, drbd_bm_bits(mdev));
2604 NLA_PUT_U64(skb, T_bits_oos, drbd_bm_total_weight(mdev));
2605 if (C_SYNC_SOURCE <= mdev->state.conn &&
2606 C_PAUSED_SYNC_T >= mdev->state.conn) {
2607 NLA_PUT_U64(skb, T_bits_rs_total, mdev->rs_total);
2608 NLA_PUT_U64(skb, T_bits_rs_failed, mdev->rs_failed);
2613 switch(sib->sib_reason) {
2614 case SIB_SYNC_PROGRESS:
2615 case SIB_GET_STATUS_REPLY:
2617 case SIB_STATE_CHANGE:
2618 NLA_PUT_U32(skb, T_prev_state, sib->os.i);
2619 NLA_PUT_U32(skb, T_new_state, sib->ns.i);
2621 case SIB_HELPER_POST:
2623 T_helper_exit_code, sib->helper_exit_code);
2625 case SIB_HELPER_PRE:
2626 NLA_PUT_STRING(skb, T_helper, sib->helper_name);
2630 nla_nest_end(skb, nla);
2640 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2642 enum drbd_ret_code retcode;
2645 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2646 if (!adm_ctx.reply_skb)
2648 if (retcode != NO_ERROR)
2651 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2653 nlmsg_free(adm_ctx.reply_skb);
2657 drbd_adm_finish(info, retcode);
2661 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2663 struct drbd_conf *mdev;
2664 struct drbd_genlmsghdr *dh;
2665 struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2666 struct drbd_tconn *tconn = NULL;
2667 struct drbd_tconn *tmp;
2668 unsigned volume = cb->args[1];
2670 /* Open coded, deferred, iteration:
2671 * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2672 * idr_for_each_entry(&tconn->volumes, mdev, i) {
2676 * where tconn is cb->args[0];
2677 * and i is cb->args[1];
2679 * cb->args[2] indicates if we shall loop over all resources,
2680 * or just dump all volumes of a single resource.
2682 * This may miss entries inserted after this dump started,
2683 * or entries deleted before they are reached.
2685 * We need to make sure the mdev won't disappear while
2686 * we are looking at it, and revalidate our iterators
2687 * on each iteration.
2690 /* synchronize with conn_create()/conn_destroy() */
2692 /* revalidate iterator position */
2693 list_for_each_entry_rcu(tmp, &drbd_tconns, all_tconn) {
2695 /* first iteration */
2707 mdev = idr_get_next(&tconn->volumes, &volume);
2709 /* No more volumes to dump on this tconn.
2710 * Advance tconn iterator. */
2711 pos = list_entry_rcu(tconn->all_tconn.next,
2712 struct drbd_tconn, all_tconn);
2713 /* Did we dump any volume on this tconn yet? */
2715 /* If we reached the end of the list,
2716 * or only a single resource dump was requested,
2718 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2726 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).pid,
2727 cb->nlh->nlmsg_seq, &drbd_genl_family,
2728 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2733 /* this is a tconn without a single volume */
2735 dh->ret_code = NO_ERROR;
2736 if (nla_put_drbd_cfg_context(skb, tconn->name, VOLUME_UNSPECIFIED))
2737 genlmsg_cancel(skb, dh);
2739 genlmsg_end(skb, dh);
2743 D_ASSERT(mdev->vnr == volume);
2744 D_ASSERT(mdev->tconn == tconn);
2746 dh->minor = mdev_to_minor(mdev);
2747 dh->ret_code = NO_ERROR;
2749 if (nla_put_status_info(skb, mdev, NULL)) {
2750 genlmsg_cancel(skb, dh);
2753 genlmsg_end(skb, dh);
2758 /* where to start the next iteration */
2759 cb->args[0] = (long)pos;
2760 cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2762 /* No more tconns/volumes/minors found results in an empty skb.
2763 * Which will terminate the dump. */
2768 * Request status of all resources, or of all volumes within a single resource.
2770 * This is a dump, as the answer may not fit in a single reply skb otherwise.
2771 * Which means we cannot use the family->attrbuf or other such members, because
2772 * dump is NOT protected by the genl_lock(). During dump, we only have access
2773 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2775 * Once things are setup properly, we call into get_one_status().
2777 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2779 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2781 const char *conn_name;
2782 struct drbd_tconn *tconn;
2784 /* Is this a followup call? */
2786 /* ... of a single resource dump,
2787 * and the resource iterator has been advanced already? */
2788 if (cb->args[2] && cb->args[2] != cb->args[0])
2789 return 0; /* DONE. */
2793 /* First call (from netlink_dump_start). We need to figure out
2794 * which resource(s) the user wants us to dump. */
2795 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2796 nlmsg_attrlen(cb->nlh, hdrlen),
2797 DRBD_NLA_CFG_CONTEXT);
2799 /* No explicit context given. Dump all. */
2802 nla = nla_find_nested(nla, __nla_type(T_ctx_conn_name));
2803 /* context given, but no name present? */
2806 conn_name = nla_data(nla);
2807 tconn = conn_get_by_name(conn_name);
2812 kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2814 /* prime iterators, and set "filter" mode mark:
2815 * only dump this tconn. */
2816 cb->args[0] = (long)tconn;
2817 /* cb->args[1] = 0; passed in this way. */
2818 cb->args[2] = (long)tconn;
2821 return get_one_status(skb, cb);
2824 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2826 enum drbd_ret_code retcode;
2827 struct timeout_parms tp;
2830 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2831 if (!adm_ctx.reply_skb)
2833 if (retcode != NO_ERROR)
2837 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2838 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2841 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2843 nlmsg_free(adm_ctx.reply_skb);
2847 drbd_adm_finish(info, retcode);
2851 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2853 struct drbd_conf *mdev;
2854 enum drbd_ret_code retcode;
2856 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2857 if (!adm_ctx.reply_skb)
2859 if (retcode != NO_ERROR)
2862 mdev = adm_ctx.mdev;
2863 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2864 /* resume from last known position, if possible */
2865 struct start_ov_parms parms =
2866 { .ov_start_sector = mdev->ov_start_sector };
2867 int err = start_ov_parms_from_attrs(&parms, info);
2869 retcode = ERR_MANDATORY_TAG;
2870 drbd_msg_put_info(from_attrs_err_to_txt(err));
2873 /* w_make_ov_request expects position to be aligned */
2874 mdev->ov_start_sector = parms.ov_start_sector & ~BM_SECT_PER_BIT;
2876 /* If there is still bitmap IO pending, e.g. previous resync or verify
2877 * just being finished, wait for it before requesting a new resync. */
2878 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2879 retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2881 drbd_adm_finish(info, retcode);
2886 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
2888 struct drbd_conf *mdev;
2889 enum drbd_ret_code retcode;
2890 int skip_initial_sync = 0;
2892 struct new_c_uuid_parms args;
2894 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2895 if (!adm_ctx.reply_skb)
2897 if (retcode != NO_ERROR)
2900 mdev = adm_ctx.mdev;
2901 memset(&args, 0, sizeof(args));
2902 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
2903 err = new_c_uuid_parms_from_attrs(&args, info);
2905 retcode = ERR_MANDATORY_TAG;
2906 drbd_msg_put_info(from_attrs_err_to_txt(err));
2911 mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
2913 if (!get_ldev(mdev)) {
2914 retcode = ERR_NO_DISK;
2918 /* this is "skip initial sync", assume to be clean */
2919 if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
2920 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
2921 dev_info(DEV, "Preparing to skip initial sync\n");
2922 skip_initial_sync = 1;
2923 } else if (mdev->state.conn != C_STANDALONE) {
2924 retcode = ERR_CONNECTED;
2928 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
2929 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
2931 if (args.clear_bm) {
2932 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
2933 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
2935 dev_err(DEV, "Writing bitmap failed with %d\n",err);
2936 retcode = ERR_IO_MD_DISK;
2938 if (skip_initial_sync) {
2939 drbd_send_uuids_skip_initial_sync(mdev);
2940 _drbd_uuid_set(mdev, UI_BITMAP, 0);
2941 drbd_print_uuids(mdev, "cleared bitmap UUID");
2942 spin_lock_irq(&mdev->tconn->req_lock);
2943 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
2945 spin_unlock_irq(&mdev->tconn->req_lock);
2953 mutex_unlock(mdev->state_mutex);
2955 drbd_adm_finish(info, retcode);
2959 static enum drbd_ret_code
2960 drbd_check_conn_name(const char *name)
2962 if (!name || !name[0]) {
2963 drbd_msg_put_info("connection name missing");
2964 return ERR_MANDATORY_TAG;
2966 /* if we want to use these in sysfs/configfs/debugfs some day,
2967 * we must not allow slashes */
2968 if (strchr(name, '/')) {
2969 drbd_msg_put_info("invalid connection name");
2970 return ERR_INVALID_REQUEST;
2975 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
2977 enum drbd_ret_code retcode;
2979 retcode = drbd_adm_prepare(skb, info, 0);
2980 if (!adm_ctx.reply_skb)
2982 if (retcode != NO_ERROR)
2985 retcode = drbd_check_conn_name(adm_ctx.conn_name);
2986 if (retcode != NO_ERROR)
2989 if (adm_ctx.tconn) {
2990 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
2991 retcode = ERR_INVALID_REQUEST;
2992 drbd_msg_put_info("resource exists");
2994 /* else: still NO_ERROR */
2998 if (!conn_create(adm_ctx.conn_name))
2999 retcode = ERR_NOMEM;
3001 drbd_adm_finish(info, retcode);
3005 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
3007 struct drbd_genlmsghdr *dh = info->userhdr;
3008 enum drbd_ret_code retcode;
3010 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3011 if (!adm_ctx.reply_skb)
3013 if (retcode != NO_ERROR)
3016 /* FIXME drop minor_count parameter, limit to MINORMASK */
3017 if (dh->minor >= minor_count) {
3018 drbd_msg_put_info("requested minor out of range");
3019 retcode = ERR_INVALID_REQUEST;
3022 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3023 drbd_msg_put_info("requested volume id out of range");
3024 retcode = ERR_INVALID_REQUEST;
3028 /* drbd_adm_prepare made sure already
3029 * that mdev->tconn and mdev->vnr match the request. */
3031 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3032 retcode = ERR_MINOR_EXISTS;
3033 /* else: still NO_ERROR */
3037 retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3039 drbd_adm_finish(info, retcode);
3043 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3045 if (mdev->state.disk == D_DISKLESS &&
3046 /* no need to be mdev->state.conn == C_STANDALONE &&
3047 * we may want to delete a minor from a live replication group.
3049 mdev->state.role == R_SECONDARY) {
3050 idr_remove(&mdev->tconn->volumes, mdev->vnr);
3051 idr_remove(&minors, mdev_to_minor(mdev));
3052 del_gendisk(mdev->vdisk);
3054 kref_put(&mdev->kref, &drbd_minor_destroy);
3057 return ERR_MINOR_CONFIGURED;
3060 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3062 enum drbd_ret_code retcode;
3064 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3065 if (!adm_ctx.reply_skb)
3067 if (retcode != NO_ERROR)
3070 retcode = adm_delete_minor(adm_ctx.mdev);
3072 drbd_adm_finish(info, retcode);
3076 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3078 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3079 struct drbd_conf *mdev;
3082 retcode = drbd_adm_prepare(skb, info, 0);
3083 if (!adm_ctx.reply_skb)
3085 if (retcode != NO_ERROR)
3088 if (!adm_ctx.tconn) {
3089 retcode = ERR_RES_NOT_KNOWN;
3094 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3095 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3096 if (retcode < SS_SUCCESS) {
3097 drbd_msg_put_info("failed to demote");
3102 retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3103 if (retcode < SS_SUCCESS) {
3104 drbd_msg_put_info("failed to disconnect");
3109 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3110 retcode = adm_detach(mdev);
3111 if (retcode < SS_SUCCESS) {
3112 drbd_msg_put_info("failed to detach");
3117 /* If we reach this, all volumes (of this tconn) are Secondary,
3118 * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3119 * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3120 drbd_thread_stop(&adm_ctx.tconn->worker);
3122 /* Now, nothing can fail anymore */
3124 /* delete volumes */
3125 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3126 retcode = adm_delete_minor(mdev);
3127 if (retcode != NO_ERROR) {
3128 /* "can not happen" */
3129 drbd_msg_put_info("failed to delete volume");
3134 /* delete connection */
3135 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3136 list_del_rcu(&adm_ctx.tconn->all_tconn);
3138 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3142 /* "can not happen" */
3143 retcode = ERR_RES_IN_USE;
3144 drbd_msg_put_info("failed to delete connection");
3148 drbd_adm_finish(info, retcode);
3152 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
3154 enum drbd_ret_code retcode;
3156 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONN);
3157 if (!adm_ctx.reply_skb)
3159 if (retcode != NO_ERROR)
3162 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3163 list_del_rcu(&adm_ctx.tconn->all_tconn);
3165 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3169 retcode = ERR_RES_IN_USE;
3172 if (retcode == NO_ERROR)
3173 drbd_thread_stop(&adm_ctx.tconn->worker);
3175 drbd_adm_finish(info, retcode);
3179 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3181 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3182 struct sk_buff *msg;
3183 struct drbd_genlmsghdr *d_out;
3187 seq = atomic_inc_return(&drbd_genl_seq);
3188 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3193 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3194 if (!d_out) /* cannot happen, but anyways. */
3195 goto nla_put_failure;
3196 d_out->minor = mdev_to_minor(mdev);
3197 d_out->ret_code = NO_ERROR;
3199 if (nla_put_status_info(msg, mdev, sib))
3200 goto nla_put_failure;
3201 genlmsg_end(msg, d_out);
3202 err = drbd_genl_multicast_events(msg, 0);
3203 /* msg has been consumed or freed in netlink_broadcast() */
3204 if (err && err != -ESRCH)
3212 dev_err(DEV, "Error %d while broadcasting event. "
3213 "Event seq:%u sib_reason:%u\n",
3214 err, seq, sib->sib_reason);