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! */
97 struct nlattr *my_addr;
98 struct nlattr *peer_addr;
101 struct sk_buff *reply_skb;
102 /* pointer into reply buffer */
103 struct drbd_genlmsghdr *reply_dh;
104 /* resolved from attributes, if possible */
105 struct drbd_conf *mdev;
106 struct drbd_tconn *tconn;
109 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
111 genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
112 if (genlmsg_reply(skb, info))
113 printk(KERN_ERR "drbd: error sending genl reply\n");
116 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
117 * reason it could fail was no space in skb, and there are 4k available. */
118 int drbd_msg_put_info(const char *info)
120 struct sk_buff *skb = adm_ctx.reply_skb;
124 if (!info || !info[0])
127 nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
131 err = nla_put_string(skb, T_info_text, info);
133 nla_nest_cancel(skb, nla);
136 nla_nest_end(skb, nla);
140 /* This would be a good candidate for a "pre_doit" hook,
141 * and per-family private info->pointers.
142 * But we need to stay compatible with older kernels.
143 * If it returns successfully, adm_ctx members are valid.
145 #define DRBD_ADM_NEED_MINOR 1
146 #define DRBD_ADM_NEED_RESOURCE 2
147 #define DRBD_ADM_NEED_CONNECTION 4
148 static int drbd_adm_prepare(struct sk_buff *skb, struct genl_info *info,
151 struct drbd_genlmsghdr *d_in = info->userhdr;
152 const u8 cmd = info->genlhdr->cmd;
155 memset(&adm_ctx, 0, sizeof(adm_ctx));
157 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
158 if (cmd != DRBD_ADM_GET_STATUS && !capable(CAP_NET_ADMIN))
161 adm_ctx.reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
162 if (!adm_ctx.reply_skb) {
167 adm_ctx.reply_dh = genlmsg_put_reply(adm_ctx.reply_skb,
168 info, &drbd_genl_family, 0, cmd);
169 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
171 if (!adm_ctx.reply_dh) {
176 adm_ctx.reply_dh->minor = d_in->minor;
177 adm_ctx.reply_dh->ret_code = NO_ERROR;
179 adm_ctx.volume = VOLUME_UNSPECIFIED;
180 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
182 /* parse and validate only */
183 err = drbd_cfg_context_from_attrs(NULL, info);
187 /* It was present, and valid,
188 * copy it over to the reply skb. */
189 err = nla_put_nohdr(adm_ctx.reply_skb,
190 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
191 info->attrs[DRBD_NLA_CFG_CONTEXT]);
195 /* and assign stuff to the global adm_ctx */
196 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
198 adm_ctx.volume = nla_get_u32(nla);
199 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
201 adm_ctx.resource_name = nla_data(nla);
202 adm_ctx.my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)];
203 adm_ctx.peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)];
204 if ((adm_ctx.my_addr &&
205 nla_len(adm_ctx.my_addr) > sizeof(adm_ctx.tconn->my_addr)) ||
206 (adm_ctx.peer_addr &&
207 nla_len(adm_ctx.peer_addr) > sizeof(adm_ctx.tconn->peer_addr))) {
213 adm_ctx.minor = d_in->minor;
214 adm_ctx.mdev = minor_to_mdev(d_in->minor);
215 adm_ctx.tconn = conn_get_by_name(adm_ctx.resource_name);
217 if (!adm_ctx.mdev && (flags & DRBD_ADM_NEED_MINOR)) {
218 drbd_msg_put_info("unknown minor");
219 return ERR_MINOR_INVALID;
221 if (!adm_ctx.tconn && (flags & DRBD_ADM_NEED_RESOURCE)) {
222 drbd_msg_put_info("unknown resource");
223 return ERR_INVALID_REQUEST;
226 if (flags & DRBD_ADM_NEED_CONNECTION) {
227 if (adm_ctx.tconn && !(flags & DRBD_ADM_NEED_RESOURCE)) {
228 drbd_msg_put_info("no resource name expected");
229 return ERR_INVALID_REQUEST;
232 drbd_msg_put_info("no minor number expected");
233 return ERR_INVALID_REQUEST;
235 if (adm_ctx.my_addr && adm_ctx.peer_addr)
236 adm_ctx.tconn = conn_get_by_addrs(nla_data(adm_ctx.my_addr),
237 nla_len(adm_ctx.my_addr),
238 nla_data(adm_ctx.peer_addr),
239 nla_len(adm_ctx.peer_addr));
240 if (!adm_ctx.tconn) {
241 drbd_msg_put_info("unknown connection");
242 return ERR_INVALID_REQUEST;
246 /* some more paranoia, if the request was over-determined */
247 if (adm_ctx.mdev && adm_ctx.tconn &&
248 adm_ctx.mdev->tconn != adm_ctx.tconn) {
249 pr_warning("request: minor=%u, resource=%s; but that minor belongs to connection %s\n",
250 adm_ctx.minor, adm_ctx.resource_name,
251 adm_ctx.mdev->tconn->name);
252 drbd_msg_put_info("minor exists in different resource");
253 return ERR_INVALID_REQUEST;
256 adm_ctx.volume != VOLUME_UNSPECIFIED &&
257 adm_ctx.volume != adm_ctx.mdev->vnr) {
258 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
259 adm_ctx.minor, adm_ctx.volume,
260 adm_ctx.mdev->vnr, adm_ctx.mdev->tconn->name);
261 drbd_msg_put_info("minor exists as different volume");
262 return ERR_INVALID_REQUEST;
268 nlmsg_free(adm_ctx.reply_skb);
269 adm_ctx.reply_skb = NULL;
273 static int drbd_adm_finish(struct genl_info *info, int retcode)
276 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
277 adm_ctx.tconn = NULL;
280 if (!adm_ctx.reply_skb)
283 adm_ctx.reply_dh->ret_code = retcode;
284 drbd_adm_send_reply(adm_ctx.reply_skb, info);
288 static void setup_khelper_env(struct drbd_tconn *tconn, char **envp)
292 /* FIXME: A future version will not allow this case. */
293 if (tconn->my_addr_len == 0 || tconn->peer_addr_len == 0)
296 switch (((struct sockaddr *)&tconn->peer_addr)->sa_family) {
299 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
300 &((struct sockaddr_in6 *)&tconn->peer_addr)->sin6_addr);
304 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
305 &((struct sockaddr_in *)&tconn->peer_addr)->sin_addr);
309 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
310 &((struct sockaddr_in *)&tconn->peer_addr)->sin_addr);
312 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
315 int drbd_khelper(struct drbd_conf *mdev, char *cmd)
317 char *envp[] = { "HOME=/",
319 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
320 (char[20]) { }, /* address family */
321 (char[60]) { }, /* address */
324 char *argv[] = {usermode_helper, cmd, mb, NULL };
325 struct drbd_tconn *tconn = mdev->tconn;
329 if (current == tconn->worker.task)
330 set_bit(CALLBACK_PENDING, &tconn->flags);
332 snprintf(mb, 12, "minor-%d", mdev_to_minor(mdev));
333 setup_khelper_env(tconn, envp);
335 /* The helper may take some time.
336 * write out any unsynced meta data changes now */
339 dev_info(DEV, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
340 sib.sib_reason = SIB_HELPER_PRE;
341 sib.helper_name = cmd;
342 drbd_bcast_event(mdev, &sib);
343 ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
345 dev_warn(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
346 usermode_helper, cmd, mb,
347 (ret >> 8) & 0xff, ret);
349 dev_info(DEV, "helper command: %s %s %s exit code %u (0x%x)\n",
350 usermode_helper, cmd, mb,
351 (ret >> 8) & 0xff, ret);
352 sib.sib_reason = SIB_HELPER_POST;
353 sib.helper_exit_code = ret;
354 drbd_bcast_event(mdev, &sib);
356 if (current == tconn->worker.task)
357 clear_bit(CALLBACK_PENDING, &tconn->flags);
359 if (ret < 0) /* Ignore any ERRNOs we got. */
365 int conn_khelper(struct drbd_tconn *tconn, char *cmd)
367 char *envp[] = { "HOME=/",
369 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
370 (char[20]) { }, /* address family */
371 (char[60]) { }, /* address */
373 char *argv[] = {usermode_helper, cmd, tconn->name, NULL };
376 setup_khelper_env(tconn, envp);
379 conn_info(tconn, "helper command: %s %s %s\n", usermode_helper, cmd, tconn->name);
380 /* TODO: conn_bcast_event() ?? */
382 ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
384 conn_warn(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
385 usermode_helper, cmd, tconn->name,
386 (ret >> 8) & 0xff, ret);
388 conn_info(tconn, "helper command: %s %s %s exit code %u (0x%x)\n",
389 usermode_helper, cmd, tconn->name,
390 (ret >> 8) & 0xff, ret);
391 /* TODO: conn_bcast_event() ?? */
393 if (ret < 0) /* Ignore any ERRNOs we got. */
399 static enum drbd_fencing_p highest_fencing_policy(struct drbd_tconn *tconn)
401 enum drbd_fencing_p fp = FP_NOT_AVAIL;
402 struct drbd_conf *mdev;
406 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
407 if (get_ldev_if_state(mdev, D_CONSISTENT)) {
408 fp = max_t(enum drbd_fencing_p, fp,
409 rcu_dereference(mdev->ldev->disk_conf)->fencing);
418 bool conn_try_outdate_peer(struct drbd_tconn *tconn)
420 union drbd_state mask = { };
421 union drbd_state val = { };
422 enum drbd_fencing_p fp;
426 if (tconn->cstate >= C_WF_REPORT_PARAMS) {
427 conn_err(tconn, "Expected cstate < C_WF_REPORT_PARAMS\n");
431 fp = highest_fencing_policy(tconn);
434 conn_warn(tconn, "Not fencing peer, I'm not even Consistent myself.\n");
441 r = conn_khelper(tconn, "fence-peer");
443 switch ((r>>8) & 0xff) {
444 case 3: /* peer is inconsistent */
445 ex_to_string = "peer is inconsistent or worse";
447 val.pdsk = D_INCONSISTENT;
449 case 4: /* peer got outdated, or was already outdated */
450 ex_to_string = "peer was fenced";
452 val.pdsk = D_OUTDATED;
454 case 5: /* peer was down */
455 if (conn_highest_disk(tconn) == D_UP_TO_DATE) {
456 /* we will(have) create(d) a new UUID anyways... */
457 ex_to_string = "peer is unreachable, assumed to be dead";
459 val.pdsk = D_OUTDATED;
461 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
464 case 6: /* Peer is primary, voluntarily outdate myself.
465 * This is useful when an unconnected R_SECONDARY is asked to
466 * become R_PRIMARY, but finds the other peer being active. */
467 ex_to_string = "peer is active";
468 conn_warn(tconn, "Peer is primary, outdating myself.\n");
470 val.disk = D_OUTDATED;
473 if (fp != FP_STONITH)
474 conn_err(tconn, "fence-peer() = 7 && fencing != Stonith !!!\n");
475 ex_to_string = "peer was stonithed";
477 val.pdsk = D_OUTDATED;
480 /* The script is broken ... */
481 conn_err(tconn, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
482 return false; /* Eventually leave IO frozen */
485 conn_info(tconn, "fence-peer helper returned %d (%s)\n",
486 (r>>8) & 0xff, ex_to_string);
491 conn_request_state(tconn, mask, val, CS_VERBOSE);
492 here, because we might were able to re-establish the connection in the
494 spin_lock_irq(&tconn->req_lock);
495 if (tconn->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &tconn->flags))
496 _conn_request_state(tconn, mask, val, CS_VERBOSE);
497 spin_unlock_irq(&tconn->req_lock);
499 return conn_highest_pdsk(tconn) <= D_OUTDATED;
502 static int _try_outdate_peer_async(void *data)
504 struct drbd_tconn *tconn = (struct drbd_tconn *)data;
506 conn_try_outdate_peer(tconn);
508 kref_put(&tconn->kref, &conn_destroy);
512 void conn_try_outdate_peer_async(struct drbd_tconn *tconn)
514 struct task_struct *opa;
516 kref_get(&tconn->kref);
517 opa = kthread_run(_try_outdate_peer_async, tconn, "drbd_async_h");
519 conn_err(tconn, "out of mem, failed to invoke fence-peer helper\n");
520 kref_put(&tconn->kref, &conn_destroy);
525 drbd_set_role(struct drbd_conf *mdev, enum drbd_role new_role, int force)
527 const int max_tries = 4;
528 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
532 union drbd_state mask, val;
534 if (new_role == R_PRIMARY)
535 request_ping(mdev->tconn); /* Detect a dead peer ASAP */
537 mutex_lock(mdev->state_mutex);
539 mask.i = 0; mask.role = R_MASK;
540 val.i = 0; val.role = new_role;
542 while (try++ < max_tries) {
543 rv = _drbd_request_state(mdev, mask, val, CS_WAIT_COMPLETE);
545 /* in case we first succeeded to outdate,
546 * but now suddenly could establish a connection */
547 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
553 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
554 (mdev->state.disk < D_UP_TO_DATE &&
555 mdev->state.disk >= D_INCONSISTENT)) {
557 val.disk = D_UP_TO_DATE;
562 if (rv == SS_NO_UP_TO_DATE_DISK &&
563 mdev->state.disk == D_CONSISTENT && mask.pdsk == 0) {
564 D_ASSERT(mdev->state.pdsk == D_UNKNOWN);
566 if (conn_try_outdate_peer(mdev->tconn)) {
567 val.disk = D_UP_TO_DATE;
573 if (rv == SS_NOTHING_TO_DO)
575 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
576 if (!conn_try_outdate_peer(mdev->tconn) && force) {
577 dev_warn(DEV, "Forced into split brain situation!\n");
579 val.pdsk = D_OUTDATED;
584 if (rv == SS_TWO_PRIMARIES) {
585 /* Maybe the peer is detected as dead very soon...
586 retry at most once more in this case. */
589 nc = rcu_dereference(mdev->tconn->net_conf);
590 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
592 schedule_timeout_interruptible(timeo);
597 if (rv < SS_SUCCESS) {
598 rv = _drbd_request_state(mdev, mask, val,
599 CS_VERBOSE + CS_WAIT_COMPLETE);
610 dev_warn(DEV, "Forced to consider local data as UpToDate!\n");
612 /* Wait until nothing is on the fly :) */
613 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_pending_cnt) == 0);
615 /* FIXME also wait for all pending P_BARRIER_ACK? */
617 if (new_role == R_SECONDARY) {
618 set_disk_ro(mdev->vdisk, true);
619 if (get_ldev(mdev)) {
620 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
624 mutex_lock(&mdev->tconn->conf_update);
625 nc = mdev->tconn->net_conf;
627 nc->discard_my_data = 0; /* without copy; single bit op is atomic */
628 mutex_unlock(&mdev->tconn->conf_update);
630 set_disk_ro(mdev->vdisk, false);
631 if (get_ldev(mdev)) {
632 if (((mdev->state.conn < C_CONNECTED ||
633 mdev->state.pdsk <= D_FAILED)
634 && mdev->ldev->md.uuid[UI_BITMAP] == 0) || forced)
635 drbd_uuid_new_current(mdev);
637 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
642 /* writeout of activity log covered areas of the bitmap
643 * to stable storage done in after state change already */
645 if (mdev->state.conn >= C_WF_REPORT_PARAMS) {
646 /* if this was forced, we should consider sync */
648 drbd_send_uuids(mdev);
649 drbd_send_current_state(mdev);
654 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
656 mutex_unlock(mdev->state_mutex);
660 static const char *from_attrs_err_to_txt(int err)
662 return err == -ENOMSG ? "required attribute missing" :
663 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
664 err == -EEXIST ? "can not change invariant setting" :
665 "invalid attribute value";
668 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
670 struct set_role_parms parms;
672 enum drbd_ret_code retcode;
674 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
675 if (!adm_ctx.reply_skb)
677 if (retcode != NO_ERROR)
680 memset(&parms, 0, sizeof(parms));
681 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
682 err = set_role_parms_from_attrs(&parms, info);
684 retcode = ERR_MANDATORY_TAG;
685 drbd_msg_put_info(from_attrs_err_to_txt(err));
690 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
691 retcode = drbd_set_role(adm_ctx.mdev, R_PRIMARY, parms.assume_uptodate);
693 retcode = drbd_set_role(adm_ctx.mdev, R_SECONDARY, 0);
695 drbd_adm_finish(info, retcode);
699 /* Initializes the md.*_offset members, so we are able to find
700 * the on disk meta data.
702 * We currently have two possible layouts:
704 * |----------- md_size_sect ------------------|
705 * [ 4k superblock ][ activity log ][ Bitmap ]
707 * | bm_offset = al_offset + X |
708 * ==> bitmap sectors = md_size_sect - bm_offset
711 * |----------- md_size_sect ------------------|
712 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
714 * | bm_offset = al_offset - Y |
715 * ==> bitmap sectors = Y = al_offset - bm_offset
717 * Activity log size used to be fixed 32kB,
718 * but is about to become configurable.
720 static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
721 struct drbd_backing_dev *bdev)
723 sector_t md_size_sect = 0;
724 unsigned int al_size_sect = bdev->md.al_size_4k * 8;
726 bdev->md.md_offset = drbd_md_ss(bdev);
728 switch (bdev->md.meta_dev_idx) {
730 /* v07 style fixed size indexed meta data */
731 bdev->md.md_size_sect = MD_128MB_SECT;
732 bdev->md.al_offset = MD_4kB_SECT;
733 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
735 case DRBD_MD_INDEX_FLEX_EXT:
736 /* just occupy the full device; unit: sectors */
737 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
738 bdev->md.al_offset = MD_4kB_SECT;
739 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
741 case DRBD_MD_INDEX_INTERNAL:
742 case DRBD_MD_INDEX_FLEX_INT:
743 /* al size is still fixed */
744 bdev->md.al_offset = -al_size_sect;
745 /* we need (slightly less than) ~ this much bitmap sectors: */
746 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
747 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
748 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
749 md_size_sect = ALIGN(md_size_sect, 8);
751 /* plus the "drbd meta data super block",
752 * and the activity log; */
753 md_size_sect += MD_4kB_SECT + al_size_sect;
755 bdev->md.md_size_sect = md_size_sect;
756 /* bitmap offset is adjusted by 'super' block size */
757 bdev->md.bm_offset = -md_size_sect + MD_4kB_SECT;
762 /* input size is expected to be in KB */
763 char *ppsize(char *buf, unsigned long long size)
765 /* Needs 9 bytes at max including trailing NUL:
766 * -1ULL ==> "16384 EB" */
767 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
769 while (size >= 10000 && base < sizeof(units)-1) {
771 size = (size >> 10) + !!(size & (1<<9));
774 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
779 /* there is still a theoretical deadlock when called from receiver
780 * on an D_INCONSISTENT R_PRIMARY:
781 * remote READ does inc_ap_bio, receiver would need to receive answer
782 * packet from remote to dec_ap_bio again.
783 * receiver receive_sizes(), comes here,
784 * waits for ap_bio_cnt == 0. -> deadlock.
785 * but this cannot happen, actually, because:
786 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
787 * (not connected, or bad/no disk on peer):
788 * see drbd_fail_request_early, ap_bio_cnt is zero.
789 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
790 * peer may not initiate a resize.
792 /* Note these are not to be confused with
793 * drbd_adm_suspend_io/drbd_adm_resume_io,
794 * which are (sub) state changes triggered by admin (drbdsetup),
795 * and can be long lived.
796 * This changes an mdev->flag, is triggered by drbd internals,
797 * and should be short-lived. */
798 void drbd_suspend_io(struct drbd_conf *mdev)
800 set_bit(SUSPEND_IO, &mdev->flags);
801 if (drbd_suspended(mdev))
803 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
806 void drbd_resume_io(struct drbd_conf *mdev)
808 clear_bit(SUSPEND_IO, &mdev->flags);
809 wake_up(&mdev->misc_wait);
813 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
814 * @mdev: DRBD device.
816 * Returns 0 on success, negative return values indicate errors.
817 * You should call drbd_md_sync() after calling this function.
819 enum determine_dev_size drbd_determine_dev_size(struct drbd_conf *mdev, enum dds_flags flags) __must_hold(local)
821 sector_t prev_first_sect, prev_size; /* previous meta location */
822 sector_t la_size_sect, u_size;
826 int md_moved, la_size_changed;
827 enum determine_dev_size rv = unchanged;
830 * application request passes inc_ap_bio,
831 * but then cannot get an AL-reference.
832 * this function later may wait on ap_bio_cnt == 0. -> deadlock.
835 * Suspend IO right here.
836 * still lock the act_log to not trigger ASSERTs there.
838 drbd_suspend_io(mdev);
840 /* no wait necessary anymore, actually we could assert that */
841 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
843 prev_first_sect = drbd_md_first_sector(mdev->ldev);
844 prev_size = mdev->ldev->md.md_size_sect;
845 la_size_sect = mdev->ldev->md.la_size_sect;
847 /* TODO: should only be some assert here, not (re)init... */
848 drbd_md_set_sector_offsets(mdev, mdev->ldev);
851 u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
853 size = drbd_new_dev_size(mdev, mdev->ldev, u_size, flags & DDSF_FORCED);
855 if (drbd_get_capacity(mdev->this_bdev) != size ||
856 drbd_bm_capacity(mdev) != size) {
858 err = drbd_bm_resize(mdev, size, !(flags & DDSF_NO_RESYNC));
860 /* currently there is only one error: ENOMEM! */
861 size = drbd_bm_capacity(mdev)>>1;
863 dev_err(DEV, "OUT OF MEMORY! "
864 "Could not allocate bitmap!\n");
866 dev_err(DEV, "BM resizing failed. "
867 "Leaving size unchanged at size = %lu KB\n",
868 (unsigned long)size);
872 /* racy, see comments above. */
873 drbd_set_my_capacity(mdev, size);
874 mdev->ldev->md.la_size_sect = size;
875 dev_info(DEV, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
876 (unsigned long long)size>>1);
878 if (rv == dev_size_error)
881 la_size_changed = (la_size_sect != mdev->ldev->md.la_size_sect);
883 md_moved = prev_first_sect != drbd_md_first_sector(mdev->ldev)
884 || prev_size != mdev->ldev->md.md_size_sect;
886 if (la_size_changed || md_moved) {
889 drbd_al_shrink(mdev); /* All extents inactive. */
890 dev_info(DEV, "Writing the whole bitmap, %s\n",
891 la_size_changed && md_moved ? "size changed and md moved" :
892 la_size_changed ? "size changed" : "md moved");
893 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
894 err = drbd_bitmap_io(mdev, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
895 "size changed", BM_LOCKED_MASK);
900 drbd_md_mark_dirty(mdev);
903 if (size > la_size_sect)
905 if (size < la_size_sect)
908 lc_unlock(mdev->act_log);
909 wake_up(&mdev->al_wait);
910 drbd_resume_io(mdev);
916 drbd_new_dev_size(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
917 sector_t u_size, int assume_peer_has_space)
919 sector_t p_size = mdev->p_size; /* partner's disk size. */
920 sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
921 sector_t m_size; /* my size */
924 m_size = drbd_get_max_capacity(bdev);
926 if (mdev->state.conn < C_CONNECTED && assume_peer_has_space) {
927 dev_warn(DEV, "Resize while not connected was forced by the user!\n");
931 if (p_size && m_size) {
932 size = min_t(sector_t, p_size, m_size);
936 if (m_size && m_size < size)
938 if (p_size && p_size < size)
949 dev_err(DEV, "Both nodes diskless!\n");
953 dev_err(DEV, "Requested disk size is too big (%lu > %lu)\n",
954 (unsigned long)u_size>>1, (unsigned long)size>>1);
963 * drbd_check_al_size() - Ensures that the AL is of the right size
964 * @mdev: DRBD device.
966 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
967 * failed, and 0 on success. You should call drbd_md_sync() after you called
970 static int drbd_check_al_size(struct drbd_conf *mdev, struct disk_conf *dc)
972 struct lru_cache *n, *t;
973 struct lc_element *e;
978 mdev->act_log->nr_elements == dc->al_extents)
983 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
984 dc->al_extents, sizeof(struct lc_element), 0);
987 dev_err(DEV, "Cannot allocate act_log lru!\n");
990 spin_lock_irq(&mdev->al_lock);
992 for (i = 0; i < t->nr_elements; i++) {
993 e = lc_element_by_index(t, i);
995 dev_err(DEV, "refcnt(%d)==%d\n",
996 e->lc_number, e->refcnt);
1002 spin_unlock_irq(&mdev->al_lock);
1004 dev_err(DEV, "Activity log still in use!\n");
1011 drbd_md_mark_dirty(mdev); /* we changed mdev->act_log->nr_elemens */
1015 static void drbd_setup_queue_param(struct drbd_conf *mdev, unsigned int max_bio_size)
1017 struct request_queue * const q = mdev->rq_queue;
1018 unsigned int max_hw_sectors = max_bio_size >> 9;
1019 unsigned int max_segments = 0;
1021 if (get_ldev_if_state(mdev, D_ATTACHING)) {
1022 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1024 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1026 max_segments = rcu_dereference(mdev->ldev->disk_conf)->max_bio_bvecs;
1031 blk_queue_logical_block_size(q, 512);
1032 blk_queue_max_hw_sectors(q, max_hw_sectors);
1033 /* This is the workaround for "bio would need to, but cannot, be split" */
1034 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1035 blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
1037 if (get_ldev_if_state(mdev, D_ATTACHING)) {
1038 struct request_queue * const b = mdev->ldev->backing_bdev->bd_disk->queue;
1040 blk_queue_stack_limits(q, b);
1042 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1043 dev_info(DEV, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1044 q->backing_dev_info.ra_pages,
1045 b->backing_dev_info.ra_pages);
1046 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1052 void drbd_reconsider_max_bio_size(struct drbd_conf *mdev)
1054 unsigned int now, new, local, peer;
1056 now = queue_max_hw_sectors(mdev->rq_queue) << 9;
1057 local = mdev->local_max_bio_size; /* Eventually last known value, from volatile memory */
1058 peer = mdev->peer_max_bio_size; /* Eventually last known value, from meta data */
1060 if (get_ldev_if_state(mdev, D_ATTACHING)) {
1061 local = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1062 mdev->local_max_bio_size = local;
1065 local = min(local, DRBD_MAX_BIO_SIZE);
1067 /* We may ignore peer limits if the peer is modern enough.
1068 Because new from 8.3.8 onwards the peer can use multiple
1069 BIOs for a single peer_request */
1070 if (mdev->state.conn >= C_CONNECTED) {
1071 if (mdev->tconn->agreed_pro_version < 94)
1072 peer = min( mdev->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1073 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1074 else if (mdev->tconn->agreed_pro_version == 94)
1075 peer = DRBD_MAX_SIZE_H80_PACKET;
1076 else if (mdev->tconn->agreed_pro_version < 100)
1077 peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */
1079 peer = DRBD_MAX_BIO_SIZE;
1082 new = min(local, peer);
1084 if (mdev->state.role == R_PRIMARY && new < now)
1085 dev_err(DEV, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1088 dev_info(DEV, "max BIO size = %u\n", new);
1090 drbd_setup_queue_param(mdev, new);
1093 /* Starts the worker thread */
1094 static void conn_reconfig_start(struct drbd_tconn *tconn)
1096 drbd_thread_start(&tconn->worker);
1097 conn_flush_workqueue(tconn);
1100 /* if still unconfigured, stops worker again. */
1101 static void conn_reconfig_done(struct drbd_tconn *tconn)
1104 spin_lock_irq(&tconn->req_lock);
1105 stop_threads = conn_all_vols_unconf(tconn) &&
1106 tconn->cstate == C_STANDALONE;
1107 spin_unlock_irq(&tconn->req_lock);
1109 /* asender is implicitly stopped by receiver
1110 * in conn_disconnect() */
1111 drbd_thread_stop(&tconn->receiver);
1112 drbd_thread_stop(&tconn->worker);
1116 /* Make sure IO is suspended before calling this function(). */
1117 static void drbd_suspend_al(struct drbd_conf *mdev)
1121 if (!lc_try_lock(mdev->act_log)) {
1122 dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
1126 drbd_al_shrink(mdev);
1127 spin_lock_irq(&mdev->tconn->req_lock);
1128 if (mdev->state.conn < C_CONNECTED)
1129 s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
1130 spin_unlock_irq(&mdev->tconn->req_lock);
1131 lc_unlock(mdev->act_log);
1134 dev_info(DEV, "Suspended AL updates\n");
1138 static bool should_set_defaults(struct genl_info *info)
1140 unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1141 return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1144 static void enforce_disk_conf_limits(struct disk_conf *dc)
1146 if (dc->al_extents < DRBD_AL_EXTENTS_MIN)
1147 dc->al_extents = DRBD_AL_EXTENTS_MIN;
1148 if (dc->al_extents > DRBD_AL_EXTENTS_MAX)
1149 dc->al_extents = DRBD_AL_EXTENTS_MAX;
1151 if (dc->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1152 dc->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1155 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1157 enum drbd_ret_code retcode;
1158 struct drbd_conf *mdev;
1159 struct disk_conf *new_disk_conf, *old_disk_conf;
1160 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1163 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1164 if (!adm_ctx.reply_skb)
1166 if (retcode != NO_ERROR)
1169 mdev = adm_ctx.mdev;
1171 /* we also need a disk
1172 * to change the options on */
1173 if (!get_ldev(mdev)) {
1174 retcode = ERR_NO_DISK;
1178 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1179 if (!new_disk_conf) {
1180 retcode = ERR_NOMEM;
1184 mutex_lock(&mdev->tconn->conf_update);
1185 old_disk_conf = mdev->ldev->disk_conf;
1186 *new_disk_conf = *old_disk_conf;
1187 if (should_set_defaults(info))
1188 set_disk_conf_defaults(new_disk_conf);
1190 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1191 if (err && err != -ENOMSG) {
1192 retcode = ERR_MANDATORY_TAG;
1193 drbd_msg_put_info(from_attrs_err_to_txt(err));
1196 if (!expect(new_disk_conf->resync_rate >= 1))
1197 new_disk_conf->resync_rate = 1;
1199 enforce_disk_conf_limits(new_disk_conf);
1201 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1202 if (fifo_size != mdev->rs_plan_s->size) {
1203 new_plan = fifo_alloc(fifo_size);
1205 dev_err(DEV, "kmalloc of fifo_buffer failed");
1206 retcode = ERR_NOMEM;
1211 drbd_suspend_io(mdev);
1212 wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
1213 drbd_al_shrink(mdev);
1214 err = drbd_check_al_size(mdev, new_disk_conf);
1215 lc_unlock(mdev->act_log);
1216 wake_up(&mdev->al_wait);
1217 drbd_resume_io(mdev);
1220 retcode = ERR_NOMEM;
1224 write_lock_irq(&global_state_lock);
1225 retcode = drbd_resync_after_valid(mdev, new_disk_conf->resync_after);
1226 if (retcode == NO_ERROR) {
1227 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
1228 drbd_resync_after_changed(mdev);
1230 write_unlock_irq(&global_state_lock);
1232 if (retcode != NO_ERROR)
1236 old_plan = mdev->rs_plan_s;
1237 rcu_assign_pointer(mdev->rs_plan_s, new_plan);
1240 mutex_unlock(&mdev->tconn->conf_update);
1242 if (new_disk_conf->al_updates)
1243 mdev->ldev->md.flags &= ~MDF_AL_DISABLED;
1245 mdev->ldev->md.flags |= MDF_AL_DISABLED;
1247 if (new_disk_conf->md_flushes)
1248 clear_bit(MD_NO_FUA, &mdev->flags);
1250 set_bit(MD_NO_FUA, &mdev->flags);
1252 drbd_bump_write_ordering(mdev->tconn, WO_bdev_flush);
1256 if (mdev->state.conn >= C_CONNECTED)
1257 drbd_send_sync_param(mdev);
1260 kfree(old_disk_conf);
1262 mod_timer(&mdev->request_timer, jiffies + HZ);
1266 mutex_unlock(&mdev->tconn->conf_update);
1268 kfree(new_disk_conf);
1273 drbd_adm_finish(info, retcode);
1277 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1279 struct drbd_conf *mdev;
1281 enum drbd_ret_code retcode;
1282 enum determine_dev_size dd;
1283 sector_t max_possible_sectors;
1284 sector_t min_md_device_sectors;
1285 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1286 struct disk_conf *new_disk_conf = NULL;
1287 struct block_device *bdev;
1288 struct lru_cache *resync_lru = NULL;
1289 struct fifo_buffer *new_plan = NULL;
1290 union drbd_state ns, os;
1291 enum drbd_state_rv rv;
1292 struct net_conf *nc;
1294 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1295 if (!adm_ctx.reply_skb)
1297 if (retcode != NO_ERROR)
1300 mdev = adm_ctx.mdev;
1301 conn_reconfig_start(mdev->tconn);
1303 /* if you want to reconfigure, please tear down first */
1304 if (mdev->state.disk > D_DISKLESS) {
1305 retcode = ERR_DISK_CONFIGURED;
1308 /* It may just now have detached because of IO error. Make sure
1309 * drbd_ldev_destroy is done already, we may end up here very fast,
1310 * e.g. if someone calls attach from the on-io-error handler,
1311 * to realize a "hot spare" feature (not that I'd recommend that) */
1312 wait_event(mdev->misc_wait, !atomic_read(&mdev->local_cnt));
1314 /* make sure there is no leftover from previous force-detach attempts */
1315 clear_bit(FORCE_DETACH, &mdev->flags);
1316 clear_bit(WAS_IO_ERROR, &mdev->flags);
1317 clear_bit(WAS_READ_ERROR, &mdev->flags);
1319 /* and no leftover from previously aborted resync or verify, either */
1321 mdev->rs_failed = 0;
1322 atomic_set(&mdev->rs_pending_cnt, 0);
1324 /* allocation not in the IO path, drbdsetup context */
1325 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1327 retcode = ERR_NOMEM;
1330 spin_lock_init(&nbc->md.uuid_lock);
1332 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1333 if (!new_disk_conf) {
1334 retcode = ERR_NOMEM;
1337 nbc->disk_conf = new_disk_conf;
1339 set_disk_conf_defaults(new_disk_conf);
1340 err = disk_conf_from_attrs(new_disk_conf, info);
1342 retcode = ERR_MANDATORY_TAG;
1343 drbd_msg_put_info(from_attrs_err_to_txt(err));
1347 enforce_disk_conf_limits(new_disk_conf);
1349 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1351 retcode = ERR_NOMEM;
1355 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1356 retcode = ERR_MD_IDX_INVALID;
1361 nc = rcu_dereference(mdev->tconn->net_conf);
1363 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1365 retcode = ERR_STONITH_AND_PROT_A;
1371 bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1372 FMODE_READ | FMODE_WRITE | FMODE_EXCL, mdev);
1374 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1376 retcode = ERR_OPEN_DISK;
1379 nbc->backing_bdev = bdev;
1382 * meta_dev_idx >= 0: external fixed size, possibly multiple
1383 * drbd sharing one meta device. TODO in that case, paranoia
1384 * check that [md_bdev, meta_dev_idx] is not yet used by some
1385 * other drbd minor! (if you use drbd.conf + drbdadm, that
1386 * should check it for you already; but if you don't, or
1387 * someone fooled it, we need to double check here)
1389 bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1390 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1391 (new_disk_conf->meta_dev_idx < 0) ?
1392 (void *)mdev : (void *)drbd_m_holder);
1394 dev_err(DEV, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1396 retcode = ERR_OPEN_MD_DISK;
1399 nbc->md_bdev = bdev;
1401 if ((nbc->backing_bdev == nbc->md_bdev) !=
1402 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1403 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1404 retcode = ERR_MD_IDX_INVALID;
1408 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1409 1, 61, sizeof(struct bm_extent),
1410 offsetof(struct bm_extent, lce));
1412 retcode = ERR_NOMEM;
1416 /* Read our meta data super block early.
1417 * This also sets other on-disk offsets. */
1418 retcode = drbd_md_read(mdev, nbc);
1419 if (retcode != NO_ERROR)
1422 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1423 dev_err(DEV, "max capacity %llu smaller than disk size %llu\n",
1424 (unsigned long long) drbd_get_max_capacity(nbc),
1425 (unsigned long long) new_disk_conf->disk_size);
1426 retcode = ERR_DISK_TOO_SMALL;
1430 if (new_disk_conf->meta_dev_idx < 0) {
1431 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1432 /* at least one MB, otherwise it does not make sense */
1433 min_md_device_sectors = (2<<10);
1435 max_possible_sectors = DRBD_MAX_SECTORS;
1436 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1439 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1440 retcode = ERR_MD_DISK_TOO_SMALL;
1441 dev_warn(DEV, "refusing attach: md-device too small, "
1442 "at least %llu sectors needed for this meta-disk type\n",
1443 (unsigned long long) min_md_device_sectors);
1447 /* Make sure the new disk is big enough
1448 * (we may currently be R_PRIMARY with no local disk...) */
1449 if (drbd_get_max_capacity(nbc) <
1450 drbd_get_capacity(mdev->this_bdev)) {
1451 retcode = ERR_DISK_TOO_SMALL;
1455 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1457 if (nbc->known_size > max_possible_sectors) {
1458 dev_warn(DEV, "==> truncating very big lower level device "
1459 "to currently maximum possible %llu sectors <==\n",
1460 (unsigned long long) max_possible_sectors);
1461 if (new_disk_conf->meta_dev_idx >= 0)
1462 dev_warn(DEV, "==>> using internal or flexible "
1463 "meta data may help <<==\n");
1466 drbd_suspend_io(mdev);
1467 /* also wait for the last barrier ack. */
1468 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1469 * We need a way to either ignore barrier acks for barriers sent before a device
1470 * was attached, or a way to wait for all pending barrier acks to come in.
1471 * As barriers are counted per resource,
1472 * we'd need to suspend io on all devices of a resource.
1474 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_pending_cnt) || drbd_suspended(mdev));
1475 /* and for any other previously queued work */
1476 drbd_flush_workqueue(mdev);
1478 rv = _drbd_request_state(mdev, NS(disk, D_ATTACHING), CS_VERBOSE);
1479 retcode = rv; /* FIXME: Type mismatch. */
1480 drbd_resume_io(mdev);
1481 if (rv < SS_SUCCESS)
1484 if (!get_ldev_if_state(mdev, D_ATTACHING))
1485 goto force_diskless;
1487 if (!mdev->bitmap) {
1488 if (drbd_bm_init(mdev)) {
1489 retcode = ERR_NOMEM;
1490 goto force_diskless_dec;
1494 if (mdev->state.conn < C_CONNECTED &&
1495 mdev->state.role == R_PRIMARY &&
1496 (mdev->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1497 dev_err(DEV, "Can only attach to data with current UUID=%016llX\n",
1498 (unsigned long long)mdev->ed_uuid);
1499 retcode = ERR_DATA_NOT_CURRENT;
1500 goto force_diskless_dec;
1503 /* Since we are diskless, fix the activity log first... */
1504 if (drbd_check_al_size(mdev, new_disk_conf)) {
1505 retcode = ERR_NOMEM;
1506 goto force_diskless_dec;
1509 /* Prevent shrinking of consistent devices ! */
1510 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1511 drbd_new_dev_size(mdev, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1512 dev_warn(DEV, "refusing to truncate a consistent device\n");
1513 retcode = ERR_DISK_TOO_SMALL;
1514 goto force_diskless_dec;
1517 /* Reset the "barriers don't work" bits here, then force meta data to
1518 * be written, to ensure we determine if barriers are supported. */
1519 if (new_disk_conf->md_flushes)
1520 clear_bit(MD_NO_FUA, &mdev->flags);
1522 set_bit(MD_NO_FUA, &mdev->flags);
1524 /* Point of no return reached.
1525 * Devices and memory are no longer released by error cleanup below.
1526 * now mdev takes over responsibility, and the state engine should
1527 * clean it up somewhere. */
1528 D_ASSERT(mdev->ldev == NULL);
1530 mdev->resync = resync_lru;
1531 mdev->rs_plan_s = new_plan;
1534 new_disk_conf = NULL;
1537 drbd_bump_write_ordering(mdev->tconn, WO_bdev_flush);
1539 if (drbd_md_test_flag(mdev->ldev, MDF_CRASHED_PRIMARY))
1540 set_bit(CRASHED_PRIMARY, &mdev->flags);
1542 clear_bit(CRASHED_PRIMARY, &mdev->flags);
1544 if (drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1545 !(mdev->state.role == R_PRIMARY && mdev->tconn->susp_nod))
1546 set_bit(CRASHED_PRIMARY, &mdev->flags);
1553 drbd_reconsider_max_bio_size(mdev);
1555 /* If I am currently not R_PRIMARY,
1556 * but meta data primary indicator is set,
1557 * I just now recover from a hard crash,
1558 * and have been R_PRIMARY before that crash.
1560 * Now, if I had no connection before that crash
1561 * (have been degraded R_PRIMARY), chances are that
1562 * I won't find my peer now either.
1564 * In that case, and _only_ in that case,
1565 * we use the degr-wfc-timeout instead of the default,
1566 * so we can automatically recover from a crash of a
1567 * degraded but active "cluster" after a certain timeout.
1569 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
1570 if (mdev->state.role != R_PRIMARY &&
1571 drbd_md_test_flag(mdev->ldev, MDF_PRIMARY_IND) &&
1572 !drbd_md_test_flag(mdev->ldev, MDF_CONNECTED_IND))
1573 set_bit(USE_DEGR_WFC_T, &mdev->flags);
1575 dd = drbd_determine_dev_size(mdev, 0);
1576 if (dd == dev_size_error) {
1577 retcode = ERR_NOMEM_BITMAP;
1578 goto force_diskless_dec;
1579 } else if (dd == grew)
1580 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
1582 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC) ||
1583 (test_bit(CRASHED_PRIMARY, &mdev->flags) &&
1584 drbd_md_test_flag(mdev->ldev, MDF_AL_DISABLED))) {
1585 dev_info(DEV, "Assuming that all blocks are out of sync "
1586 "(aka FullSync)\n");
1587 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write,
1588 "set_n_write from attaching", BM_LOCKED_MASK)) {
1589 retcode = ERR_IO_MD_DISK;
1590 goto force_diskless_dec;
1593 if (drbd_bitmap_io(mdev, &drbd_bm_read,
1594 "read from attaching", BM_LOCKED_MASK)) {
1595 retcode = ERR_IO_MD_DISK;
1596 goto force_diskless_dec;
1600 if (_drbd_bm_total_weight(mdev) == drbd_bm_bits(mdev))
1601 drbd_suspend_al(mdev); /* IO is still suspended here... */
1603 spin_lock_irq(&mdev->tconn->req_lock);
1604 os = drbd_read_state(mdev);
1606 /* If MDF_CONSISTENT is not set go into inconsistent state,
1607 otherwise investigate MDF_WasUpToDate...
1608 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1609 otherwise into D_CONSISTENT state.
1611 if (drbd_md_test_flag(mdev->ldev, MDF_CONSISTENT)) {
1612 if (drbd_md_test_flag(mdev->ldev, MDF_WAS_UP_TO_DATE))
1613 ns.disk = D_CONSISTENT;
1615 ns.disk = D_OUTDATED;
1617 ns.disk = D_INCONSISTENT;
1620 if (drbd_md_test_flag(mdev->ldev, MDF_PEER_OUT_DATED))
1621 ns.pdsk = D_OUTDATED;
1624 if (ns.disk == D_CONSISTENT &&
1625 (ns.pdsk == D_OUTDATED || rcu_dereference(mdev->ldev->disk_conf)->fencing == FP_DONT_CARE))
1626 ns.disk = D_UP_TO_DATE;
1628 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1629 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1630 this point, because drbd_request_state() modifies these
1633 if (rcu_dereference(mdev->ldev->disk_conf)->al_updates)
1634 mdev->ldev->md.flags &= ~MDF_AL_DISABLED;
1636 mdev->ldev->md.flags |= MDF_AL_DISABLED;
1640 /* In case we are C_CONNECTED postpone any decision on the new disk
1641 state after the negotiation phase. */
1642 if (mdev->state.conn == C_CONNECTED) {
1643 mdev->new_state_tmp.i = ns.i;
1645 ns.disk = D_NEGOTIATING;
1647 /* We expect to receive up-to-date UUIDs soon.
1648 To avoid a race in receive_state, free p_uuid while
1649 holding req_lock. I.e. atomic with the state change */
1650 kfree(mdev->p_uuid);
1651 mdev->p_uuid = NULL;
1654 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1655 spin_unlock_irq(&mdev->tconn->req_lock);
1657 if (rv < SS_SUCCESS)
1658 goto force_diskless_dec;
1660 mod_timer(&mdev->request_timer, jiffies + HZ);
1662 if (mdev->state.role == R_PRIMARY)
1663 mdev->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1665 mdev->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1667 drbd_md_mark_dirty(mdev);
1670 kobject_uevent(&disk_to_dev(mdev->vdisk)->kobj, KOBJ_CHANGE);
1672 conn_reconfig_done(mdev->tconn);
1673 drbd_adm_finish(info, retcode);
1679 drbd_force_state(mdev, NS(disk, D_DISKLESS));
1682 conn_reconfig_done(mdev->tconn);
1684 if (nbc->backing_bdev)
1685 blkdev_put(nbc->backing_bdev,
1686 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1688 blkdev_put(nbc->md_bdev,
1689 FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1692 kfree(new_disk_conf);
1693 lc_destroy(resync_lru);
1697 drbd_adm_finish(info, retcode);
1701 static int adm_detach(struct drbd_conf *mdev, int force)
1703 enum drbd_state_rv retcode;
1707 set_bit(FORCE_DETACH, &mdev->flags);
1708 drbd_force_state(mdev, NS(disk, D_FAILED));
1709 retcode = SS_SUCCESS;
1713 drbd_suspend_io(mdev); /* so no-one is stuck in drbd_al_begin_io */
1714 drbd_md_get_buffer(mdev); /* make sure there is no in-flight meta-data IO */
1715 retcode = drbd_request_state(mdev, NS(disk, D_FAILED));
1716 drbd_md_put_buffer(mdev);
1717 /* D_FAILED will transition to DISKLESS. */
1718 ret = wait_event_interruptible(mdev->misc_wait,
1719 mdev->state.disk != D_FAILED);
1720 drbd_resume_io(mdev);
1721 if ((int)retcode == (int)SS_IS_DISKLESS)
1722 retcode = SS_NOTHING_TO_DO;
1729 /* Detaching the disk is a process in multiple stages. First we need to lock
1730 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1731 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1732 * internal references as well.
1733 * Only then we have finally detached. */
1734 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1736 enum drbd_ret_code retcode;
1737 struct detach_parms parms = { };
1740 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
1741 if (!adm_ctx.reply_skb)
1743 if (retcode != NO_ERROR)
1746 if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
1747 err = detach_parms_from_attrs(&parms, info);
1749 retcode = ERR_MANDATORY_TAG;
1750 drbd_msg_put_info(from_attrs_err_to_txt(err));
1755 retcode = adm_detach(adm_ctx.mdev, parms.force_detach);
1757 drbd_adm_finish(info, retcode);
1761 static bool conn_resync_running(struct drbd_tconn *tconn)
1763 struct drbd_conf *mdev;
1768 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1769 if (mdev->state.conn == C_SYNC_SOURCE ||
1770 mdev->state.conn == C_SYNC_TARGET ||
1771 mdev->state.conn == C_PAUSED_SYNC_S ||
1772 mdev->state.conn == C_PAUSED_SYNC_T) {
1782 static bool conn_ov_running(struct drbd_tconn *tconn)
1784 struct drbd_conf *mdev;
1789 idr_for_each_entry(&tconn->volumes, mdev, vnr) {
1790 if (mdev->state.conn == C_VERIFY_S ||
1791 mdev->state.conn == C_VERIFY_T) {
1801 static enum drbd_ret_code
1802 _check_net_options(struct drbd_tconn *tconn, struct net_conf *old_conf, struct net_conf *new_conf)
1804 struct drbd_conf *mdev;
1807 if (old_conf && tconn->cstate == C_WF_REPORT_PARAMS && tconn->agreed_pro_version < 100) {
1808 if (new_conf->wire_protocol != old_conf->wire_protocol)
1809 return ERR_NEED_APV_100;
1811 if (new_conf->two_primaries != old_conf->two_primaries)
1812 return ERR_NEED_APV_100;
1814 if (strcmp(new_conf->integrity_alg, old_conf->integrity_alg))
1815 return ERR_NEED_APV_100;
1818 if (!new_conf->two_primaries &&
1819 conn_highest_role(tconn) == R_PRIMARY &&
1820 conn_highest_peer(tconn) == R_PRIMARY)
1821 return ERR_NEED_ALLOW_TWO_PRI;
1823 if (new_conf->two_primaries &&
1824 (new_conf->wire_protocol != DRBD_PROT_C))
1825 return ERR_NOT_PROTO_C;
1827 idr_for_each_entry(&tconn->volumes, mdev, i) {
1828 if (get_ldev(mdev)) {
1829 enum drbd_fencing_p fp = rcu_dereference(mdev->ldev->disk_conf)->fencing;
1831 if (new_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
1832 return ERR_STONITH_AND_PROT_A;
1834 if (mdev->state.role == R_PRIMARY && new_conf->discard_my_data)
1835 return ERR_DISCARD_IMPOSSIBLE;
1838 if (new_conf->on_congestion != OC_BLOCK && new_conf->wire_protocol != DRBD_PROT_A)
1839 return ERR_CONG_NOT_PROTO_A;
1844 static enum drbd_ret_code
1845 check_net_options(struct drbd_tconn *tconn, struct net_conf *new_conf)
1847 static enum drbd_ret_code rv;
1848 struct drbd_conf *mdev;
1852 rv = _check_net_options(tconn, rcu_dereference(tconn->net_conf), new_conf);
1855 /* tconn->volumes protected by genl_lock() here */
1856 idr_for_each_entry(&tconn->volumes, mdev, i) {
1857 if (!mdev->bitmap) {
1858 if(drbd_bm_init(mdev))
1867 struct crypto_hash *verify_tfm;
1868 struct crypto_hash *csums_tfm;
1869 struct crypto_hash *cram_hmac_tfm;
1870 struct crypto_hash *integrity_tfm;
1874 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
1879 *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
1888 static enum drbd_ret_code
1889 alloc_crypto(struct crypto *crypto, struct net_conf *new_conf)
1891 char hmac_name[CRYPTO_MAX_ALG_NAME];
1892 enum drbd_ret_code rv;
1894 rv = alloc_hash(&crypto->csums_tfm, new_conf->csums_alg,
1898 rv = alloc_hash(&crypto->verify_tfm, new_conf->verify_alg,
1902 rv = alloc_hash(&crypto->integrity_tfm, new_conf->integrity_alg,
1906 if (new_conf->cram_hmac_alg[0] != 0) {
1907 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
1908 new_conf->cram_hmac_alg);
1910 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
1917 static void free_crypto(struct crypto *crypto)
1919 crypto_free_hash(crypto->cram_hmac_tfm);
1920 crypto_free_hash(crypto->integrity_tfm);
1921 crypto_free_hash(crypto->csums_tfm);
1922 crypto_free_hash(crypto->verify_tfm);
1925 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
1927 enum drbd_ret_code retcode;
1928 struct drbd_tconn *tconn;
1929 struct net_conf *old_conf, *new_conf = NULL;
1931 int ovr; /* online verify running */
1932 int rsr; /* re-sync running */
1933 struct crypto crypto = { };
1935 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
1936 if (!adm_ctx.reply_skb)
1938 if (retcode != NO_ERROR)
1941 tconn = adm_ctx.tconn;
1943 new_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
1945 retcode = ERR_NOMEM;
1949 conn_reconfig_start(tconn);
1951 mutex_lock(&tconn->data.mutex);
1952 mutex_lock(&tconn->conf_update);
1953 old_conf = tconn->net_conf;
1956 drbd_msg_put_info("net conf missing, try connect");
1957 retcode = ERR_INVALID_REQUEST;
1961 *new_conf = *old_conf;
1962 if (should_set_defaults(info))
1963 set_net_conf_defaults(new_conf);
1965 err = net_conf_from_attrs_for_change(new_conf, info);
1966 if (err && err != -ENOMSG) {
1967 retcode = ERR_MANDATORY_TAG;
1968 drbd_msg_put_info(from_attrs_err_to_txt(err));
1972 retcode = check_net_options(tconn, new_conf);
1973 if (retcode != NO_ERROR)
1976 /* re-sync running */
1977 rsr = conn_resync_running(tconn);
1978 if (rsr && strcmp(new_conf->csums_alg, old_conf->csums_alg)) {
1979 retcode = ERR_CSUMS_RESYNC_RUNNING;
1983 /* online verify running */
1984 ovr = conn_ov_running(tconn);
1985 if (ovr && strcmp(new_conf->verify_alg, old_conf->verify_alg)) {
1986 retcode = ERR_VERIFY_RUNNING;
1990 retcode = alloc_crypto(&crypto, new_conf);
1991 if (retcode != NO_ERROR)
1994 rcu_assign_pointer(tconn->net_conf, new_conf);
1997 crypto_free_hash(tconn->csums_tfm);
1998 tconn->csums_tfm = crypto.csums_tfm;
1999 crypto.csums_tfm = NULL;
2002 crypto_free_hash(tconn->verify_tfm);
2003 tconn->verify_tfm = crypto.verify_tfm;
2004 crypto.verify_tfm = NULL;
2007 crypto_free_hash(tconn->integrity_tfm);
2008 tconn->integrity_tfm = crypto.integrity_tfm;
2009 if (tconn->cstate >= C_WF_REPORT_PARAMS && tconn->agreed_pro_version >= 100)
2010 /* Do this without trying to take tconn->data.mutex again. */
2011 __drbd_send_protocol(tconn, P_PROTOCOL_UPDATE);
2013 crypto_free_hash(tconn->cram_hmac_tfm);
2014 tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2016 mutex_unlock(&tconn->conf_update);
2017 mutex_unlock(&tconn->data.mutex);
2021 if (tconn->cstate >= C_WF_REPORT_PARAMS)
2022 drbd_send_sync_param(minor_to_mdev(conn_lowest_minor(tconn)));
2027 mutex_unlock(&tconn->conf_update);
2028 mutex_unlock(&tconn->data.mutex);
2029 free_crypto(&crypto);
2032 conn_reconfig_done(tconn);
2034 drbd_adm_finish(info, retcode);
2038 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2040 struct drbd_conf *mdev;
2041 struct net_conf *old_conf, *new_conf = NULL;
2042 struct crypto crypto = { };
2043 struct drbd_tconn *tconn;
2044 enum drbd_ret_code retcode;
2048 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2050 if (!adm_ctx.reply_skb)
2052 if (retcode != NO_ERROR)
2054 if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2055 drbd_msg_put_info("connection endpoint(s) missing");
2056 retcode = ERR_INVALID_REQUEST;
2060 /* No need for _rcu here. All reconfiguration is
2061 * strictly serialized on genl_lock(). We are protected against
2062 * concurrent reconfiguration/addition/deletion */
2063 list_for_each_entry(tconn, &drbd_tconns, all_tconn) {
2064 if (nla_len(adm_ctx.my_addr) == tconn->my_addr_len &&
2065 !memcmp(nla_data(adm_ctx.my_addr), &tconn->my_addr, tconn->my_addr_len)) {
2066 retcode = ERR_LOCAL_ADDR;
2070 if (nla_len(adm_ctx.peer_addr) == tconn->peer_addr_len &&
2071 !memcmp(nla_data(adm_ctx.peer_addr), &tconn->peer_addr, tconn->peer_addr_len)) {
2072 retcode = ERR_PEER_ADDR;
2077 tconn = adm_ctx.tconn;
2078 conn_reconfig_start(tconn);
2080 if (tconn->cstate > C_STANDALONE) {
2081 retcode = ERR_NET_CONFIGURED;
2085 /* allocation not in the IO path, drbdsetup / netlink process context */
2086 new_conf = kzalloc(sizeof(*new_conf), GFP_KERNEL);
2088 retcode = ERR_NOMEM;
2092 set_net_conf_defaults(new_conf);
2094 err = net_conf_from_attrs(new_conf, info);
2095 if (err && err != -ENOMSG) {
2096 retcode = ERR_MANDATORY_TAG;
2097 drbd_msg_put_info(from_attrs_err_to_txt(err));
2101 retcode = check_net_options(tconn, new_conf);
2102 if (retcode != NO_ERROR)
2105 retcode = alloc_crypto(&crypto, new_conf);
2106 if (retcode != NO_ERROR)
2109 ((char *)new_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2111 conn_flush_workqueue(tconn);
2113 mutex_lock(&tconn->conf_update);
2114 old_conf = tconn->net_conf;
2116 retcode = ERR_NET_CONFIGURED;
2117 mutex_unlock(&tconn->conf_update);
2120 rcu_assign_pointer(tconn->net_conf, new_conf);
2122 conn_free_crypto(tconn);
2123 tconn->cram_hmac_tfm = crypto.cram_hmac_tfm;
2124 tconn->integrity_tfm = crypto.integrity_tfm;
2125 tconn->csums_tfm = crypto.csums_tfm;
2126 tconn->verify_tfm = crypto.verify_tfm;
2128 tconn->my_addr_len = nla_len(adm_ctx.my_addr);
2129 memcpy(&tconn->my_addr, nla_data(adm_ctx.my_addr), tconn->my_addr_len);
2130 tconn->peer_addr_len = nla_len(adm_ctx.peer_addr);
2131 memcpy(&tconn->peer_addr, nla_data(adm_ctx.peer_addr), tconn->peer_addr_len);
2133 mutex_unlock(&tconn->conf_update);
2136 idr_for_each_entry(&tconn->volumes, mdev, i) {
2142 retcode = conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2144 conn_reconfig_done(tconn);
2145 drbd_adm_finish(info, retcode);
2149 free_crypto(&crypto);
2152 conn_reconfig_done(tconn);
2154 drbd_adm_finish(info, retcode);
2158 static enum drbd_state_rv conn_try_disconnect(struct drbd_tconn *tconn, bool force)
2160 enum drbd_state_rv rv;
2162 rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2163 force ? CS_HARD : 0);
2166 case SS_NOTHING_TO_DO:
2168 case SS_ALREADY_STANDALONE:
2170 case SS_PRIMARY_NOP:
2171 /* Our state checking code wants to see the peer outdated. */
2172 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2173 pdsk, D_OUTDATED), CS_VERBOSE);
2175 case SS_CW_FAILED_BY_PEER:
2176 /* The peer probably wants to see us outdated. */
2177 rv = conn_request_state(tconn, NS2(conn, C_DISCONNECTING,
2178 disk, D_OUTDATED), 0);
2179 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2180 rv = conn_request_state(tconn, NS(conn, C_DISCONNECTING),
2185 /* no special handling necessary */
2188 if (rv >= SS_SUCCESS) {
2189 enum drbd_state_rv rv2;
2190 /* No one else can reconfigure the network while I am here.
2191 * The state handling only uses drbd_thread_stop_nowait(),
2192 * we want to really wait here until the receiver is no more.
2194 drbd_thread_stop(&adm_ctx.tconn->receiver);
2196 /* Race breaker. This additional state change request may be
2197 * necessary, if this was a forced disconnect during a receiver
2198 * restart. We may have "killed" the receiver thread just
2199 * after drbdd_init() returned. Typically, we should be
2200 * C_STANDALONE already, now, and this becomes a no-op.
2202 rv2 = conn_request_state(tconn, NS(conn, C_STANDALONE),
2203 CS_VERBOSE | CS_HARD);
2204 if (rv2 < SS_SUCCESS)
2206 "unexpected rv2=%d in conn_try_disconnect()\n",
2212 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2214 struct disconnect_parms parms;
2215 struct drbd_tconn *tconn;
2216 enum drbd_state_rv rv;
2217 enum drbd_ret_code retcode;
2220 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_CONNECTION);
2221 if (!adm_ctx.reply_skb)
2223 if (retcode != NO_ERROR)
2226 tconn = adm_ctx.tconn;
2227 memset(&parms, 0, sizeof(parms));
2228 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2229 err = disconnect_parms_from_attrs(&parms, info);
2231 retcode = ERR_MANDATORY_TAG;
2232 drbd_msg_put_info(from_attrs_err_to_txt(err));
2237 rv = conn_try_disconnect(tconn, parms.force_disconnect);
2238 if (rv < SS_SUCCESS)
2239 retcode = rv; /* FIXME: Type mismatch. */
2243 drbd_adm_finish(info, retcode);
2247 void resync_after_online_grow(struct drbd_conf *mdev)
2249 int iass; /* I am sync source */
2251 dev_info(DEV, "Resync of new storage after online grow\n");
2252 if (mdev->state.role != mdev->state.peer)
2253 iass = (mdev->state.role == R_PRIMARY);
2255 iass = test_bit(RESOLVE_CONFLICTS, &mdev->tconn->flags);
2258 drbd_start_resync(mdev, C_SYNC_SOURCE);
2260 _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2263 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2265 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2266 struct resize_parms rs;
2267 struct drbd_conf *mdev;
2268 enum drbd_ret_code retcode;
2269 enum determine_dev_size dd;
2270 enum dds_flags ddsf;
2274 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2275 if (!adm_ctx.reply_skb)
2277 if (retcode != NO_ERROR)
2280 memset(&rs, 0, sizeof(struct resize_parms));
2281 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2282 err = resize_parms_from_attrs(&rs, info);
2284 retcode = ERR_MANDATORY_TAG;
2285 drbd_msg_put_info(from_attrs_err_to_txt(err));
2290 mdev = adm_ctx.mdev;
2291 if (mdev->state.conn > C_CONNECTED) {
2292 retcode = ERR_RESIZE_RESYNC;
2296 if (mdev->state.role == R_SECONDARY &&
2297 mdev->state.peer == R_SECONDARY) {
2298 retcode = ERR_NO_PRIMARY;
2302 if (!get_ldev(mdev)) {
2303 retcode = ERR_NO_DISK;
2307 if (rs.no_resync && mdev->tconn->agreed_pro_version < 93) {
2308 retcode = ERR_NEED_APV_93;
2313 u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
2315 if (u_size != (sector_t)rs.resize_size) {
2316 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2317 if (!new_disk_conf) {
2318 retcode = ERR_NOMEM;
2323 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev))
2324 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2326 if (new_disk_conf) {
2327 mutex_lock(&mdev->tconn->conf_update);
2328 old_disk_conf = mdev->ldev->disk_conf;
2329 *new_disk_conf = *old_disk_conf;
2330 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2331 rcu_assign_pointer(mdev->ldev->disk_conf, new_disk_conf);
2332 mutex_unlock(&mdev->tconn->conf_update);
2334 kfree(old_disk_conf);
2337 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2338 dd = drbd_determine_dev_size(mdev, ddsf);
2341 if (dd == dev_size_error) {
2342 retcode = ERR_NOMEM_BITMAP;
2346 if (mdev->state.conn == C_CONNECTED) {
2348 set_bit(RESIZE_PENDING, &mdev->flags);
2350 drbd_send_uuids(mdev);
2351 drbd_send_sizes(mdev, 1, ddsf);
2355 drbd_adm_finish(info, retcode);
2363 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2365 enum drbd_ret_code retcode;
2366 struct drbd_tconn *tconn;
2367 struct res_opts res_opts;
2370 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
2371 if (!adm_ctx.reply_skb)
2373 if (retcode != NO_ERROR)
2375 tconn = adm_ctx.tconn;
2377 res_opts = tconn->res_opts;
2378 if (should_set_defaults(info))
2379 set_res_opts_defaults(&res_opts);
2381 err = res_opts_from_attrs(&res_opts, info);
2382 if (err && err != -ENOMSG) {
2383 retcode = ERR_MANDATORY_TAG;
2384 drbd_msg_put_info(from_attrs_err_to_txt(err));
2388 err = set_resource_options(tconn, &res_opts);
2390 retcode = ERR_INVALID_REQUEST;
2392 retcode = ERR_NOMEM;
2396 drbd_adm_finish(info, retcode);
2400 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2402 struct drbd_conf *mdev;
2403 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2405 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2406 if (!adm_ctx.reply_skb)
2408 if (retcode != NO_ERROR)
2411 mdev = adm_ctx.mdev;
2413 /* If there is still bitmap IO pending, probably because of a previous
2414 * resync just being finished, wait for it before requesting a new resync.
2415 * Also wait for it's after_state_ch(). */
2416 drbd_suspend_io(mdev);
2417 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2418 drbd_flush_workqueue(mdev);
2420 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T), CS_ORDERED);
2422 if (retcode < SS_SUCCESS && retcode != SS_NEED_CONNECTION)
2423 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2425 while (retcode == SS_NEED_CONNECTION) {
2426 spin_lock_irq(&mdev->tconn->req_lock);
2427 if (mdev->state.conn < C_CONNECTED)
2428 retcode = _drbd_set_state(_NS(mdev, disk, D_INCONSISTENT), CS_VERBOSE, NULL);
2429 spin_unlock_irq(&mdev->tconn->req_lock);
2431 if (retcode != SS_NEED_CONNECTION)
2434 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_T));
2436 drbd_resume_io(mdev);
2439 drbd_adm_finish(info, retcode);
2443 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2444 union drbd_state mask, union drbd_state val)
2446 enum drbd_ret_code retcode;
2448 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2449 if (!adm_ctx.reply_skb)
2451 if (retcode != NO_ERROR)
2454 retcode = drbd_request_state(adm_ctx.mdev, mask, val);
2456 drbd_adm_finish(info, retcode);
2460 static int drbd_bmio_set_susp_al(struct drbd_conf *mdev)
2464 rv = drbd_bmio_set_n_write(mdev);
2465 drbd_suspend_al(mdev);
2469 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2471 int retcode; /* drbd_ret_code, drbd_state_rv */
2472 struct drbd_conf *mdev;
2474 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2475 if (!adm_ctx.reply_skb)
2477 if (retcode != NO_ERROR)
2480 mdev = adm_ctx.mdev;
2482 /* If there is still bitmap IO pending, probably because of a previous
2483 * resync just being finished, wait for it before requesting a new resync.
2484 * Also wait for it's after_state_ch(). */
2485 drbd_suspend_io(mdev);
2486 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2487 drbd_flush_workqueue(mdev);
2489 retcode = _drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S), CS_ORDERED);
2490 if (retcode < SS_SUCCESS) {
2491 if (retcode == SS_NEED_CONNECTION && mdev->state.role == R_PRIMARY) {
2492 /* The peer will get a resync upon connect anyways.
2493 * Just make that into a full resync. */
2494 retcode = drbd_request_state(mdev, NS(pdsk, D_INCONSISTENT));
2495 if (retcode >= SS_SUCCESS) {
2496 if (drbd_bitmap_io(mdev, &drbd_bmio_set_susp_al,
2497 "set_n_write from invalidate_peer",
2498 BM_LOCKED_SET_ALLOWED))
2499 retcode = ERR_IO_MD_DISK;
2502 retcode = drbd_request_state(mdev, NS(conn, C_STARTING_SYNC_S));
2504 drbd_resume_io(mdev);
2507 drbd_adm_finish(info, retcode);
2511 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2513 enum drbd_ret_code retcode;
2515 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2516 if (!adm_ctx.reply_skb)
2518 if (retcode != NO_ERROR)
2521 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2522 retcode = ERR_PAUSE_IS_SET;
2524 drbd_adm_finish(info, retcode);
2528 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2530 union drbd_dev_state s;
2531 enum drbd_ret_code retcode;
2533 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2534 if (!adm_ctx.reply_skb)
2536 if (retcode != NO_ERROR)
2539 if (drbd_request_state(adm_ctx.mdev, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2540 s = adm_ctx.mdev->state;
2541 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2542 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2543 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2545 retcode = ERR_PAUSE_IS_CLEAR;
2550 drbd_adm_finish(info, retcode);
2554 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2556 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2559 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2561 struct drbd_conf *mdev;
2562 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2564 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2565 if (!adm_ctx.reply_skb)
2567 if (retcode != NO_ERROR)
2570 mdev = adm_ctx.mdev;
2571 if (test_bit(NEW_CUR_UUID, &mdev->flags)) {
2572 drbd_uuid_new_current(mdev);
2573 clear_bit(NEW_CUR_UUID, &mdev->flags);
2575 drbd_suspend_io(mdev);
2576 retcode = drbd_request_state(mdev, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2577 if (retcode == SS_SUCCESS) {
2578 if (mdev->state.conn < C_CONNECTED)
2579 tl_clear(mdev->tconn);
2580 if (mdev->state.disk == D_DISKLESS || mdev->state.disk == D_FAILED)
2581 tl_restart(mdev->tconn, FAIL_FROZEN_DISK_IO);
2583 drbd_resume_io(mdev);
2586 drbd_adm_finish(info, retcode);
2590 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2592 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2595 int nla_put_drbd_cfg_context(struct sk_buff *skb, struct drbd_tconn *tconn, unsigned vnr)
2598 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2600 goto nla_put_failure;
2601 if (vnr != VOLUME_UNSPECIFIED &&
2602 nla_put_u32(skb, T_ctx_volume, vnr))
2603 goto nla_put_failure;
2604 if (nla_put_string(skb, T_ctx_resource_name, tconn->name))
2605 goto nla_put_failure;
2606 if (tconn->my_addr_len &&
2607 nla_put(skb, T_ctx_my_addr, tconn->my_addr_len, &tconn->my_addr))
2608 goto nla_put_failure;
2609 if (tconn->peer_addr_len &&
2610 nla_put(skb, T_ctx_peer_addr, tconn->peer_addr_len, &tconn->peer_addr))
2611 goto nla_put_failure;
2612 nla_nest_end(skb, nla);
2617 nla_nest_cancel(skb, nla);
2621 int nla_put_status_info(struct sk_buff *skb, struct drbd_conf *mdev,
2622 const struct sib_info *sib)
2624 struct state_info *si = NULL; /* for sizeof(si->member); */
2625 struct net_conf *nc;
2629 int exclude_sensitive;
2631 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2632 * to. So we better exclude_sensitive information.
2634 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2635 * in the context of the requesting user process. Exclude sensitive
2636 * information, unless current has superuser.
2638 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2639 * relies on the current implementation of netlink_dump(), which
2640 * executes the dump callback successively from netlink_recvmsg(),
2641 * always in the context of the receiving process */
2642 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2644 got_ldev = get_ldev(mdev);
2646 /* We need to add connection name and volume number information still.
2647 * Minor number is in drbd_genlmsghdr. */
2648 if (nla_put_drbd_cfg_context(skb, mdev->tconn, mdev->vnr))
2649 goto nla_put_failure;
2651 if (res_opts_to_skb(skb, &mdev->tconn->res_opts, exclude_sensitive))
2652 goto nla_put_failure;
2656 if (disk_conf_to_skb(skb, rcu_dereference(mdev->ldev->disk_conf), exclude_sensitive))
2657 goto nla_put_failure;
2659 nc = rcu_dereference(mdev->tconn->net_conf);
2661 err = net_conf_to_skb(skb, nc, exclude_sensitive);
2664 goto nla_put_failure;
2666 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2668 goto nla_put_failure;
2669 if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
2670 nla_put_u32(skb, T_current_state, mdev->state.i) ||
2671 nla_put_u64(skb, T_ed_uuid, mdev->ed_uuid) ||
2672 nla_put_u64(skb, T_capacity, drbd_get_capacity(mdev->this_bdev)) ||
2673 nla_put_u64(skb, T_send_cnt, mdev->send_cnt) ||
2674 nla_put_u64(skb, T_recv_cnt, mdev->recv_cnt) ||
2675 nla_put_u64(skb, T_read_cnt, mdev->read_cnt) ||
2676 nla_put_u64(skb, T_writ_cnt, mdev->writ_cnt) ||
2677 nla_put_u64(skb, T_al_writ_cnt, mdev->al_writ_cnt) ||
2678 nla_put_u64(skb, T_bm_writ_cnt, mdev->bm_writ_cnt) ||
2679 nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&mdev->ap_bio_cnt)) ||
2680 nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&mdev->ap_pending_cnt)) ||
2681 nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&mdev->rs_pending_cnt)))
2682 goto nla_put_failure;
2687 spin_lock_irq(&mdev->ldev->md.uuid_lock);
2688 err = nla_put(skb, T_uuids, sizeof(si->uuids), mdev->ldev->md.uuid);
2689 spin_unlock_irq(&mdev->ldev->md.uuid_lock);
2692 goto nla_put_failure;
2694 if (nla_put_u32(skb, T_disk_flags, mdev->ldev->md.flags) ||
2695 nla_put_u64(skb, T_bits_total, drbd_bm_bits(mdev)) ||
2696 nla_put_u64(skb, T_bits_oos, drbd_bm_total_weight(mdev)))
2697 goto nla_put_failure;
2698 if (C_SYNC_SOURCE <= mdev->state.conn &&
2699 C_PAUSED_SYNC_T >= mdev->state.conn) {
2700 if (nla_put_u64(skb, T_bits_rs_total, mdev->rs_total) ||
2701 nla_put_u64(skb, T_bits_rs_failed, mdev->rs_failed))
2702 goto nla_put_failure;
2707 switch(sib->sib_reason) {
2708 case SIB_SYNC_PROGRESS:
2709 case SIB_GET_STATUS_REPLY:
2711 case SIB_STATE_CHANGE:
2712 if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
2713 nla_put_u32(skb, T_new_state, sib->ns.i))
2714 goto nla_put_failure;
2716 case SIB_HELPER_POST:
2717 if (nla_put_u32(skb, T_helper_exit_code,
2718 sib->helper_exit_code))
2719 goto nla_put_failure;
2721 case SIB_HELPER_PRE:
2722 if (nla_put_string(skb, T_helper, sib->helper_name))
2723 goto nla_put_failure;
2727 nla_nest_end(skb, nla);
2737 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
2739 enum drbd_ret_code retcode;
2742 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2743 if (!adm_ctx.reply_skb)
2745 if (retcode != NO_ERROR)
2748 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.mdev, NULL);
2750 nlmsg_free(adm_ctx.reply_skb);
2754 drbd_adm_finish(info, retcode);
2758 int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
2760 struct drbd_conf *mdev;
2761 struct drbd_genlmsghdr *dh;
2762 struct drbd_tconn *pos = (struct drbd_tconn*)cb->args[0];
2763 struct drbd_tconn *tconn = NULL;
2764 struct drbd_tconn *tmp;
2765 unsigned volume = cb->args[1];
2767 /* Open coded, deferred, iteration:
2768 * list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2769 * idr_for_each_entry(&tconn->volumes, mdev, i) {
2773 * where tconn is cb->args[0];
2774 * and i is cb->args[1];
2776 * cb->args[2] indicates if we shall loop over all resources,
2777 * or just dump all volumes of a single resource.
2779 * This may miss entries inserted after this dump started,
2780 * or entries deleted before they are reached.
2782 * We need to make sure the mdev won't disappear while
2783 * we are looking at it, and revalidate our iterators
2784 * on each iteration.
2787 /* synchronize with conn_create()/conn_destroy() */
2789 /* revalidate iterator position */
2790 list_for_each_entry_rcu(tmp, &drbd_tconns, all_tconn) {
2792 /* first iteration */
2804 mdev = idr_get_next(&tconn->volumes, &volume);
2806 /* No more volumes to dump on this tconn.
2807 * Advance tconn iterator. */
2808 pos = list_entry_rcu(tconn->all_tconn.next,
2809 struct drbd_tconn, all_tconn);
2810 /* Did we dump any volume on this tconn yet? */
2812 /* If we reached the end of the list,
2813 * or only a single resource dump was requested,
2815 if (&pos->all_tconn == &drbd_tconns || cb->args[2])
2823 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2824 cb->nlh->nlmsg_seq, &drbd_genl_family,
2825 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
2830 /* This is a tconn without a single volume.
2831 * Suprisingly enough, it may have a network
2833 struct net_conf *nc;
2835 dh->ret_code = NO_ERROR;
2836 if (nla_put_drbd_cfg_context(skb, tconn, VOLUME_UNSPECIFIED))
2838 nc = rcu_dereference(tconn->net_conf);
2839 if (nc && net_conf_to_skb(skb, nc, 1) != 0)
2844 D_ASSERT(mdev->vnr == volume);
2845 D_ASSERT(mdev->tconn == tconn);
2847 dh->minor = mdev_to_minor(mdev);
2848 dh->ret_code = NO_ERROR;
2850 if (nla_put_status_info(skb, mdev, NULL)) {
2852 genlmsg_cancel(skb, dh);
2856 genlmsg_end(skb, dh);
2861 /* where to start the next iteration */
2862 cb->args[0] = (long)pos;
2863 cb->args[1] = (pos == tconn) ? volume + 1 : 0;
2865 /* No more tconns/volumes/minors found results in an empty skb.
2866 * Which will terminate the dump. */
2871 * Request status of all resources, or of all volumes within a single resource.
2873 * This is a dump, as the answer may not fit in a single reply skb otherwise.
2874 * Which means we cannot use the family->attrbuf or other such members, because
2875 * dump is NOT protected by the genl_lock(). During dump, we only have access
2876 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
2878 * Once things are setup properly, we call into get_one_status().
2880 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
2882 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
2884 const char *resource_name;
2885 struct drbd_tconn *tconn;
2888 /* Is this a followup call? */
2890 /* ... of a single resource dump,
2891 * and the resource iterator has been advanced already? */
2892 if (cb->args[2] && cb->args[2] != cb->args[0])
2893 return 0; /* DONE. */
2897 /* First call (from netlink_dump_start). We need to figure out
2898 * which resource(s) the user wants us to dump. */
2899 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
2900 nlmsg_attrlen(cb->nlh, hdrlen),
2901 DRBD_NLA_CFG_CONTEXT);
2903 /* No explicit context given. Dump all. */
2906 maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
2907 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
2909 return PTR_ERR(nla);
2910 /* context given, but no name present? */
2913 resource_name = nla_data(nla);
2914 tconn = conn_get_by_name(resource_name);
2919 kref_put(&tconn->kref, &conn_destroy); /* get_one_status() (re)validates tconn by itself */
2921 /* prime iterators, and set "filter" mode mark:
2922 * only dump this tconn. */
2923 cb->args[0] = (long)tconn;
2924 /* cb->args[1] = 0; passed in this way. */
2925 cb->args[2] = (long)tconn;
2928 return get_one_status(skb, cb);
2931 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
2933 enum drbd_ret_code retcode;
2934 struct timeout_parms tp;
2937 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2938 if (!adm_ctx.reply_skb)
2940 if (retcode != NO_ERROR)
2944 adm_ctx.mdev->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
2945 test_bit(USE_DEGR_WFC_T, &adm_ctx.mdev->flags) ? UT_DEGRADED :
2948 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
2950 nlmsg_free(adm_ctx.reply_skb);
2954 drbd_adm_finish(info, retcode);
2958 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
2960 struct drbd_conf *mdev;
2961 enum drbd_ret_code retcode;
2962 struct start_ov_parms parms;
2964 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
2965 if (!adm_ctx.reply_skb)
2967 if (retcode != NO_ERROR)
2970 mdev = adm_ctx.mdev;
2972 /* resume from last known position, if possible */
2973 parms.ov_start_sector = mdev->ov_start_sector;
2974 parms.ov_stop_sector = ULLONG_MAX;
2975 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
2976 int err = start_ov_parms_from_attrs(&parms, info);
2978 retcode = ERR_MANDATORY_TAG;
2979 drbd_msg_put_info(from_attrs_err_to_txt(err));
2983 /* w_make_ov_request expects position to be aligned */
2984 mdev->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
2985 mdev->ov_stop_sector = parms.ov_stop_sector;
2987 /* If there is still bitmap IO pending, e.g. previous resync or verify
2988 * just being finished, wait for it before requesting a new resync. */
2989 drbd_suspend_io(mdev);
2990 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
2991 retcode = drbd_request_state(mdev,NS(conn,C_VERIFY_S));
2992 drbd_resume_io(mdev);
2994 drbd_adm_finish(info, retcode);
2999 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
3001 struct drbd_conf *mdev;
3002 enum drbd_ret_code retcode;
3003 int skip_initial_sync = 0;
3005 struct new_c_uuid_parms args;
3007 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3008 if (!adm_ctx.reply_skb)
3010 if (retcode != NO_ERROR)
3013 mdev = adm_ctx.mdev;
3014 memset(&args, 0, sizeof(args));
3015 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
3016 err = new_c_uuid_parms_from_attrs(&args, info);
3018 retcode = ERR_MANDATORY_TAG;
3019 drbd_msg_put_info(from_attrs_err_to_txt(err));
3024 mutex_lock(mdev->state_mutex); /* Protects us against serialized state changes. */
3026 if (!get_ldev(mdev)) {
3027 retcode = ERR_NO_DISK;
3031 /* this is "skip initial sync", assume to be clean */
3032 if (mdev->state.conn == C_CONNECTED && mdev->tconn->agreed_pro_version >= 90 &&
3033 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
3034 dev_info(DEV, "Preparing to skip initial sync\n");
3035 skip_initial_sync = 1;
3036 } else if (mdev->state.conn != C_STANDALONE) {
3037 retcode = ERR_CONNECTED;
3041 drbd_uuid_set(mdev, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
3042 drbd_uuid_new_current(mdev); /* New current, previous to UI_BITMAP */
3044 if (args.clear_bm) {
3045 err = drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
3046 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
3048 dev_err(DEV, "Writing bitmap failed with %d\n",err);
3049 retcode = ERR_IO_MD_DISK;
3051 if (skip_initial_sync) {
3052 drbd_send_uuids_skip_initial_sync(mdev);
3053 _drbd_uuid_set(mdev, UI_BITMAP, 0);
3054 drbd_print_uuids(mdev, "cleared bitmap UUID");
3055 spin_lock_irq(&mdev->tconn->req_lock);
3056 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3058 spin_unlock_irq(&mdev->tconn->req_lock);
3066 mutex_unlock(mdev->state_mutex);
3068 drbd_adm_finish(info, retcode);
3072 static enum drbd_ret_code
3073 drbd_check_resource_name(const char *name)
3075 if (!name || !name[0]) {
3076 drbd_msg_put_info("resource name missing");
3077 return ERR_MANDATORY_TAG;
3079 /* if we want to use these in sysfs/configfs/debugfs some day,
3080 * we must not allow slashes */
3081 if (strchr(name, '/')) {
3082 drbd_msg_put_info("invalid resource name");
3083 return ERR_INVALID_REQUEST;
3088 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
3090 enum drbd_ret_code retcode;
3091 struct res_opts res_opts;
3094 retcode = drbd_adm_prepare(skb, info, 0);
3095 if (!adm_ctx.reply_skb)
3097 if (retcode != NO_ERROR)
3100 set_res_opts_defaults(&res_opts);
3101 err = res_opts_from_attrs(&res_opts, info);
3102 if (err && err != -ENOMSG) {
3103 retcode = ERR_MANDATORY_TAG;
3104 drbd_msg_put_info(from_attrs_err_to_txt(err));
3108 retcode = drbd_check_resource_name(adm_ctx.resource_name);
3109 if (retcode != NO_ERROR)
3112 if (adm_ctx.tconn) {
3113 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
3114 retcode = ERR_INVALID_REQUEST;
3115 drbd_msg_put_info("resource exists");
3117 /* else: still NO_ERROR */
3121 if (!conn_create(adm_ctx.resource_name, &res_opts))
3122 retcode = ERR_NOMEM;
3124 drbd_adm_finish(info, retcode);
3128 int drbd_adm_add_minor(struct sk_buff *skb, struct genl_info *info)
3130 struct drbd_genlmsghdr *dh = info->userhdr;
3131 enum drbd_ret_code retcode;
3133 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3134 if (!adm_ctx.reply_skb)
3136 if (retcode != NO_ERROR)
3139 if (dh->minor > MINORMASK) {
3140 drbd_msg_put_info("requested minor out of range");
3141 retcode = ERR_INVALID_REQUEST;
3144 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
3145 drbd_msg_put_info("requested volume id out of range");
3146 retcode = ERR_INVALID_REQUEST;
3150 /* drbd_adm_prepare made sure already
3151 * that mdev->tconn and mdev->vnr match the request. */
3153 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
3154 retcode = ERR_MINOR_EXISTS;
3155 /* else: still NO_ERROR */
3159 retcode = conn_new_minor(adm_ctx.tconn, dh->minor, adm_ctx.volume);
3161 drbd_adm_finish(info, retcode);
3165 static enum drbd_ret_code adm_delete_minor(struct drbd_conf *mdev)
3167 if (mdev->state.disk == D_DISKLESS &&
3168 /* no need to be mdev->state.conn == C_STANDALONE &&
3169 * we may want to delete a minor from a live replication group.
3171 mdev->state.role == R_SECONDARY) {
3172 _drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS),
3173 CS_VERBOSE + CS_WAIT_COMPLETE);
3174 idr_remove(&mdev->tconn->volumes, mdev->vnr);
3175 idr_remove(&minors, mdev_to_minor(mdev));
3176 destroy_workqueue(mdev->submit.wq);
3177 del_gendisk(mdev->vdisk);
3179 kref_put(&mdev->kref, &drbd_minor_destroy);
3182 return ERR_MINOR_CONFIGURED;
3185 int drbd_adm_delete_minor(struct sk_buff *skb, struct genl_info *info)
3187 enum drbd_ret_code retcode;
3189 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_MINOR);
3190 if (!adm_ctx.reply_skb)
3192 if (retcode != NO_ERROR)
3195 retcode = adm_delete_minor(adm_ctx.mdev);
3197 drbd_adm_finish(info, retcode);
3201 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
3203 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3204 struct drbd_conf *mdev;
3207 retcode = drbd_adm_prepare(skb, info, 0);
3208 if (!adm_ctx.reply_skb)
3210 if (retcode != NO_ERROR)
3213 if (!adm_ctx.tconn) {
3214 retcode = ERR_RES_NOT_KNOWN;
3219 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3220 retcode = drbd_set_role(mdev, R_SECONDARY, 0);
3221 if (retcode < SS_SUCCESS) {
3222 drbd_msg_put_info("failed to demote");
3227 retcode = conn_try_disconnect(adm_ctx.tconn, 0);
3228 if (retcode < SS_SUCCESS) {
3229 drbd_msg_put_info("failed to disconnect");
3234 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3235 retcode = adm_detach(mdev, 0);
3236 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
3237 drbd_msg_put_info("failed to detach");
3242 /* If we reach this, all volumes (of this tconn) are Secondary,
3243 * Disconnected, Diskless, aka Unconfigured. Make sure all threads have
3244 * actually stopped, state handling only does drbd_thread_stop_nowait(). */
3245 drbd_thread_stop(&adm_ctx.tconn->worker);
3247 /* Now, nothing can fail anymore */
3249 /* delete volumes */
3250 idr_for_each_entry(&adm_ctx.tconn->volumes, mdev, i) {
3251 retcode = adm_delete_minor(mdev);
3252 if (retcode != NO_ERROR) {
3253 /* "can not happen" */
3254 drbd_msg_put_info("failed to delete volume");
3259 /* delete connection */
3260 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3261 list_del_rcu(&adm_ctx.tconn->all_tconn);
3263 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3267 /* "can not happen" */
3268 retcode = ERR_RES_IN_USE;
3269 drbd_msg_put_info("failed to delete connection");
3273 drbd_adm_finish(info, retcode);
3277 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
3279 enum drbd_ret_code retcode;
3281 retcode = drbd_adm_prepare(skb, info, DRBD_ADM_NEED_RESOURCE);
3282 if (!adm_ctx.reply_skb)
3284 if (retcode != NO_ERROR)
3287 if (conn_lowest_minor(adm_ctx.tconn) < 0) {
3288 list_del_rcu(&adm_ctx.tconn->all_tconn);
3290 kref_put(&adm_ctx.tconn->kref, &conn_destroy);
3294 retcode = ERR_RES_IN_USE;
3297 if (retcode == NO_ERROR)
3298 drbd_thread_stop(&adm_ctx.tconn->worker);
3300 drbd_adm_finish(info, retcode);
3304 void drbd_bcast_event(struct drbd_conf *mdev, const struct sib_info *sib)
3306 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
3307 struct sk_buff *msg;
3308 struct drbd_genlmsghdr *d_out;
3312 if (sib->sib_reason == SIB_SYNC_PROGRESS) {
3313 if (time_after(jiffies, mdev->rs_last_bcast + HZ))
3314 mdev->rs_last_bcast = jiffies;
3319 seq = atomic_inc_return(&drbd_genl_seq);
3320 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
3325 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
3326 if (!d_out) /* cannot happen, but anyways. */
3327 goto nla_put_failure;
3328 d_out->minor = mdev_to_minor(mdev);
3329 d_out->ret_code = NO_ERROR;
3331 if (nla_put_status_info(msg, mdev, sib))
3332 goto nla_put_failure;
3333 genlmsg_end(msg, d_out);
3334 err = drbd_genl_multicast_events(msg, 0);
3335 /* msg has been consumed or freed in netlink_broadcast() */
3336 if (err && err != -ESRCH)
3344 dev_err(DEV, "Error %d while broadcasting event. "
3345 "Event seq:%u sib_reason:%u\n",
3346 err, seq, sib->sib_reason);