1 // SPDX-License-Identifier: GPL-2.0-only
5 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
7 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
8 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
9 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16 #include <linux/module.h>
17 #include <linux/drbd.h>
20 #include <linux/file.h>
21 #include <linux/slab.h>
22 #include <linux/blkpg.h>
23 #include <linux/cpumask.h>
25 #include "drbd_protocol.h"
27 #include "drbd_state_change.h"
28 #include <asm/unaligned.h>
29 #include <linux/drbd_limits.h>
30 #include <linux/kthread.h>
32 #include <net/genetlink.h>
35 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
36 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
38 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info);
39 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info);
41 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info);
42 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info);
43 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
45 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
46 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
47 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
48 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
49 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
50 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
51 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
52 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
53 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
54 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
55 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
56 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
57 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
58 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
59 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
60 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
61 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
62 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
63 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
64 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
66 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
67 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb);
68 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb);
69 int drbd_adm_dump_devices_done(struct netlink_callback *cb);
70 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb);
71 int drbd_adm_dump_connections_done(struct netlink_callback *cb);
72 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb);
73 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb);
74 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb);
76 #include <linux/drbd_genl_api.h>
78 #include <linux/genl_magic_func.h>
80 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
81 static atomic_t notify_genl_seq = ATOMIC_INIT(2); /* two. */
83 DEFINE_MUTEX(notification_mutex);
85 /* used blkdev_get_by_path, to claim our meta data device(s) */
86 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
88 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
90 genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
91 if (genlmsg_reply(skb, info))
92 pr_err("error sending genl reply\n");
95 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
96 * reason it could fail was no space in skb, and there are 4k available. */
97 static int drbd_msg_put_info(struct sk_buff *skb, const char *info)
102 if (!info || !info[0])
105 nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_REPLY);
109 err = nla_put_string(skb, T_info_text, info);
111 nla_nest_cancel(skb, nla);
114 nla_nest_end(skb, nla);
119 static int drbd_msg_sprintf_info(struct sk_buff *skb, const char *fmt, ...)
122 struct nlattr *nla, *txt;
126 nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_REPLY);
130 txt = nla_reserve(skb, T_info_text, 256);
132 nla_nest_cancel(skb, nla);
136 len = vscnprintf(nla_data(txt), 256, fmt, args);
139 /* maybe: retry with larger reserve, if truncated */
140 txt->nla_len = nla_attr_size(len+1);
141 nlmsg_trim(skb, (char*)txt + NLA_ALIGN(txt->nla_len));
142 nla_nest_end(skb, nla);
147 /* This would be a good candidate for a "pre_doit" hook,
148 * and per-family private info->pointers.
149 * But we need to stay compatible with older kernels.
150 * If it returns successfully, adm_ctx members are valid.
152 * At this point, we still rely on the global genl_lock().
153 * If we want to avoid that, and allow "genl_family.parallel_ops", we may need
154 * to add additional synchronization against object destruction/modification.
156 #define DRBD_ADM_NEED_MINOR 1
157 #define DRBD_ADM_NEED_RESOURCE 2
158 #define DRBD_ADM_NEED_CONNECTION 4
159 static int drbd_adm_prepare(struct drbd_config_context *adm_ctx,
160 struct sk_buff *skb, struct genl_info *info, unsigned flags)
162 struct drbd_genlmsghdr *d_in = genl_info_userhdr(info);
163 const u8 cmd = info->genlhdr->cmd;
166 memset(adm_ctx, 0, sizeof(*adm_ctx));
168 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
169 if (cmd != DRBD_ADM_GET_STATUS && !capable(CAP_NET_ADMIN))
172 adm_ctx->reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
173 if (!adm_ctx->reply_skb) {
178 adm_ctx->reply_dh = genlmsg_put_reply(adm_ctx->reply_skb,
179 info, &drbd_genl_family, 0, cmd);
180 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
182 if (!adm_ctx->reply_dh) {
187 adm_ctx->reply_dh->minor = d_in->minor;
188 adm_ctx->reply_dh->ret_code = NO_ERROR;
190 adm_ctx->volume = VOLUME_UNSPECIFIED;
191 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
193 /* parse and validate only */
194 err = drbd_cfg_context_from_attrs(NULL, info);
198 /* It was present, and valid,
199 * copy it over to the reply skb. */
200 err = nla_put_nohdr(adm_ctx->reply_skb,
201 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
202 info->attrs[DRBD_NLA_CFG_CONTEXT]);
206 /* and assign stuff to the adm_ctx */
207 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
209 adm_ctx->volume = nla_get_u32(nla);
210 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
212 adm_ctx->resource_name = nla_data(nla);
213 adm_ctx->my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)];
214 adm_ctx->peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)];
215 if ((adm_ctx->my_addr &&
216 nla_len(adm_ctx->my_addr) > sizeof(adm_ctx->connection->my_addr)) ||
217 (adm_ctx->peer_addr &&
218 nla_len(adm_ctx->peer_addr) > sizeof(adm_ctx->connection->peer_addr))) {
224 adm_ctx->minor = d_in->minor;
225 adm_ctx->device = minor_to_device(d_in->minor);
227 /* We are protected by the global genl_lock().
228 * But we may explicitly drop it/retake it in drbd_adm_set_role(),
229 * so make sure this object stays around. */
231 kref_get(&adm_ctx->device->kref);
233 if (adm_ctx->resource_name) {
234 adm_ctx->resource = drbd_find_resource(adm_ctx->resource_name);
237 if (!adm_ctx->device && (flags & DRBD_ADM_NEED_MINOR)) {
238 drbd_msg_put_info(adm_ctx->reply_skb, "unknown minor");
239 return ERR_MINOR_INVALID;
241 if (!adm_ctx->resource && (flags & DRBD_ADM_NEED_RESOURCE)) {
242 drbd_msg_put_info(adm_ctx->reply_skb, "unknown resource");
243 if (adm_ctx->resource_name)
244 return ERR_RES_NOT_KNOWN;
245 return ERR_INVALID_REQUEST;
248 if (flags & DRBD_ADM_NEED_CONNECTION) {
249 if (adm_ctx->resource) {
250 drbd_msg_put_info(adm_ctx->reply_skb, "no resource name expected");
251 return ERR_INVALID_REQUEST;
253 if (adm_ctx->device) {
254 drbd_msg_put_info(adm_ctx->reply_skb, "no minor number expected");
255 return ERR_INVALID_REQUEST;
257 if (adm_ctx->my_addr && adm_ctx->peer_addr)
258 adm_ctx->connection = conn_get_by_addrs(nla_data(adm_ctx->my_addr),
259 nla_len(adm_ctx->my_addr),
260 nla_data(adm_ctx->peer_addr),
261 nla_len(adm_ctx->peer_addr));
262 if (!adm_ctx->connection) {
263 drbd_msg_put_info(adm_ctx->reply_skb, "unknown connection");
264 return ERR_INVALID_REQUEST;
268 /* some more paranoia, if the request was over-determined */
269 if (adm_ctx->device && adm_ctx->resource &&
270 adm_ctx->device->resource != adm_ctx->resource) {
271 pr_warn("request: minor=%u, resource=%s; but that minor belongs to resource %s\n",
272 adm_ctx->minor, adm_ctx->resource->name,
273 adm_ctx->device->resource->name);
274 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists in different resource");
275 return ERR_INVALID_REQUEST;
277 if (adm_ctx->device &&
278 adm_ctx->volume != VOLUME_UNSPECIFIED &&
279 adm_ctx->volume != adm_ctx->device->vnr) {
280 pr_warn("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
281 adm_ctx->minor, adm_ctx->volume,
282 adm_ctx->device->vnr, adm_ctx->device->resource->name);
283 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists as different volume");
284 return ERR_INVALID_REQUEST;
287 /* still, provide adm_ctx->resource always, if possible. */
288 if (!adm_ctx->resource) {
289 adm_ctx->resource = adm_ctx->device ? adm_ctx->device->resource
290 : adm_ctx->connection ? adm_ctx->connection->resource : NULL;
291 if (adm_ctx->resource)
292 kref_get(&adm_ctx->resource->kref);
298 nlmsg_free(adm_ctx->reply_skb);
299 adm_ctx->reply_skb = NULL;
303 static int drbd_adm_finish(struct drbd_config_context *adm_ctx,
304 struct genl_info *info, int retcode)
306 if (adm_ctx->device) {
307 kref_put(&adm_ctx->device->kref, drbd_destroy_device);
308 adm_ctx->device = NULL;
310 if (adm_ctx->connection) {
311 kref_put(&adm_ctx->connection->kref, &drbd_destroy_connection);
312 adm_ctx->connection = NULL;
314 if (adm_ctx->resource) {
315 kref_put(&adm_ctx->resource->kref, drbd_destroy_resource);
316 adm_ctx->resource = NULL;
319 if (!adm_ctx->reply_skb)
322 adm_ctx->reply_dh->ret_code = retcode;
323 drbd_adm_send_reply(adm_ctx->reply_skb, info);
327 static void setup_khelper_env(struct drbd_connection *connection, char **envp)
331 /* FIXME: A future version will not allow this case. */
332 if (connection->my_addr_len == 0 || connection->peer_addr_len == 0)
335 switch (((struct sockaddr *)&connection->peer_addr)->sa_family) {
338 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
339 &((struct sockaddr_in6 *)&connection->peer_addr)->sin6_addr);
343 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
344 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
348 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
349 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
351 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
354 int drbd_khelper(struct drbd_device *device, char *cmd)
356 char *envp[] = { "HOME=/",
358 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
359 (char[20]) { }, /* address family */
360 (char[60]) { }, /* address */
363 char *argv[] = {drbd_usermode_helper, cmd, mb, NULL };
364 struct drbd_connection *connection = first_peer_device(device)->connection;
368 if (current == connection->worker.task)
369 set_bit(CALLBACK_PENDING, &connection->flags);
371 snprintf(mb, 14, "minor-%d", device_to_minor(device));
372 setup_khelper_env(connection, envp);
374 /* The helper may take some time.
375 * write out any unsynced meta data changes now */
376 drbd_md_sync(device);
378 drbd_info(device, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, mb);
379 sib.sib_reason = SIB_HELPER_PRE;
380 sib.helper_name = cmd;
381 drbd_bcast_event(device, &sib);
382 notify_helper(NOTIFY_CALL, device, connection, cmd, 0);
383 ret = call_usermodehelper(drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
385 drbd_warn(device, "helper command: %s %s %s exit code %u (0x%x)\n",
386 drbd_usermode_helper, cmd, mb,
387 (ret >> 8) & 0xff, ret);
389 drbd_info(device, "helper command: %s %s %s exit code %u (0x%x)\n",
390 drbd_usermode_helper, cmd, mb,
391 (ret >> 8) & 0xff, ret);
392 sib.sib_reason = SIB_HELPER_POST;
393 sib.helper_exit_code = ret;
394 drbd_bcast_event(device, &sib);
395 notify_helper(NOTIFY_RESPONSE, device, connection, cmd, ret);
397 if (current == connection->worker.task)
398 clear_bit(CALLBACK_PENDING, &connection->flags);
400 if (ret < 0) /* Ignore any ERRNOs we got. */
406 enum drbd_peer_state conn_khelper(struct drbd_connection *connection, char *cmd)
408 char *envp[] = { "HOME=/",
410 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
411 (char[20]) { }, /* address family */
412 (char[60]) { }, /* address */
414 char *resource_name = connection->resource->name;
415 char *argv[] = {drbd_usermode_helper, cmd, resource_name, NULL };
418 setup_khelper_env(connection, envp);
419 conn_md_sync(connection);
421 drbd_info(connection, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, resource_name);
422 /* TODO: conn_bcast_event() ?? */
423 notify_helper(NOTIFY_CALL, NULL, connection, cmd, 0);
425 ret = call_usermodehelper(drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
427 drbd_warn(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
428 drbd_usermode_helper, cmd, resource_name,
429 (ret >> 8) & 0xff, ret);
431 drbd_info(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
432 drbd_usermode_helper, cmd, resource_name,
433 (ret >> 8) & 0xff, ret);
434 /* TODO: conn_bcast_event() ?? */
435 notify_helper(NOTIFY_RESPONSE, NULL, connection, cmd, ret);
437 if (ret < 0) /* Ignore any ERRNOs we got. */
443 static enum drbd_fencing_p highest_fencing_policy(struct drbd_connection *connection)
445 enum drbd_fencing_p fp = FP_NOT_AVAIL;
446 struct drbd_peer_device *peer_device;
450 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
451 struct drbd_device *device = peer_device->device;
452 if (get_ldev_if_state(device, D_CONSISTENT)) {
453 struct disk_conf *disk_conf =
454 rcu_dereference(peer_device->device->ldev->disk_conf);
455 fp = max_t(enum drbd_fencing_p, fp, disk_conf->fencing);
464 static bool resource_is_supended(struct drbd_resource *resource)
466 return resource->susp || resource->susp_fen || resource->susp_nod;
469 bool conn_try_outdate_peer(struct drbd_connection *connection)
471 struct drbd_resource * const resource = connection->resource;
472 unsigned int connect_cnt;
473 union drbd_state mask = { };
474 union drbd_state val = { };
475 enum drbd_fencing_p fp;
479 spin_lock_irq(&resource->req_lock);
480 if (connection->cstate >= C_WF_REPORT_PARAMS) {
481 drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n");
482 spin_unlock_irq(&resource->req_lock);
486 connect_cnt = connection->connect_cnt;
487 spin_unlock_irq(&resource->req_lock);
489 fp = highest_fencing_policy(connection);
492 drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n");
493 spin_lock_irq(&resource->req_lock);
494 if (connection->cstate < C_WF_REPORT_PARAMS) {
495 _conn_request_state(connection,
496 (union drbd_state) { { .susp_fen = 1 } },
497 (union drbd_state) { { .susp_fen = 0 } },
498 CS_VERBOSE | CS_HARD | CS_DC_SUSP);
499 /* We are no longer suspended due to the fencing policy.
500 * We may still be suspended due to the on-no-data-accessible policy.
501 * If that was OND_IO_ERROR, fail pending requests. */
502 if (!resource_is_supended(resource))
503 _tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
505 /* Else: in case we raced with a connection handshake,
506 * let the handshake figure out if we maybe can RESEND,
507 * and do not resume/fail pending requests here.
508 * Worst case is we stay suspended for now, which may be
509 * resolved by either re-establishing the replication link, or
510 * the next link failure, or eventually the administrator. */
511 spin_unlock_irq(&resource->req_lock);
519 r = conn_khelper(connection, "fence-peer");
521 switch ((r>>8) & 0xff) {
522 case P_INCONSISTENT: /* peer is inconsistent */
523 ex_to_string = "peer is inconsistent or worse";
525 val.pdsk = D_INCONSISTENT;
527 case P_OUTDATED: /* peer got outdated, or was already outdated */
528 ex_to_string = "peer was fenced";
530 val.pdsk = D_OUTDATED;
532 case P_DOWN: /* peer was down */
533 if (conn_highest_disk(connection) == D_UP_TO_DATE) {
534 /* we will(have) create(d) a new UUID anyways... */
535 ex_to_string = "peer is unreachable, assumed to be dead";
537 val.pdsk = D_OUTDATED;
539 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
542 case P_PRIMARY: /* Peer is primary, voluntarily outdate myself.
543 * This is useful when an unconnected R_SECONDARY is asked to
544 * become R_PRIMARY, but finds the other peer being active. */
545 ex_to_string = "peer is active";
546 drbd_warn(connection, "Peer is primary, outdating myself.\n");
548 val.disk = D_OUTDATED;
551 /* THINK: do we need to handle this
552 * like case 4, or more like case 5? */
553 if (fp != FP_STONITH)
554 drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n");
555 ex_to_string = "peer was stonithed";
557 val.pdsk = D_OUTDATED;
560 /* The script is broken ... */
561 drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
562 return false; /* Eventually leave IO frozen */
565 drbd_info(connection, "fence-peer helper returned %d (%s)\n",
566 (r>>8) & 0xff, ex_to_string);
569 conn_request_state(connection, mask, val, CS_VERBOSE);
570 here, because we might were able to re-establish the connection in the
572 spin_lock_irq(&resource->req_lock);
573 if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) {
574 if (connection->connect_cnt != connect_cnt)
575 /* In case the connection was established and droped
576 while the fence-peer handler was running, ignore it */
577 drbd_info(connection, "Ignoring fence-peer exit code\n");
579 _conn_request_state(connection, mask, val, CS_VERBOSE);
581 spin_unlock_irq(&resource->req_lock);
583 return conn_highest_pdsk(connection) <= D_OUTDATED;
586 static int _try_outdate_peer_async(void *data)
588 struct drbd_connection *connection = (struct drbd_connection *)data;
590 conn_try_outdate_peer(connection);
592 kref_put(&connection->kref, drbd_destroy_connection);
596 void conn_try_outdate_peer_async(struct drbd_connection *connection)
598 struct task_struct *opa;
600 kref_get(&connection->kref);
601 /* We may have just sent a signal to this thread
602 * to get it out of some blocking network function.
603 * Clear signals; otherwise kthread_run(), which internally uses
604 * wait_on_completion_killable(), will mistake our pending signal
605 * for a new fatal signal and fail. */
606 flush_signals(current);
607 opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h");
609 drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n");
610 kref_put(&connection->kref, drbd_destroy_connection);
615 drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force)
617 struct drbd_peer_device *const peer_device = first_peer_device(device);
618 struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL;
619 const int max_tries = 4;
620 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
624 union drbd_state mask, val;
626 if (new_role == R_PRIMARY) {
627 struct drbd_connection *connection;
629 /* Detect dead peers as soon as possible. */
632 for_each_connection(connection, device->resource)
633 request_ping(connection);
637 mutex_lock(device->state_mutex);
639 mask.i = 0; mask.role = R_MASK;
640 val.i = 0; val.role = new_role;
642 while (try++ < max_tries) {
643 rv = _drbd_request_state_holding_state_mutex(device, mask, val, CS_WAIT_COMPLETE);
645 /* in case we first succeeded to outdate,
646 * but now suddenly could establish a connection */
647 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
653 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
654 (device->state.disk < D_UP_TO_DATE &&
655 device->state.disk >= D_INCONSISTENT)) {
657 val.disk = D_UP_TO_DATE;
662 if (rv == SS_NO_UP_TO_DATE_DISK &&
663 device->state.disk == D_CONSISTENT && mask.pdsk == 0) {
664 D_ASSERT(device, device->state.pdsk == D_UNKNOWN);
666 if (conn_try_outdate_peer(connection)) {
667 val.disk = D_UP_TO_DATE;
673 if (rv == SS_NOTHING_TO_DO)
675 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
676 if (!conn_try_outdate_peer(connection) && force) {
677 drbd_warn(device, "Forced into split brain situation!\n");
679 val.pdsk = D_OUTDATED;
684 if (rv == SS_TWO_PRIMARIES) {
685 /* Maybe the peer is detected as dead very soon...
686 retry at most once more in this case. */
687 if (try < max_tries) {
691 nc = rcu_dereference(connection->net_conf);
692 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
694 schedule_timeout_interruptible(timeo);
698 if (rv < SS_SUCCESS) {
699 rv = _drbd_request_state(device, mask, val,
700 CS_VERBOSE + CS_WAIT_COMPLETE);
711 drbd_warn(device, "Forced to consider local data as UpToDate!\n");
713 /* Wait until nothing is on the fly :) */
714 wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0);
716 /* FIXME also wait for all pending P_BARRIER_ACK? */
718 if (new_role == R_SECONDARY) {
719 if (get_ldev(device)) {
720 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
724 mutex_lock(&device->resource->conf_update);
725 nc = connection->net_conf;
727 nc->discard_my_data = 0; /* without copy; single bit op is atomic */
728 mutex_unlock(&device->resource->conf_update);
730 if (get_ldev(device)) {
731 if (((device->state.conn < C_CONNECTED ||
732 device->state.pdsk <= D_FAILED)
733 && device->ldev->md.uuid[UI_BITMAP] == 0) || forced)
734 drbd_uuid_new_current(device);
736 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
741 /* writeout of activity log covered areas of the bitmap
742 * to stable storage done in after state change already */
744 if (device->state.conn >= C_WF_REPORT_PARAMS) {
745 /* if this was forced, we should consider sync */
747 drbd_send_uuids(peer_device);
748 drbd_send_current_state(peer_device);
751 drbd_md_sync(device);
752 set_disk_ro(device->vdisk, new_role == R_SECONDARY);
753 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
755 mutex_unlock(device->state_mutex);
759 static const char *from_attrs_err_to_txt(int err)
761 return err == -ENOMSG ? "required attribute missing" :
762 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
763 err == -EEXIST ? "can not change invariant setting" :
764 "invalid attribute value";
767 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
769 struct drbd_config_context adm_ctx;
770 struct set_role_parms parms;
772 enum drbd_ret_code retcode;
773 enum drbd_state_rv rv;
775 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
776 if (!adm_ctx.reply_skb)
778 if (retcode != NO_ERROR)
781 memset(&parms, 0, sizeof(parms));
782 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
783 err = set_role_parms_from_attrs(&parms, info);
785 retcode = ERR_MANDATORY_TAG;
786 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
791 mutex_lock(&adm_ctx.resource->adm_mutex);
793 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
794 rv = drbd_set_role(adm_ctx.device, R_PRIMARY, parms.assume_uptodate);
796 rv = drbd_set_role(adm_ctx.device, R_SECONDARY, 0);
798 mutex_unlock(&adm_ctx.resource->adm_mutex);
800 drbd_adm_finish(&adm_ctx, info, rv);
803 drbd_adm_finish(&adm_ctx, info, retcode);
807 /* Initializes the md.*_offset members, so we are able to find
808 * the on disk meta data.
810 * We currently have two possible layouts:
812 * |----------- md_size_sect ------------------|
813 * [ 4k superblock ][ activity log ][ Bitmap ]
815 * | bm_offset = al_offset + X |
816 * ==> bitmap sectors = md_size_sect - bm_offset
819 * |----------- md_size_sect ------------------|
820 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
822 * | bm_offset = al_offset - Y |
823 * ==> bitmap sectors = Y = al_offset - bm_offset
825 * Activity log size used to be fixed 32kB,
826 * but is about to become configurable.
828 static void drbd_md_set_sector_offsets(struct drbd_device *device,
829 struct drbd_backing_dev *bdev)
831 sector_t md_size_sect = 0;
832 unsigned int al_size_sect = bdev->md.al_size_4k * 8;
834 bdev->md.md_offset = drbd_md_ss(bdev);
836 switch (bdev->md.meta_dev_idx) {
838 /* v07 style fixed size indexed meta data */
839 bdev->md.md_size_sect = MD_128MB_SECT;
840 bdev->md.al_offset = MD_4kB_SECT;
841 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
843 case DRBD_MD_INDEX_FLEX_EXT:
844 /* just occupy the full device; unit: sectors */
845 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
846 bdev->md.al_offset = MD_4kB_SECT;
847 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
849 case DRBD_MD_INDEX_INTERNAL:
850 case DRBD_MD_INDEX_FLEX_INT:
851 /* al size is still fixed */
852 bdev->md.al_offset = -al_size_sect;
853 /* we need (slightly less than) ~ this much bitmap sectors: */
854 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
855 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
856 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
857 md_size_sect = ALIGN(md_size_sect, 8);
859 /* plus the "drbd meta data super block",
860 * and the activity log; */
861 md_size_sect += MD_4kB_SECT + al_size_sect;
863 bdev->md.md_size_sect = md_size_sect;
864 /* bitmap offset is adjusted by 'super' block size */
865 bdev->md.bm_offset = -md_size_sect + MD_4kB_SECT;
870 /* input size is expected to be in KB */
871 char *ppsize(char *buf, unsigned long long size)
873 /* Needs 9 bytes at max including trailing NUL:
874 * -1ULL ==> "16384 EB" */
875 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
877 while (size >= 10000 && base < sizeof(units)-1) {
879 size = (size >> 10) + !!(size & (1<<9));
882 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
887 /* there is still a theoretical deadlock when called from receiver
888 * on an D_INCONSISTENT R_PRIMARY:
889 * remote READ does inc_ap_bio, receiver would need to receive answer
890 * packet from remote to dec_ap_bio again.
891 * receiver receive_sizes(), comes here,
892 * waits for ap_bio_cnt == 0. -> deadlock.
893 * but this cannot happen, actually, because:
894 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
895 * (not connected, or bad/no disk on peer):
896 * see drbd_fail_request_early, ap_bio_cnt is zero.
897 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
898 * peer may not initiate a resize.
900 /* Note these are not to be confused with
901 * drbd_adm_suspend_io/drbd_adm_resume_io,
902 * which are (sub) state changes triggered by admin (drbdsetup),
903 * and can be long lived.
904 * This changes an device->flag, is triggered by drbd internals,
905 * and should be short-lived. */
906 /* It needs to be a counter, since multiple threads might
907 independently suspend and resume IO. */
908 void drbd_suspend_io(struct drbd_device *device)
910 atomic_inc(&device->suspend_cnt);
911 if (drbd_suspended(device))
913 wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt));
916 void drbd_resume_io(struct drbd_device *device)
918 if (atomic_dec_and_test(&device->suspend_cnt))
919 wake_up(&device->misc_wait);
923 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
924 * @device: DRBD device.
926 * Returns 0 on success, negative return values indicate errors.
927 * You should call drbd_md_sync() after calling this function.
929 enum determine_dev_size
930 drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local)
932 struct md_offsets_and_sizes {
933 u64 last_agreed_sect;
940 u32 al_stripe_size_4k;
942 sector_t u_size, size;
943 struct drbd_md *md = &device->ldev->md;
946 int md_moved, la_size_changed;
947 enum determine_dev_size rv = DS_UNCHANGED;
949 /* We may change the on-disk offsets of our meta data below. Lock out
950 * anything that may cause meta data IO, to avoid acting on incomplete
951 * layout changes or scribbling over meta data that is in the process
954 * Move is not exactly correct, btw, currently we have all our meta
955 * data in core memory, to "move" it we just write it all out, there
957 drbd_suspend_io(device);
958 buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */
960 drbd_resume_io(device);
964 /* remember current offset and sizes */
965 prev.last_agreed_sect = md->la_size_sect;
966 prev.md_offset = md->md_offset;
967 prev.al_offset = md->al_offset;
968 prev.bm_offset = md->bm_offset;
969 prev.md_size_sect = md->md_size_sect;
970 prev.al_stripes = md->al_stripes;
971 prev.al_stripe_size_4k = md->al_stripe_size_4k;
974 /* rs is non NULL if we should change the AL layout only */
975 md->al_stripes = rs->al_stripes;
976 md->al_stripe_size_4k = rs->al_stripe_size / 4;
977 md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4;
980 drbd_md_set_sector_offsets(device, device->ldev);
983 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
985 size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED);
987 if (size < prev.last_agreed_sect) {
988 if (rs && u_size == 0) {
989 /* Remove "rs &&" later. This check should always be active, but
990 right now the receiver expects the permissive behavior */
991 drbd_warn(device, "Implicit shrink not allowed. "
992 "Use --size=%llus for explicit shrink.\n",
993 (unsigned long long)size);
994 rv = DS_ERROR_SHRINK;
997 rv = DS_ERROR_SPACE_MD;
998 if (rv != DS_UNCHANGED)
1002 if (get_capacity(device->vdisk) != size ||
1003 drbd_bm_capacity(device) != size) {
1005 err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC));
1006 if (unlikely(err)) {
1007 /* currently there is only one error: ENOMEM! */
1008 size = drbd_bm_capacity(device);
1010 drbd_err(device, "OUT OF MEMORY! "
1011 "Could not allocate bitmap!\n");
1013 drbd_err(device, "BM resizing failed. "
1014 "Leaving size unchanged\n");
1018 /* racy, see comments above. */
1019 drbd_set_my_capacity(device, size);
1020 md->la_size_sect = size;
1025 la_size_changed = (prev.last_agreed_sect != md->la_size_sect);
1027 md_moved = prev.md_offset != md->md_offset
1028 || prev.md_size_sect != md->md_size_sect;
1030 if (la_size_changed || md_moved || rs) {
1033 /* We do some synchronous IO below, which may take some time.
1034 * Clear the timer, to avoid scary "timer expired!" messages,
1035 * "Superblock" is written out at least twice below, anyways. */
1036 del_timer(&device->md_sync_timer);
1038 /* We won't change the "al-extents" setting, we just may need
1039 * to move the on-disk location of the activity log ringbuffer.
1040 * Lock for transaction is good enough, it may well be "dirty"
1041 * or even "starving". */
1042 wait_event(device->al_wait, lc_try_lock_for_transaction(device->act_log));
1044 /* mark current on-disk bitmap and activity log as unreliable */
1045 prev_flags = md->flags;
1046 md->flags |= MDF_FULL_SYNC | MDF_AL_DISABLED;
1047 drbd_md_write(device, buffer);
1049 drbd_al_initialize(device, buffer);
1051 drbd_info(device, "Writing the whole bitmap, %s\n",
1052 la_size_changed && md_moved ? "size changed and md moved" :
1053 la_size_changed ? "size changed" : "md moved");
1054 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
1055 drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
1056 "size changed", BM_LOCKED_MASK, NULL);
1058 /* on-disk bitmap and activity log is authoritative again
1059 * (unless there was an IO error meanwhile...) */
1060 md->flags = prev_flags;
1061 drbd_md_write(device, buffer);
1064 drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
1065 md->al_stripes, md->al_stripe_size_4k * 4);
1068 if (size > prev.last_agreed_sect)
1069 rv = prev.last_agreed_sect ? DS_GREW : DS_GREW_FROM_ZERO;
1070 if (size < prev.last_agreed_sect)
1075 /* restore previous offset and sizes */
1076 md->la_size_sect = prev.last_agreed_sect;
1077 md->md_offset = prev.md_offset;
1078 md->al_offset = prev.al_offset;
1079 md->bm_offset = prev.bm_offset;
1080 md->md_size_sect = prev.md_size_sect;
1081 md->al_stripes = prev.al_stripes;
1082 md->al_stripe_size_4k = prev.al_stripe_size_4k;
1083 md->al_size_4k = (u64)prev.al_stripes * prev.al_stripe_size_4k;
1085 lc_unlock(device->act_log);
1086 wake_up(&device->al_wait);
1087 drbd_md_put_buffer(device);
1088 drbd_resume_io(device);
1094 drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev,
1095 sector_t u_size, int assume_peer_has_space)
1097 sector_t p_size = device->p_size; /* partner's disk size. */
1098 sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
1099 sector_t m_size; /* my size */
1102 m_size = drbd_get_max_capacity(bdev);
1104 if (device->state.conn < C_CONNECTED && assume_peer_has_space) {
1105 drbd_warn(device, "Resize while not connected was forced by the user!\n");
1109 if (p_size && m_size) {
1110 size = min_t(sector_t, p_size, m_size);
1113 size = la_size_sect;
1114 if (m_size && m_size < size)
1116 if (p_size && p_size < size)
1127 drbd_err(device, "Both nodes diskless!\n");
1131 drbd_err(device, "Requested disk size is too big (%lu > %lu)\n",
1132 (unsigned long)u_size>>1, (unsigned long)size>>1);
1141 * drbd_check_al_size() - Ensures that the AL is of the right size
1142 * @device: DRBD device.
1144 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1145 * failed, and 0 on success. You should call drbd_md_sync() after you called
1148 static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1150 struct lru_cache *n, *t;
1151 struct lc_element *e;
1152 unsigned int in_use;
1155 if (device->act_log &&
1156 device->act_log->nr_elements == dc->al_extents)
1160 t = device->act_log;
1161 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
1162 dc->al_extents, sizeof(struct lc_element), 0);
1165 drbd_err(device, "Cannot allocate act_log lru!\n");
1168 spin_lock_irq(&device->al_lock);
1170 for (i = 0; i < t->nr_elements; i++) {
1171 e = lc_element_by_index(t, i);
1173 drbd_err(device, "refcnt(%d)==%d\n",
1174 e->lc_number, e->refcnt);
1175 in_use += e->refcnt;
1179 device->act_log = n;
1180 spin_unlock_irq(&device->al_lock);
1182 drbd_err(device, "Activity log still in use!\n");
1188 drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */
1192 static void blk_queue_discard_granularity(struct request_queue *q, unsigned int granularity)
1194 q->limits.discard_granularity = granularity;
1197 static unsigned int drbd_max_discard_sectors(struct drbd_connection *connection)
1199 /* when we introduced REQ_WRITE_SAME support, we also bumped
1200 * our maximum supported batch bio size used for discards. */
1201 if (connection->agreed_features & DRBD_FF_WSAME)
1202 return DRBD_MAX_BBIO_SECTORS;
1203 /* before, with DRBD <= 8.4.6, we only allowed up to one AL_EXTENT_SIZE. */
1204 return AL_EXTENT_SIZE >> 9;
1207 static void decide_on_discard_support(struct drbd_device *device,
1208 struct drbd_backing_dev *bdev)
1210 struct drbd_connection *connection =
1211 first_peer_device(device)->connection;
1212 struct request_queue *q = device->rq_queue;
1213 unsigned int max_discard_sectors;
1215 if (bdev && !bdev_max_discard_sectors(bdev->backing_bdev))
1218 if (connection->cstate >= C_CONNECTED &&
1219 !(connection->agreed_features & DRBD_FF_TRIM)) {
1220 drbd_info(connection,
1221 "peer DRBD too old, does not support TRIM: disabling discards\n");
1226 * We don't care for the granularity, really.
1228 * Stacking limits below should fix it for the local device. Whether or
1229 * not it is a suitable granularity on the remote device is not our
1230 * problem, really. If you care, you need to use devices with similar
1231 * topology on all peers.
1233 blk_queue_discard_granularity(q, 512);
1234 max_discard_sectors = drbd_max_discard_sectors(connection);
1235 blk_queue_max_discard_sectors(q, max_discard_sectors);
1236 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
1240 blk_queue_discard_granularity(q, 0);
1241 blk_queue_max_discard_sectors(q, 0);
1244 static void fixup_write_zeroes(struct drbd_device *device, struct request_queue *q)
1246 /* Fixup max_write_zeroes_sectors after blk_stack_limits():
1247 * if we can handle "zeroes" efficiently on the protocol,
1248 * we want to do that, even if our backend does not announce
1249 * max_write_zeroes_sectors itself. */
1250 struct drbd_connection *connection = first_peer_device(device)->connection;
1251 /* If the peer announces WZEROES support, use it. Otherwise, rather
1252 * send explicit zeroes than rely on some discard-zeroes-data magic. */
1253 if (connection->agreed_features & DRBD_FF_WZEROES)
1254 q->limits.max_write_zeroes_sectors = DRBD_MAX_BBIO_SECTORS;
1256 q->limits.max_write_zeroes_sectors = 0;
1259 static void fixup_discard_support(struct drbd_device *device, struct request_queue *q)
1261 unsigned int max_discard = device->rq_queue->limits.max_discard_sectors;
1262 unsigned int discard_granularity =
1263 device->rq_queue->limits.discard_granularity >> SECTOR_SHIFT;
1265 if (discard_granularity > max_discard) {
1266 blk_queue_discard_granularity(q, 0);
1267 blk_queue_max_discard_sectors(q, 0);
1271 static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev,
1272 unsigned int max_bio_size, struct o_qlim *o)
1274 struct request_queue * const q = device->rq_queue;
1275 unsigned int max_hw_sectors = max_bio_size >> 9;
1276 unsigned int max_segments = 0;
1277 struct request_queue *b = NULL;
1278 struct disk_conf *dc;
1281 b = bdev->backing_bdev->bd_disk->queue;
1283 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1285 dc = rcu_dereference(device->ldev->disk_conf);
1286 max_segments = dc->max_bio_bvecs;
1289 blk_set_stacking_limits(&q->limits);
1292 blk_queue_max_hw_sectors(q, max_hw_sectors);
1293 /* This is the workaround for "bio would need to, but cannot, be split" */
1294 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1295 blk_queue_segment_boundary(q, PAGE_SIZE-1);
1296 decide_on_discard_support(device, bdev);
1299 blk_stack_limits(&q->limits, &b->limits, 0);
1300 disk_update_readahead(device->vdisk);
1302 fixup_write_zeroes(device, q);
1303 fixup_discard_support(device, q);
1306 void drbd_reconsider_queue_parameters(struct drbd_device *device, struct drbd_backing_dev *bdev, struct o_qlim *o)
1308 unsigned int now, new, local, peer;
1310 now = queue_max_hw_sectors(device->rq_queue) << 9;
1311 local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */
1312 peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */
1315 local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9;
1316 device->local_max_bio_size = local;
1318 local = min(local, DRBD_MAX_BIO_SIZE);
1320 /* We may ignore peer limits if the peer is modern enough.
1321 Because new from 8.3.8 onwards the peer can use multiple
1322 BIOs for a single peer_request */
1323 if (device->state.conn >= C_WF_REPORT_PARAMS) {
1324 if (first_peer_device(device)->connection->agreed_pro_version < 94)
1325 peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1326 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1327 else if (first_peer_device(device)->connection->agreed_pro_version == 94)
1328 peer = DRBD_MAX_SIZE_H80_PACKET;
1329 else if (first_peer_device(device)->connection->agreed_pro_version < 100)
1330 peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */
1332 peer = DRBD_MAX_BIO_SIZE;
1334 /* We may later detach and re-attach on a disconnected Primary.
1335 * Avoid this setting to jump back in that case.
1336 * We want to store what we know the peer DRBD can handle,
1337 * not what the peer IO backend can handle. */
1338 if (peer > device->peer_max_bio_size)
1339 device->peer_max_bio_size = peer;
1341 new = min(local, peer);
1343 if (device->state.role == R_PRIMARY && new < now)
1344 drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1347 drbd_info(device, "max BIO size = %u\n", new);
1349 drbd_setup_queue_param(device, bdev, new, o);
1352 /* Starts the worker thread */
1353 static void conn_reconfig_start(struct drbd_connection *connection)
1355 drbd_thread_start(&connection->worker);
1356 drbd_flush_workqueue(&connection->sender_work);
1359 /* if still unconfigured, stops worker again. */
1360 static void conn_reconfig_done(struct drbd_connection *connection)
1363 spin_lock_irq(&connection->resource->req_lock);
1364 stop_threads = conn_all_vols_unconf(connection) &&
1365 connection->cstate == C_STANDALONE;
1366 spin_unlock_irq(&connection->resource->req_lock);
1368 /* ack_receiver thread and ack_sender workqueue are implicitly
1369 * stopped by receiver in conn_disconnect() */
1370 drbd_thread_stop(&connection->receiver);
1371 drbd_thread_stop(&connection->worker);
1375 /* Make sure IO is suspended before calling this function(). */
1376 static void drbd_suspend_al(struct drbd_device *device)
1380 if (!lc_try_lock(device->act_log)) {
1381 drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1385 drbd_al_shrink(device);
1386 spin_lock_irq(&device->resource->req_lock);
1387 if (device->state.conn < C_CONNECTED)
1388 s = !test_and_set_bit(AL_SUSPENDED, &device->flags);
1389 spin_unlock_irq(&device->resource->req_lock);
1390 lc_unlock(device->act_log);
1393 drbd_info(device, "Suspended AL updates\n");
1397 static bool should_set_defaults(struct genl_info *info)
1399 struct drbd_genlmsghdr *dh = genl_info_userhdr(info);
1401 return 0 != (dh->flags & DRBD_GENL_F_SET_DEFAULTS);
1404 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1406 /* This is limited by 16 bit "slot" numbers,
1407 * and by available on-disk context storage.
1409 * Also (u16)~0 is special (denotes a "free" extent).
1411 * One transaction occupies one 4kB on-disk block,
1412 * we have n such blocks in the on disk ring buffer,
1413 * the "current" transaction may fail (n-1),
1414 * and there is 919 slot numbers context information per transaction.
1416 * 72 transaction blocks amounts to more than 2**16 context slots,
1417 * so cap there first.
1419 const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1420 const unsigned int sufficient_on_disk =
1421 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1422 /AL_CONTEXT_PER_TRANSACTION;
1424 unsigned int al_size_4k = bdev->md.al_size_4k;
1426 if (al_size_4k > sufficient_on_disk)
1429 return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1432 static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1434 return a->disk_barrier != b->disk_barrier ||
1435 a->disk_flushes != b->disk_flushes ||
1436 a->disk_drain != b->disk_drain;
1439 static void sanitize_disk_conf(struct drbd_device *device, struct disk_conf *disk_conf,
1440 struct drbd_backing_dev *nbc)
1442 struct block_device *bdev = nbc->backing_bdev;
1444 if (disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1445 disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1446 if (disk_conf->al_extents > drbd_al_extents_max(nbc))
1447 disk_conf->al_extents = drbd_al_extents_max(nbc);
1449 if (!bdev_max_discard_sectors(bdev)) {
1450 if (disk_conf->rs_discard_granularity) {
1451 disk_conf->rs_discard_granularity = 0; /* disable feature */
1452 drbd_info(device, "rs_discard_granularity feature disabled\n");
1456 if (disk_conf->rs_discard_granularity) {
1457 int orig_value = disk_conf->rs_discard_granularity;
1458 sector_t discard_size = bdev_max_discard_sectors(bdev) << 9;
1459 unsigned int discard_granularity = bdev_discard_granularity(bdev);
1462 if (discard_granularity > disk_conf->rs_discard_granularity)
1463 disk_conf->rs_discard_granularity = discard_granularity;
1465 remainder = disk_conf->rs_discard_granularity %
1466 discard_granularity;
1467 disk_conf->rs_discard_granularity += remainder;
1469 if (disk_conf->rs_discard_granularity > discard_size)
1470 disk_conf->rs_discard_granularity = discard_size;
1472 if (disk_conf->rs_discard_granularity != orig_value)
1473 drbd_info(device, "rs_discard_granularity changed to %d\n",
1474 disk_conf->rs_discard_granularity);
1478 static int disk_opts_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1482 if (device->act_log &&
1483 device->act_log->nr_elements == dc->al_extents)
1486 drbd_suspend_io(device);
1487 /* If IO completion is currently blocked, we would likely wait
1488 * "forever" for the activity log to become unused. So we don't. */
1489 if (atomic_read(&device->ap_bio_cnt))
1492 wait_event(device->al_wait, lc_try_lock(device->act_log));
1493 drbd_al_shrink(device);
1494 err = drbd_check_al_size(device, dc);
1495 lc_unlock(device->act_log);
1496 wake_up(&device->al_wait);
1498 drbd_resume_io(device);
1502 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1504 struct drbd_config_context adm_ctx;
1505 enum drbd_ret_code retcode;
1506 struct drbd_device *device;
1507 struct disk_conf *new_disk_conf, *old_disk_conf;
1508 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1510 unsigned int fifo_size;
1512 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1513 if (!adm_ctx.reply_skb)
1515 if (retcode != NO_ERROR)
1518 device = adm_ctx.device;
1519 mutex_lock(&adm_ctx.resource->adm_mutex);
1521 /* we also need a disk
1522 * to change the options on */
1523 if (!get_ldev(device)) {
1524 retcode = ERR_NO_DISK;
1528 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1529 if (!new_disk_conf) {
1530 retcode = ERR_NOMEM;
1534 mutex_lock(&device->resource->conf_update);
1535 old_disk_conf = device->ldev->disk_conf;
1536 *new_disk_conf = *old_disk_conf;
1537 if (should_set_defaults(info))
1538 set_disk_conf_defaults(new_disk_conf);
1540 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1541 if (err && err != -ENOMSG) {
1542 retcode = ERR_MANDATORY_TAG;
1543 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1547 if (!expect(device, new_disk_conf->resync_rate >= 1))
1548 new_disk_conf->resync_rate = 1;
1550 sanitize_disk_conf(device, new_disk_conf, device->ldev);
1552 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1553 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1555 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1556 if (fifo_size != device->rs_plan_s->size) {
1557 new_plan = fifo_alloc(fifo_size);
1559 drbd_err(device, "kmalloc of fifo_buffer failed");
1560 retcode = ERR_NOMEM;
1565 err = disk_opts_check_al_size(device, new_disk_conf);
1567 /* Could be just "busy". Ignore?
1568 * Introduce dedicated error code? */
1569 drbd_msg_put_info(adm_ctx.reply_skb,
1570 "Try again without changing current al-extents setting");
1571 retcode = ERR_NOMEM;
1575 lock_all_resources();
1576 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1577 if (retcode == NO_ERROR) {
1578 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1579 drbd_resync_after_changed(device);
1581 unlock_all_resources();
1583 if (retcode != NO_ERROR)
1587 old_plan = device->rs_plan_s;
1588 rcu_assign_pointer(device->rs_plan_s, new_plan);
1591 mutex_unlock(&device->resource->conf_update);
1593 if (new_disk_conf->al_updates)
1594 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1596 device->ldev->md.flags |= MDF_AL_DISABLED;
1598 if (new_disk_conf->md_flushes)
1599 clear_bit(MD_NO_FUA, &device->flags);
1601 set_bit(MD_NO_FUA, &device->flags);
1603 if (write_ordering_changed(old_disk_conf, new_disk_conf))
1604 drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH);
1606 if (old_disk_conf->discard_zeroes_if_aligned !=
1607 new_disk_conf->discard_zeroes_if_aligned)
1608 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
1610 drbd_md_sync(device);
1612 if (device->state.conn >= C_CONNECTED) {
1613 struct drbd_peer_device *peer_device;
1615 for_each_peer_device(peer_device, device)
1616 drbd_send_sync_param(peer_device);
1619 kvfree_rcu_mightsleep(old_disk_conf);
1621 mod_timer(&device->request_timer, jiffies + HZ);
1625 mutex_unlock(&device->resource->conf_update);
1627 kfree(new_disk_conf);
1632 mutex_unlock(&adm_ctx.resource->adm_mutex);
1634 drbd_adm_finish(&adm_ctx, info, retcode);
1638 static struct block_device *open_backing_dev(struct drbd_device *device,
1639 const char *bdev_path, void *claim_ptr, bool do_bd_link)
1641 struct block_device *bdev;
1644 bdev = blkdev_get_by_path(bdev_path, BLK_OPEN_READ | BLK_OPEN_WRITE,
1647 drbd_err(device, "open(\"%s\") failed with %ld\n",
1648 bdev_path, PTR_ERR(bdev));
1655 err = bd_link_disk_holder(bdev, device->vdisk);
1657 blkdev_put(bdev, claim_ptr);
1658 drbd_err(device, "bd_link_disk_holder(\"%s\", ...) failed with %d\n",
1660 bdev = ERR_PTR(err);
1665 static int open_backing_devices(struct drbd_device *device,
1666 struct disk_conf *new_disk_conf,
1667 struct drbd_backing_dev *nbc)
1669 struct block_device *bdev;
1671 bdev = open_backing_dev(device, new_disk_conf->backing_dev, device, true);
1673 return ERR_OPEN_DISK;
1674 nbc->backing_bdev = bdev;
1677 * meta_dev_idx >= 0: external fixed size, possibly multiple
1678 * drbd sharing one meta device. TODO in that case, paranoia
1679 * check that [md_bdev, meta_dev_idx] is not yet used by some
1680 * other drbd minor! (if you use drbd.conf + drbdadm, that
1681 * should check it for you already; but if you don't, or
1682 * someone fooled it, we need to double check here)
1684 bdev = open_backing_dev(device, new_disk_conf->meta_dev,
1685 /* claim ptr: device, if claimed exclusively; shared drbd_m_holder,
1686 * if potentially shared with other drbd minors */
1687 (new_disk_conf->meta_dev_idx < 0) ? (void*)device : (void*)drbd_m_holder,
1688 /* avoid double bd_claim_by_disk() for the same (source,target) tuple,
1689 * as would happen with internal metadata. */
1690 (new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_FLEX_INT &&
1691 new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_INTERNAL));
1693 return ERR_OPEN_MD_DISK;
1694 nbc->md_bdev = bdev;
1698 static void close_backing_dev(struct drbd_device *device, struct block_device *bdev,
1699 void *claim_ptr, bool do_bd_unlink)
1704 bd_unlink_disk_holder(bdev, device->vdisk);
1705 blkdev_put(bdev, claim_ptr);
1708 void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev)
1713 close_backing_dev(device, ldev->md_bdev,
1714 ldev->md.meta_dev_idx < 0 ?
1715 (void *)device : (void *)drbd_m_holder,
1716 ldev->md_bdev != ldev->backing_bdev);
1717 close_backing_dev(device, ldev->backing_bdev, device, true);
1719 kfree(ldev->disk_conf);
1723 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1725 struct drbd_config_context adm_ctx;
1726 struct drbd_device *device;
1727 struct drbd_peer_device *peer_device;
1728 struct drbd_connection *connection;
1730 enum drbd_ret_code retcode;
1731 enum determine_dev_size dd;
1732 sector_t max_possible_sectors;
1733 sector_t min_md_device_sectors;
1734 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1735 struct disk_conf *new_disk_conf = NULL;
1736 struct lru_cache *resync_lru = NULL;
1737 struct fifo_buffer *new_plan = NULL;
1738 union drbd_state ns, os;
1739 enum drbd_state_rv rv;
1740 struct net_conf *nc;
1742 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1743 if (!adm_ctx.reply_skb)
1745 if (retcode != NO_ERROR)
1748 device = adm_ctx.device;
1749 mutex_lock(&adm_ctx.resource->adm_mutex);
1750 peer_device = first_peer_device(device);
1751 connection = peer_device->connection;
1752 conn_reconfig_start(connection);
1754 /* if you want to reconfigure, please tear down first */
1755 if (device->state.disk > D_DISKLESS) {
1756 retcode = ERR_DISK_CONFIGURED;
1759 /* It may just now have detached because of IO error. Make sure
1760 * drbd_ldev_destroy is done already, we may end up here very fast,
1761 * e.g. if someone calls attach from the on-io-error handler,
1762 * to realize a "hot spare" feature (not that I'd recommend that) */
1763 wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1765 /* make sure there is no leftover from previous force-detach attempts */
1766 clear_bit(FORCE_DETACH, &device->flags);
1767 clear_bit(WAS_IO_ERROR, &device->flags);
1768 clear_bit(WAS_READ_ERROR, &device->flags);
1770 /* and no leftover from previously aborted resync or verify, either */
1771 device->rs_total = 0;
1772 device->rs_failed = 0;
1773 atomic_set(&device->rs_pending_cnt, 0);
1775 /* allocation not in the IO path, drbdsetup context */
1776 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1778 retcode = ERR_NOMEM;
1781 spin_lock_init(&nbc->md.uuid_lock);
1783 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1784 if (!new_disk_conf) {
1785 retcode = ERR_NOMEM;
1788 nbc->disk_conf = new_disk_conf;
1790 set_disk_conf_defaults(new_disk_conf);
1791 err = disk_conf_from_attrs(new_disk_conf, info);
1793 retcode = ERR_MANDATORY_TAG;
1794 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1798 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1799 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1801 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1803 retcode = ERR_NOMEM;
1807 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1808 retcode = ERR_MD_IDX_INVALID;
1813 nc = rcu_dereference(connection->net_conf);
1815 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1817 retcode = ERR_STONITH_AND_PROT_A;
1823 retcode = open_backing_devices(device, new_disk_conf, nbc);
1824 if (retcode != NO_ERROR)
1827 if ((nbc->backing_bdev == nbc->md_bdev) !=
1828 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1829 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1830 retcode = ERR_MD_IDX_INVALID;
1834 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1835 1, 61, sizeof(struct bm_extent),
1836 offsetof(struct bm_extent, lce));
1838 retcode = ERR_NOMEM;
1842 /* Read our meta data super block early.
1843 * This also sets other on-disk offsets. */
1844 retcode = drbd_md_read(device, nbc);
1845 if (retcode != NO_ERROR)
1848 sanitize_disk_conf(device, new_disk_conf, nbc);
1850 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1851 drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1852 (unsigned long long) drbd_get_max_capacity(nbc),
1853 (unsigned long long) new_disk_conf->disk_size);
1854 retcode = ERR_DISK_TOO_SMALL;
1858 if (new_disk_conf->meta_dev_idx < 0) {
1859 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1860 /* at least one MB, otherwise it does not make sense */
1861 min_md_device_sectors = (2<<10);
1863 max_possible_sectors = DRBD_MAX_SECTORS;
1864 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1867 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1868 retcode = ERR_MD_DISK_TOO_SMALL;
1869 drbd_warn(device, "refusing attach: md-device too small, "
1870 "at least %llu sectors needed for this meta-disk type\n",
1871 (unsigned long long) min_md_device_sectors);
1875 /* Make sure the new disk is big enough
1876 * (we may currently be R_PRIMARY with no local disk...) */
1877 if (drbd_get_max_capacity(nbc) < get_capacity(device->vdisk)) {
1878 retcode = ERR_DISK_TOO_SMALL;
1882 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1884 if (nbc->known_size > max_possible_sectors) {
1885 drbd_warn(device, "==> truncating very big lower level device "
1886 "to currently maximum possible %llu sectors <==\n",
1887 (unsigned long long) max_possible_sectors);
1888 if (new_disk_conf->meta_dev_idx >= 0)
1889 drbd_warn(device, "==>> using internal or flexible "
1890 "meta data may help <<==\n");
1893 drbd_suspend_io(device);
1894 /* also wait for the last barrier ack. */
1895 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1896 * We need a way to either ignore barrier acks for barriers sent before a device
1897 * was attached, or a way to wait for all pending barrier acks to come in.
1898 * As barriers are counted per resource,
1899 * we'd need to suspend io on all devices of a resource.
1901 wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1902 /* and for any other previously queued work */
1903 drbd_flush_workqueue(&connection->sender_work);
1905 rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1906 retcode = (enum drbd_ret_code)rv;
1907 drbd_resume_io(device);
1908 if (rv < SS_SUCCESS)
1911 if (!get_ldev_if_state(device, D_ATTACHING))
1912 goto force_diskless;
1914 if (!device->bitmap) {
1915 if (drbd_bm_init(device)) {
1916 retcode = ERR_NOMEM;
1917 goto force_diskless_dec;
1921 if (device->state.pdsk != D_UP_TO_DATE && device->ed_uuid &&
1922 (device->state.role == R_PRIMARY || device->state.peer == R_PRIMARY) &&
1923 (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1924 drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1925 (unsigned long long)device->ed_uuid);
1926 retcode = ERR_DATA_NOT_CURRENT;
1927 goto force_diskless_dec;
1930 /* Since we are diskless, fix the activity log first... */
1931 if (drbd_check_al_size(device, new_disk_conf)) {
1932 retcode = ERR_NOMEM;
1933 goto force_diskless_dec;
1936 /* Prevent shrinking of consistent devices ! */
1938 unsigned long long nsz = drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0);
1939 unsigned long long eff = nbc->md.la_size_sect;
1940 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && nsz < eff) {
1941 if (nsz == nbc->disk_conf->disk_size) {
1942 drbd_warn(device, "truncating a consistent device during attach (%llu < %llu)\n", nsz, eff);
1944 drbd_warn(device, "refusing to truncate a consistent device (%llu < %llu)\n", nsz, eff);
1945 drbd_msg_sprintf_info(adm_ctx.reply_skb,
1946 "To-be-attached device has last effective > current size, and is consistent\n"
1947 "(%llu > %llu sectors). Refusing to attach.", eff, nsz);
1948 retcode = ERR_IMPLICIT_SHRINK;
1949 goto force_diskless_dec;
1954 lock_all_resources();
1955 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1956 if (retcode != NO_ERROR) {
1957 unlock_all_resources();
1958 goto force_diskless_dec;
1961 /* Reset the "barriers don't work" bits here, then force meta data to
1962 * be written, to ensure we determine if barriers are supported. */
1963 if (new_disk_conf->md_flushes)
1964 clear_bit(MD_NO_FUA, &device->flags);
1966 set_bit(MD_NO_FUA, &device->flags);
1968 /* Point of no return reached.
1969 * Devices and memory are no longer released by error cleanup below.
1970 * now device takes over responsibility, and the state engine should
1971 * clean it up somewhere. */
1972 D_ASSERT(device, device->ldev == NULL);
1974 device->resync = resync_lru;
1975 device->rs_plan_s = new_plan;
1978 new_disk_conf = NULL;
1981 drbd_resync_after_changed(device);
1982 drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH);
1983 unlock_all_resources();
1985 if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
1986 set_bit(CRASHED_PRIMARY, &device->flags);
1988 clear_bit(CRASHED_PRIMARY, &device->flags);
1990 if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1991 !(device->state.role == R_PRIMARY && device->resource->susp_nod))
1992 set_bit(CRASHED_PRIMARY, &device->flags);
1994 device->send_cnt = 0;
1995 device->recv_cnt = 0;
1996 device->read_cnt = 0;
1997 device->writ_cnt = 0;
1999 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
2001 /* If I am currently not R_PRIMARY,
2002 * but meta data primary indicator is set,
2003 * I just now recover from a hard crash,
2004 * and have been R_PRIMARY before that crash.
2006 * Now, if I had no connection before that crash
2007 * (have been degraded R_PRIMARY), chances are that
2008 * I won't find my peer now either.
2010 * In that case, and _only_ in that case,
2011 * we use the degr-wfc-timeout instead of the default,
2012 * so we can automatically recover from a crash of a
2013 * degraded but active "cluster" after a certain timeout.
2015 clear_bit(USE_DEGR_WFC_T, &device->flags);
2016 if (device->state.role != R_PRIMARY &&
2017 drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
2018 !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
2019 set_bit(USE_DEGR_WFC_T, &device->flags);
2021 dd = drbd_determine_dev_size(device, 0, NULL);
2022 if (dd <= DS_ERROR) {
2023 retcode = ERR_NOMEM_BITMAP;
2024 goto force_diskless_dec;
2025 } else if (dd == DS_GREW)
2026 set_bit(RESYNC_AFTER_NEG, &device->flags);
2028 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
2029 (test_bit(CRASHED_PRIMARY, &device->flags) &&
2030 drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
2031 drbd_info(device, "Assuming that all blocks are out of sync "
2032 "(aka FullSync)\n");
2033 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2034 "set_n_write from attaching", BM_LOCKED_MASK,
2036 retcode = ERR_IO_MD_DISK;
2037 goto force_diskless_dec;
2040 if (drbd_bitmap_io(device, &drbd_bm_read,
2041 "read from attaching", BM_LOCKED_MASK,
2043 retcode = ERR_IO_MD_DISK;
2044 goto force_diskless_dec;
2048 if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
2049 drbd_suspend_al(device); /* IO is still suspended here... */
2051 spin_lock_irq(&device->resource->req_lock);
2052 os = drbd_read_state(device);
2054 /* If MDF_CONSISTENT is not set go into inconsistent state,
2055 otherwise investigate MDF_WasUpToDate...
2056 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
2057 otherwise into D_CONSISTENT state.
2059 if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
2060 if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
2061 ns.disk = D_CONSISTENT;
2063 ns.disk = D_OUTDATED;
2065 ns.disk = D_INCONSISTENT;
2068 if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
2069 ns.pdsk = D_OUTDATED;
2072 if (ns.disk == D_CONSISTENT &&
2073 (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
2074 ns.disk = D_UP_TO_DATE;
2076 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
2077 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
2078 this point, because drbd_request_state() modifies these
2081 if (rcu_dereference(device->ldev->disk_conf)->al_updates)
2082 device->ldev->md.flags &= ~MDF_AL_DISABLED;
2084 device->ldev->md.flags |= MDF_AL_DISABLED;
2088 /* In case we are C_CONNECTED postpone any decision on the new disk
2089 state after the negotiation phase. */
2090 if (device->state.conn == C_CONNECTED) {
2091 device->new_state_tmp.i = ns.i;
2093 ns.disk = D_NEGOTIATING;
2095 /* We expect to receive up-to-date UUIDs soon.
2096 To avoid a race in receive_state, free p_uuid while
2097 holding req_lock. I.e. atomic with the state change */
2098 kfree(device->p_uuid);
2099 device->p_uuid = NULL;
2102 rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
2103 spin_unlock_irq(&device->resource->req_lock);
2105 if (rv < SS_SUCCESS)
2106 goto force_diskless_dec;
2108 mod_timer(&device->request_timer, jiffies + HZ);
2110 if (device->state.role == R_PRIMARY)
2111 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
2113 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
2115 drbd_md_mark_dirty(device);
2116 drbd_md_sync(device);
2118 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
2120 conn_reconfig_done(connection);
2121 mutex_unlock(&adm_ctx.resource->adm_mutex);
2122 drbd_adm_finish(&adm_ctx, info, retcode);
2128 drbd_force_state(device, NS(disk, D_DISKLESS));
2129 drbd_md_sync(device);
2131 conn_reconfig_done(connection);
2133 close_backing_dev(device, nbc->md_bdev,
2134 nbc->disk_conf->meta_dev_idx < 0 ?
2135 (void *)device : (void *)drbd_m_holder,
2136 nbc->md_bdev != nbc->backing_bdev);
2137 close_backing_dev(device, nbc->backing_bdev, device, true);
2140 kfree(new_disk_conf);
2141 lc_destroy(resync_lru);
2143 mutex_unlock(&adm_ctx.resource->adm_mutex);
2145 drbd_adm_finish(&adm_ctx, info, retcode);
2149 static int adm_detach(struct drbd_device *device, int force)
2152 set_bit(FORCE_DETACH, &device->flags);
2153 drbd_force_state(device, NS(disk, D_FAILED));
2157 return drbd_request_detach_interruptible(device);
2160 /* Detaching the disk is a process in multiple stages. First we need to lock
2161 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
2162 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
2163 * internal references as well.
2164 * Only then we have finally detached. */
2165 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
2167 struct drbd_config_context adm_ctx;
2168 enum drbd_ret_code retcode;
2169 struct detach_parms parms = { };
2172 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2173 if (!adm_ctx.reply_skb)
2175 if (retcode != NO_ERROR)
2178 if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
2179 err = detach_parms_from_attrs(&parms, info);
2181 retcode = ERR_MANDATORY_TAG;
2182 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2187 mutex_lock(&adm_ctx.resource->adm_mutex);
2188 retcode = adm_detach(adm_ctx.device, parms.force_detach);
2189 mutex_unlock(&adm_ctx.resource->adm_mutex);
2191 drbd_adm_finish(&adm_ctx, info, retcode);
2195 static bool conn_resync_running(struct drbd_connection *connection)
2197 struct drbd_peer_device *peer_device;
2202 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2203 struct drbd_device *device = peer_device->device;
2204 if (device->state.conn == C_SYNC_SOURCE ||
2205 device->state.conn == C_SYNC_TARGET ||
2206 device->state.conn == C_PAUSED_SYNC_S ||
2207 device->state.conn == C_PAUSED_SYNC_T) {
2217 static bool conn_ov_running(struct drbd_connection *connection)
2219 struct drbd_peer_device *peer_device;
2224 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2225 struct drbd_device *device = peer_device->device;
2226 if (device->state.conn == C_VERIFY_S ||
2227 device->state.conn == C_VERIFY_T) {
2237 static enum drbd_ret_code
2238 _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2240 struct drbd_peer_device *peer_device;
2243 if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2244 if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2245 return ERR_NEED_APV_100;
2247 if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2248 return ERR_NEED_APV_100;
2250 if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2251 return ERR_NEED_APV_100;
2254 if (!new_net_conf->two_primaries &&
2255 conn_highest_role(connection) == R_PRIMARY &&
2256 conn_highest_peer(connection) == R_PRIMARY)
2257 return ERR_NEED_ALLOW_TWO_PRI;
2259 if (new_net_conf->two_primaries &&
2260 (new_net_conf->wire_protocol != DRBD_PROT_C))
2261 return ERR_NOT_PROTO_C;
2263 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2264 struct drbd_device *device = peer_device->device;
2265 if (get_ldev(device)) {
2266 enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2268 if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2269 return ERR_STONITH_AND_PROT_A;
2271 if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2272 return ERR_DISCARD_IMPOSSIBLE;
2275 if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2276 return ERR_CONG_NOT_PROTO_A;
2281 static enum drbd_ret_code
2282 check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2284 enum drbd_ret_code rv;
2285 struct drbd_peer_device *peer_device;
2289 rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2292 /* connection->peer_devices protected by genl_lock() here */
2293 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2294 struct drbd_device *device = peer_device->device;
2295 if (!device->bitmap) {
2296 if (drbd_bm_init(device))
2305 struct crypto_shash *verify_tfm;
2306 struct crypto_shash *csums_tfm;
2307 struct crypto_shash *cram_hmac_tfm;
2308 struct crypto_shash *integrity_tfm;
2312 alloc_shash(struct crypto_shash **tfm, char *tfm_name, int err_alg)
2317 *tfm = crypto_alloc_shash(tfm_name, 0, 0);
2326 static enum drbd_ret_code
2327 alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2329 char hmac_name[CRYPTO_MAX_ALG_NAME];
2330 enum drbd_ret_code rv;
2332 rv = alloc_shash(&crypto->csums_tfm, new_net_conf->csums_alg,
2336 rv = alloc_shash(&crypto->verify_tfm, new_net_conf->verify_alg,
2340 rv = alloc_shash(&crypto->integrity_tfm, new_net_conf->integrity_alg,
2344 if (new_net_conf->cram_hmac_alg[0] != 0) {
2345 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
2346 new_net_conf->cram_hmac_alg);
2348 rv = alloc_shash(&crypto->cram_hmac_tfm, hmac_name,
2355 static void free_crypto(struct crypto *crypto)
2357 crypto_free_shash(crypto->cram_hmac_tfm);
2358 crypto_free_shash(crypto->integrity_tfm);
2359 crypto_free_shash(crypto->csums_tfm);
2360 crypto_free_shash(crypto->verify_tfm);
2363 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2365 struct drbd_config_context adm_ctx;
2366 enum drbd_ret_code retcode;
2367 struct drbd_connection *connection;
2368 struct net_conf *old_net_conf, *new_net_conf = NULL;
2370 int ovr; /* online verify running */
2371 int rsr; /* re-sync running */
2372 struct crypto crypto = { };
2374 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2375 if (!adm_ctx.reply_skb)
2377 if (retcode != NO_ERROR)
2380 connection = adm_ctx.connection;
2381 mutex_lock(&adm_ctx.resource->adm_mutex);
2383 new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
2384 if (!new_net_conf) {
2385 retcode = ERR_NOMEM;
2389 conn_reconfig_start(connection);
2391 mutex_lock(&connection->data.mutex);
2392 mutex_lock(&connection->resource->conf_update);
2393 old_net_conf = connection->net_conf;
2395 if (!old_net_conf) {
2396 drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect");
2397 retcode = ERR_INVALID_REQUEST;
2401 *new_net_conf = *old_net_conf;
2402 if (should_set_defaults(info))
2403 set_net_conf_defaults(new_net_conf);
2405 err = net_conf_from_attrs_for_change(new_net_conf, info);
2406 if (err && err != -ENOMSG) {
2407 retcode = ERR_MANDATORY_TAG;
2408 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2412 retcode = check_net_options(connection, new_net_conf);
2413 if (retcode != NO_ERROR)
2416 /* re-sync running */
2417 rsr = conn_resync_running(connection);
2418 if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2419 retcode = ERR_CSUMS_RESYNC_RUNNING;
2423 /* online verify running */
2424 ovr = conn_ov_running(connection);
2425 if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2426 retcode = ERR_VERIFY_RUNNING;
2430 retcode = alloc_crypto(&crypto, new_net_conf);
2431 if (retcode != NO_ERROR)
2434 rcu_assign_pointer(connection->net_conf, new_net_conf);
2437 crypto_free_shash(connection->csums_tfm);
2438 connection->csums_tfm = crypto.csums_tfm;
2439 crypto.csums_tfm = NULL;
2442 crypto_free_shash(connection->verify_tfm);
2443 connection->verify_tfm = crypto.verify_tfm;
2444 crypto.verify_tfm = NULL;
2447 crypto_free_shash(connection->integrity_tfm);
2448 connection->integrity_tfm = crypto.integrity_tfm;
2449 if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2450 /* Do this without trying to take connection->data.mutex again. */
2451 __drbd_send_protocol(connection, P_PROTOCOL_UPDATE);
2453 crypto_free_shash(connection->cram_hmac_tfm);
2454 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2456 mutex_unlock(&connection->resource->conf_update);
2457 mutex_unlock(&connection->data.mutex);
2458 kvfree_rcu_mightsleep(old_net_conf);
2460 if (connection->cstate >= C_WF_REPORT_PARAMS) {
2461 struct drbd_peer_device *peer_device;
2464 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2465 drbd_send_sync_param(peer_device);
2471 mutex_unlock(&connection->resource->conf_update);
2472 mutex_unlock(&connection->data.mutex);
2473 free_crypto(&crypto);
2474 kfree(new_net_conf);
2476 conn_reconfig_done(connection);
2478 mutex_unlock(&adm_ctx.resource->adm_mutex);
2480 drbd_adm_finish(&adm_ctx, info, retcode);
2484 static void connection_to_info(struct connection_info *info,
2485 struct drbd_connection *connection)
2487 info->conn_connection_state = connection->cstate;
2488 info->conn_role = conn_highest_peer(connection);
2491 static void peer_device_to_info(struct peer_device_info *info,
2492 struct drbd_peer_device *peer_device)
2494 struct drbd_device *device = peer_device->device;
2496 info->peer_repl_state =
2497 max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn);
2498 info->peer_disk_state = device->state.pdsk;
2499 info->peer_resync_susp_user = device->state.user_isp;
2500 info->peer_resync_susp_peer = device->state.peer_isp;
2501 info->peer_resync_susp_dependency = device->state.aftr_isp;
2504 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2506 struct connection_info connection_info;
2507 enum drbd_notification_type flags;
2508 unsigned int peer_devices = 0;
2509 struct drbd_config_context adm_ctx;
2510 struct drbd_peer_device *peer_device;
2511 struct net_conf *old_net_conf, *new_net_conf = NULL;
2512 struct crypto crypto = { };
2513 struct drbd_resource *resource;
2514 struct drbd_connection *connection;
2515 enum drbd_ret_code retcode;
2516 enum drbd_state_rv rv;
2520 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2522 if (!adm_ctx.reply_skb)
2524 if (retcode != NO_ERROR)
2526 if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2527 drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing");
2528 retcode = ERR_INVALID_REQUEST;
2532 /* No need for _rcu here. All reconfiguration is
2533 * strictly serialized on genl_lock(). We are protected against
2534 * concurrent reconfiguration/addition/deletion */
2535 for_each_resource(resource, &drbd_resources) {
2536 for_each_connection(connection, resource) {
2537 if (nla_len(adm_ctx.my_addr) == connection->my_addr_len &&
2538 !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr,
2539 connection->my_addr_len)) {
2540 retcode = ERR_LOCAL_ADDR;
2544 if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len &&
2545 !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr,
2546 connection->peer_addr_len)) {
2547 retcode = ERR_PEER_ADDR;
2553 mutex_lock(&adm_ctx.resource->adm_mutex);
2554 connection = first_connection(adm_ctx.resource);
2555 conn_reconfig_start(connection);
2557 if (connection->cstate > C_STANDALONE) {
2558 retcode = ERR_NET_CONFIGURED;
2562 /* allocation not in the IO path, drbdsetup / netlink process context */
2563 new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL);
2564 if (!new_net_conf) {
2565 retcode = ERR_NOMEM;
2569 set_net_conf_defaults(new_net_conf);
2571 err = net_conf_from_attrs(new_net_conf, info);
2572 if (err && err != -ENOMSG) {
2573 retcode = ERR_MANDATORY_TAG;
2574 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2578 retcode = check_net_options(connection, new_net_conf);
2579 if (retcode != NO_ERROR)
2582 retcode = alloc_crypto(&crypto, new_net_conf);
2583 if (retcode != NO_ERROR)
2586 ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2588 drbd_flush_workqueue(&connection->sender_work);
2590 mutex_lock(&adm_ctx.resource->conf_update);
2591 old_net_conf = connection->net_conf;
2593 retcode = ERR_NET_CONFIGURED;
2594 mutex_unlock(&adm_ctx.resource->conf_update);
2597 rcu_assign_pointer(connection->net_conf, new_net_conf);
2599 conn_free_crypto(connection);
2600 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2601 connection->integrity_tfm = crypto.integrity_tfm;
2602 connection->csums_tfm = crypto.csums_tfm;
2603 connection->verify_tfm = crypto.verify_tfm;
2605 connection->my_addr_len = nla_len(adm_ctx.my_addr);
2606 memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2607 connection->peer_addr_len = nla_len(adm_ctx.peer_addr);
2608 memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2610 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2614 connection_to_info(&connection_info, connection);
2615 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2616 mutex_lock(¬ification_mutex);
2617 notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags);
2618 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2619 struct peer_device_info peer_device_info;
2621 peer_device_to_info(&peer_device_info, peer_device);
2622 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2623 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags);
2625 mutex_unlock(¬ification_mutex);
2626 mutex_unlock(&adm_ctx.resource->conf_update);
2629 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2630 struct drbd_device *device = peer_device->device;
2631 device->send_cnt = 0;
2632 device->recv_cnt = 0;
2636 rv = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2638 conn_reconfig_done(connection);
2639 mutex_unlock(&adm_ctx.resource->adm_mutex);
2640 drbd_adm_finish(&adm_ctx, info, rv);
2644 free_crypto(&crypto);
2645 kfree(new_net_conf);
2647 conn_reconfig_done(connection);
2648 mutex_unlock(&adm_ctx.resource->adm_mutex);
2650 drbd_adm_finish(&adm_ctx, info, retcode);
2654 static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2656 enum drbd_conns cstate;
2657 enum drbd_state_rv rv;
2660 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2661 force ? CS_HARD : 0);
2664 case SS_NOTHING_TO_DO:
2666 case SS_ALREADY_STANDALONE:
2668 case SS_PRIMARY_NOP:
2669 /* Our state checking code wants to see the peer outdated. */
2670 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2672 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2673 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2676 case SS_CW_FAILED_BY_PEER:
2677 spin_lock_irq(&connection->resource->req_lock);
2678 cstate = connection->cstate;
2679 spin_unlock_irq(&connection->resource->req_lock);
2680 if (cstate <= C_WF_CONNECTION)
2682 /* The peer probably wants to see us outdated. */
2683 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2684 disk, D_OUTDATED), 0);
2685 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2686 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2691 /* no special handling necessary */
2694 if (rv >= SS_SUCCESS) {
2695 enum drbd_state_rv rv2;
2696 /* No one else can reconfigure the network while I am here.
2697 * The state handling only uses drbd_thread_stop_nowait(),
2698 * we want to really wait here until the receiver is no more.
2700 drbd_thread_stop(&connection->receiver);
2702 /* Race breaker. This additional state change request may be
2703 * necessary, if this was a forced disconnect during a receiver
2704 * restart. We may have "killed" the receiver thread just
2705 * after drbd_receiver() returned. Typically, we should be
2706 * C_STANDALONE already, now, and this becomes a no-op.
2708 rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2709 CS_VERBOSE | CS_HARD);
2710 if (rv2 < SS_SUCCESS)
2711 drbd_err(connection,
2712 "unexpected rv2=%d in conn_try_disconnect()\n",
2714 /* Unlike in DRBD 9, the state engine has generated
2715 * NOTIFY_DESTROY events before clearing connection->net_conf. */
2720 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2722 struct drbd_config_context adm_ctx;
2723 struct disconnect_parms parms;
2724 struct drbd_connection *connection;
2725 enum drbd_state_rv rv;
2726 enum drbd_ret_code retcode;
2729 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2730 if (!adm_ctx.reply_skb)
2732 if (retcode != NO_ERROR)
2735 connection = adm_ctx.connection;
2736 memset(&parms, 0, sizeof(parms));
2737 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2738 err = disconnect_parms_from_attrs(&parms, info);
2740 retcode = ERR_MANDATORY_TAG;
2741 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2746 mutex_lock(&adm_ctx.resource->adm_mutex);
2747 rv = conn_try_disconnect(connection, parms.force_disconnect);
2748 mutex_unlock(&adm_ctx.resource->adm_mutex);
2749 if (rv < SS_SUCCESS) {
2750 drbd_adm_finish(&adm_ctx, info, rv);
2755 drbd_adm_finish(&adm_ctx, info, retcode);
2759 void resync_after_online_grow(struct drbd_device *device)
2761 int iass; /* I am sync source */
2763 drbd_info(device, "Resync of new storage after online grow\n");
2764 if (device->state.role != device->state.peer)
2765 iass = (device->state.role == R_PRIMARY);
2767 iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2770 drbd_start_resync(device, C_SYNC_SOURCE);
2772 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2775 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2777 struct drbd_config_context adm_ctx;
2778 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2779 struct resize_parms rs;
2780 struct drbd_device *device;
2781 enum drbd_ret_code retcode;
2782 enum determine_dev_size dd;
2783 bool change_al_layout = false;
2784 enum dds_flags ddsf;
2788 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2789 if (!adm_ctx.reply_skb)
2791 if (retcode != NO_ERROR)
2794 mutex_lock(&adm_ctx.resource->adm_mutex);
2795 device = adm_ctx.device;
2796 if (!get_ldev(device)) {
2797 retcode = ERR_NO_DISK;
2801 memset(&rs, 0, sizeof(struct resize_parms));
2802 rs.al_stripes = device->ldev->md.al_stripes;
2803 rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2804 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2805 err = resize_parms_from_attrs(&rs, info);
2807 retcode = ERR_MANDATORY_TAG;
2808 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2813 if (device->state.conn > C_CONNECTED) {
2814 retcode = ERR_RESIZE_RESYNC;
2818 if (device->state.role == R_SECONDARY &&
2819 device->state.peer == R_SECONDARY) {
2820 retcode = ERR_NO_PRIMARY;
2824 if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2825 retcode = ERR_NEED_APV_93;
2830 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2832 if (u_size != (sector_t)rs.resize_size) {
2833 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2834 if (!new_disk_conf) {
2835 retcode = ERR_NOMEM;
2840 if (device->ldev->md.al_stripes != rs.al_stripes ||
2841 device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2842 u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2844 if (al_size_k > (16 * 1024 * 1024)) {
2845 retcode = ERR_MD_LAYOUT_TOO_BIG;
2849 if (al_size_k < MD_32kB_SECT/2) {
2850 retcode = ERR_MD_LAYOUT_TOO_SMALL;
2854 if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2855 retcode = ERR_MD_LAYOUT_CONNECTED;
2859 change_al_layout = true;
2862 if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev))
2863 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
2865 if (new_disk_conf) {
2866 mutex_lock(&device->resource->conf_update);
2867 old_disk_conf = device->ldev->disk_conf;
2868 *new_disk_conf = *old_disk_conf;
2869 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2870 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2871 mutex_unlock(&device->resource->conf_update);
2872 kvfree_rcu_mightsleep(old_disk_conf);
2873 new_disk_conf = NULL;
2876 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2877 dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL);
2878 drbd_md_sync(device);
2880 if (dd == DS_ERROR) {
2881 retcode = ERR_NOMEM_BITMAP;
2883 } else if (dd == DS_ERROR_SPACE_MD) {
2884 retcode = ERR_MD_LAYOUT_NO_FIT;
2886 } else if (dd == DS_ERROR_SHRINK) {
2887 retcode = ERR_IMPLICIT_SHRINK;
2891 if (device->state.conn == C_CONNECTED) {
2893 set_bit(RESIZE_PENDING, &device->flags);
2895 drbd_send_uuids(first_peer_device(device));
2896 drbd_send_sizes(first_peer_device(device), 1, ddsf);
2900 mutex_unlock(&adm_ctx.resource->adm_mutex);
2902 drbd_adm_finish(&adm_ctx, info, retcode);
2907 kfree(new_disk_conf);
2911 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2913 struct drbd_config_context adm_ctx;
2914 enum drbd_ret_code retcode;
2915 struct res_opts res_opts;
2918 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2919 if (!adm_ctx.reply_skb)
2921 if (retcode != NO_ERROR)
2924 res_opts = adm_ctx.resource->res_opts;
2925 if (should_set_defaults(info))
2926 set_res_opts_defaults(&res_opts);
2928 err = res_opts_from_attrs(&res_opts, info);
2929 if (err && err != -ENOMSG) {
2930 retcode = ERR_MANDATORY_TAG;
2931 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2935 mutex_lock(&adm_ctx.resource->adm_mutex);
2936 err = set_resource_options(adm_ctx.resource, &res_opts);
2938 retcode = ERR_INVALID_REQUEST;
2940 retcode = ERR_NOMEM;
2942 mutex_unlock(&adm_ctx.resource->adm_mutex);
2945 drbd_adm_finish(&adm_ctx, info, retcode);
2949 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2951 struct drbd_config_context adm_ctx;
2952 struct drbd_device *device;
2953 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2955 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2956 if (!adm_ctx.reply_skb)
2958 if (retcode != NO_ERROR)
2961 device = adm_ctx.device;
2962 if (!get_ldev(device)) {
2963 retcode = ERR_NO_DISK;
2967 mutex_lock(&adm_ctx.resource->adm_mutex);
2969 /* If there is still bitmap IO pending, probably because of a previous
2970 * resync just being finished, wait for it before requesting a new resync.
2971 * Also wait for it's after_state_ch(). */
2972 drbd_suspend_io(device);
2973 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2974 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2976 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2977 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2978 * try to start a resync handshake as sync target for full sync.
2980 if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
2981 retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
2982 if (retcode >= SS_SUCCESS) {
2983 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2984 "set_n_write from invalidate", BM_LOCKED_MASK, NULL))
2985 retcode = ERR_IO_MD_DISK;
2988 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
2989 drbd_resume_io(device);
2990 mutex_unlock(&adm_ctx.resource->adm_mutex);
2993 drbd_adm_finish(&adm_ctx, info, retcode);
2997 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2998 union drbd_state mask, union drbd_state val)
3000 struct drbd_config_context adm_ctx;
3001 enum drbd_ret_code retcode;
3003 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3004 if (!adm_ctx.reply_skb)
3006 if (retcode != NO_ERROR)
3009 mutex_lock(&adm_ctx.resource->adm_mutex);
3010 retcode = drbd_request_state(adm_ctx.device, mask, val);
3011 mutex_unlock(&adm_ctx.resource->adm_mutex);
3013 drbd_adm_finish(&adm_ctx, info, retcode);
3017 static int drbd_bmio_set_susp_al(struct drbd_device *device,
3018 struct drbd_peer_device *peer_device) __must_hold(local)
3022 rv = drbd_bmio_set_n_write(device, peer_device);
3023 drbd_suspend_al(device);
3027 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
3029 struct drbd_config_context adm_ctx;
3030 int retcode; /* drbd_ret_code, drbd_state_rv */
3031 struct drbd_device *device;
3033 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3034 if (!adm_ctx.reply_skb)
3036 if (retcode != NO_ERROR)
3039 device = adm_ctx.device;
3040 if (!get_ldev(device)) {
3041 retcode = ERR_NO_DISK;
3045 mutex_lock(&adm_ctx.resource->adm_mutex);
3047 /* If there is still bitmap IO pending, probably because of a previous
3048 * resync just being finished, wait for it before requesting a new resync.
3049 * Also wait for it's after_state_ch(). */
3050 drbd_suspend_io(device);
3051 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
3052 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
3054 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
3055 * in the bitmap. Otherwise, try to start a resync handshake
3056 * as sync source for full sync.
3058 if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
3059 /* The peer will get a resync upon connect anyways. Just make that
3060 into a full resync. */
3061 retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
3062 if (retcode >= SS_SUCCESS) {
3063 if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al,
3064 "set_n_write from invalidate_peer",
3065 BM_LOCKED_SET_ALLOWED, NULL))
3066 retcode = ERR_IO_MD_DISK;
3069 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
3070 drbd_resume_io(device);
3071 mutex_unlock(&adm_ctx.resource->adm_mutex);
3074 drbd_adm_finish(&adm_ctx, info, retcode);
3078 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
3080 struct drbd_config_context adm_ctx;
3081 enum drbd_ret_code retcode;
3083 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3084 if (!adm_ctx.reply_skb)
3086 if (retcode != NO_ERROR)
3089 mutex_lock(&adm_ctx.resource->adm_mutex);
3090 if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
3091 retcode = ERR_PAUSE_IS_SET;
3092 mutex_unlock(&adm_ctx.resource->adm_mutex);
3094 drbd_adm_finish(&adm_ctx, info, retcode);
3098 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
3100 struct drbd_config_context adm_ctx;
3101 union drbd_dev_state s;
3102 enum drbd_ret_code retcode;
3104 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3105 if (!adm_ctx.reply_skb)
3107 if (retcode != NO_ERROR)
3110 mutex_lock(&adm_ctx.resource->adm_mutex);
3111 if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
3112 s = adm_ctx.device->state;
3113 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
3114 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
3115 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
3117 retcode = ERR_PAUSE_IS_CLEAR;
3120 mutex_unlock(&adm_ctx.resource->adm_mutex);
3122 drbd_adm_finish(&adm_ctx, info, retcode);
3126 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
3128 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
3131 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
3133 struct drbd_config_context adm_ctx;
3134 struct drbd_device *device;
3135 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3137 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3138 if (!adm_ctx.reply_skb)
3140 if (retcode != NO_ERROR)
3143 mutex_lock(&adm_ctx.resource->adm_mutex);
3144 device = adm_ctx.device;
3145 if (test_bit(NEW_CUR_UUID, &device->flags)) {
3146 if (get_ldev_if_state(device, D_ATTACHING)) {
3147 drbd_uuid_new_current(device);
3150 /* This is effectively a multi-stage "forced down".
3151 * The NEW_CUR_UUID bit is supposedly only set, if we
3152 * lost the replication connection, and are configured
3153 * to freeze IO and wait for some fence-peer handler.
3154 * So we still don't have a replication connection.
3155 * And now we don't have a local disk either. After
3156 * resume, we will fail all pending and new IO, because
3157 * we don't have any data anymore. Which means we will
3158 * eventually be able to terminate all users of this
3159 * device, and then take it down. By bumping the
3160 * "effective" data uuid, we make sure that you really
3161 * need to tear down before you reconfigure, we will
3162 * the refuse to re-connect or re-attach (because no
3163 * matching real data uuid exists).
3166 get_random_bytes(&val, sizeof(u64));
3167 drbd_set_ed_uuid(device, val);
3168 drbd_warn(device, "Resumed without access to data; please tear down before attempting to re-configure.\n");
3170 clear_bit(NEW_CUR_UUID, &device->flags);
3172 drbd_suspend_io(device);
3173 retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
3174 if (retcode == SS_SUCCESS) {
3175 if (device->state.conn < C_CONNECTED)
3176 tl_clear(first_peer_device(device)->connection);
3177 if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
3178 tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO);
3180 drbd_resume_io(device);
3181 mutex_unlock(&adm_ctx.resource->adm_mutex);
3183 drbd_adm_finish(&adm_ctx, info, retcode);
3187 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
3189 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
3192 static int nla_put_drbd_cfg_context(struct sk_buff *skb,
3193 struct drbd_resource *resource,
3194 struct drbd_connection *connection,
3195 struct drbd_device *device)
3198 nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_CONTEXT);
3200 goto nla_put_failure;
3202 nla_put_u32(skb, T_ctx_volume, device->vnr))
3203 goto nla_put_failure;
3204 if (nla_put_string(skb, T_ctx_resource_name, resource->name))
3205 goto nla_put_failure;
3207 if (connection->my_addr_len &&
3208 nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr))
3209 goto nla_put_failure;
3210 if (connection->peer_addr_len &&
3211 nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr))
3212 goto nla_put_failure;
3214 nla_nest_end(skb, nla);
3219 nla_nest_cancel(skb, nla);
3224 * The generic netlink dump callbacks are called outside the genl_lock(), so
3225 * they cannot use the simple attribute parsing code which uses global
3228 static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr)
3230 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3231 const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3234 nla = nla_find(nlmsg_attrdata(nlh, hdrlen), nlmsg_attrlen(nlh, hdrlen),
3235 DRBD_NLA_CFG_CONTEXT);
3238 return drbd_nla_find_nested(maxtype, nla, __nla_type(attr));
3241 static void resource_to_info(struct resource_info *, struct drbd_resource *);
3243 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb)
3245 struct drbd_genlmsghdr *dh;
3246 struct drbd_resource *resource;
3247 struct resource_info resource_info;
3248 struct resource_statistics resource_statistics;
3253 for_each_resource_rcu(resource, &drbd_resources)
3254 if (resource == (struct drbd_resource *)cb->args[0])
3255 goto found_resource;
3256 err = 0; /* resource was probably deleted */
3259 resource = list_entry(&drbd_resources,
3260 struct drbd_resource, resources);
3263 list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) {
3270 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3271 cb->nlh->nlmsg_seq, &drbd_genl_family,
3272 NLM_F_MULTI, DRBD_ADM_GET_RESOURCES);
3277 dh->ret_code = NO_ERROR;
3278 err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL);
3281 err = res_opts_to_skb(skb, &resource->res_opts, !capable(CAP_SYS_ADMIN));
3284 resource_to_info(&resource_info, resource);
3285 err = resource_info_to_skb(skb, &resource_info, !capable(CAP_SYS_ADMIN));
3288 resource_statistics.res_stat_write_ordering = resource->write_ordering;
3289 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
3292 cb->args[0] = (long)resource;
3293 genlmsg_end(skb, dh);
3303 static void device_to_statistics(struct device_statistics *s,
3304 struct drbd_device *device)
3306 memset(s, 0, sizeof(*s));
3307 s->dev_upper_blocked = !may_inc_ap_bio(device);
3308 if (get_ldev(device)) {
3309 struct drbd_md *md = &device->ldev->md;
3310 u64 *history_uuids = (u64 *)s->history_uuids;
3313 spin_lock_irq(&md->uuid_lock);
3314 s->dev_current_uuid = md->uuid[UI_CURRENT];
3315 BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1);
3316 for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++)
3317 history_uuids[n] = md->uuid[UI_HISTORY_START + n];
3318 for (; n < HISTORY_UUIDS; n++)
3319 history_uuids[n] = 0;
3320 s->history_uuids_len = HISTORY_UUIDS;
3321 spin_unlock_irq(&md->uuid_lock);
3323 s->dev_disk_flags = md->flags;
3326 s->dev_size = get_capacity(device->vdisk);
3327 s->dev_read = device->read_cnt;
3328 s->dev_write = device->writ_cnt;
3329 s->dev_al_writes = device->al_writ_cnt;
3330 s->dev_bm_writes = device->bm_writ_cnt;
3331 s->dev_upper_pending = atomic_read(&device->ap_bio_cnt);
3332 s->dev_lower_pending = atomic_read(&device->local_cnt);
3333 s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags);
3334 s->dev_exposed_data_uuid = device->ed_uuid;
3337 static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr)
3340 struct drbd_resource *resource =
3341 (struct drbd_resource *)cb->args[0];
3342 kref_put(&resource->kref, drbd_destroy_resource);
3348 int drbd_adm_dump_devices_done(struct netlink_callback *cb) {
3349 return put_resource_in_arg0(cb, 7);
3352 static void device_to_info(struct device_info *, struct drbd_device *);
3354 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb)
3356 struct nlattr *resource_filter;
3357 struct drbd_resource *resource;
3358 struct drbd_device *device;
3359 int minor, err, retcode;
3360 struct drbd_genlmsghdr *dh;
3361 struct device_info device_info;
3362 struct device_statistics device_statistics;
3363 struct idr *idr_to_search;
3365 resource = (struct drbd_resource *)cb->args[0];
3366 if (!cb->args[0] && !cb->args[1]) {
3367 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3368 if (resource_filter) {
3369 retcode = ERR_RES_NOT_KNOWN;
3370 resource = drbd_find_resource(nla_data(resource_filter));
3373 cb->args[0] = (long)resource;
3378 minor = cb->args[1];
3379 idr_to_search = resource ? &resource->devices : &drbd_devices;
3380 device = idr_get_next(idr_to_search, &minor);
3385 idr_for_each_entry_continue(idr_to_search, device, minor) {
3387 goto put_result; /* only one iteration */
3390 goto out; /* no more devices */
3393 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3394 cb->nlh->nlmsg_seq, &drbd_genl_family,
3395 NLM_F_MULTI, DRBD_ADM_GET_DEVICES);
3399 dh->ret_code = retcode;
3401 if (retcode == NO_ERROR) {
3402 dh->minor = device->minor;
3403 err = nla_put_drbd_cfg_context(skb, device->resource, NULL, device);
3406 if (get_ldev(device)) {
3407 struct disk_conf *disk_conf =
3408 rcu_dereference(device->ldev->disk_conf);
3410 err = disk_conf_to_skb(skb, disk_conf, !capable(CAP_SYS_ADMIN));
3415 device_to_info(&device_info, device);
3416 err = device_info_to_skb(skb, &device_info, !capable(CAP_SYS_ADMIN));
3420 device_to_statistics(&device_statistics, device);
3421 err = device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
3424 cb->args[1] = minor + 1;
3426 genlmsg_end(skb, dh);
3436 int drbd_adm_dump_connections_done(struct netlink_callback *cb)
3438 return put_resource_in_arg0(cb, 6);
3441 enum { SINGLE_RESOURCE, ITERATE_RESOURCES };
3443 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb)
3445 struct nlattr *resource_filter;
3446 struct drbd_resource *resource = NULL, *next_resource;
3447 struct drbd_connection *connection;
3448 int err = 0, retcode;
3449 struct drbd_genlmsghdr *dh;
3450 struct connection_info connection_info;
3451 struct connection_statistics connection_statistics;
3454 resource = (struct drbd_resource *)cb->args[0];
3456 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3457 if (resource_filter) {
3458 retcode = ERR_RES_NOT_KNOWN;
3459 resource = drbd_find_resource(nla_data(resource_filter));
3462 cb->args[0] = (long)resource;
3463 cb->args[1] = SINGLE_RESOURCE;
3467 if (list_empty(&drbd_resources))
3469 resource = list_first_entry(&drbd_resources, struct drbd_resource, resources);
3470 kref_get(&resource->kref);
3471 cb->args[0] = (long)resource;
3472 cb->args[1] = ITERATE_RESOURCES;
3477 mutex_lock(&resource->conf_update);
3480 for_each_connection_rcu(connection, resource)
3481 if (connection == (struct drbd_connection *)cb->args[2])
3482 goto found_connection;
3483 /* connection was probably deleted */
3484 goto no_more_connections;
3486 connection = list_entry(&resource->connections, struct drbd_connection, connections);
3489 list_for_each_entry_continue_rcu(connection, &resource->connections, connections) {
3490 if (!has_net_conf(connection))
3493 goto put_result; /* only one iteration */
3496 no_more_connections:
3497 if (cb->args[1] == ITERATE_RESOURCES) {
3498 for_each_resource_rcu(next_resource, &drbd_resources) {
3499 if (next_resource == resource)
3500 goto found_resource;
3502 /* resource was probably deleted */
3507 list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) {
3508 mutex_unlock(&resource->conf_update);
3509 kref_put(&resource->kref, drbd_destroy_resource);
3510 resource = next_resource;
3511 kref_get(&resource->kref);
3512 cb->args[0] = (long)resource;
3516 goto out; /* no more resources */
3519 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3520 cb->nlh->nlmsg_seq, &drbd_genl_family,
3521 NLM_F_MULTI, DRBD_ADM_GET_CONNECTIONS);
3525 dh->ret_code = retcode;
3527 if (retcode == NO_ERROR) {
3528 struct net_conf *net_conf;
3530 err = nla_put_drbd_cfg_context(skb, resource, connection, NULL);
3533 net_conf = rcu_dereference(connection->net_conf);
3535 err = net_conf_to_skb(skb, net_conf, !capable(CAP_SYS_ADMIN));
3539 connection_to_info(&connection_info, connection);
3540 err = connection_info_to_skb(skb, &connection_info, !capable(CAP_SYS_ADMIN));
3543 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
3544 err = connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
3547 cb->args[2] = (long)connection;
3549 genlmsg_end(skb, dh);
3555 mutex_unlock(&resource->conf_update);
3561 enum mdf_peer_flag {
3562 MDF_PEER_CONNECTED = 1 << 0,
3563 MDF_PEER_OUTDATED = 1 << 1,
3564 MDF_PEER_FENCING = 1 << 2,
3565 MDF_PEER_FULL_SYNC = 1 << 3,
3568 static void peer_device_to_statistics(struct peer_device_statistics *s,
3569 struct drbd_peer_device *peer_device)
3571 struct drbd_device *device = peer_device->device;
3573 memset(s, 0, sizeof(*s));
3574 s->peer_dev_received = device->recv_cnt;
3575 s->peer_dev_sent = device->send_cnt;
3576 s->peer_dev_pending = atomic_read(&device->ap_pending_cnt) +
3577 atomic_read(&device->rs_pending_cnt);
3578 s->peer_dev_unacked = atomic_read(&device->unacked_cnt);
3579 s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9);
3580 s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9);
3581 if (get_ldev(device)) {
3582 struct drbd_md *md = &device->ldev->md;
3584 spin_lock_irq(&md->uuid_lock);
3585 s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP];
3586 spin_unlock_irq(&md->uuid_lock);
3588 (drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ?
3589 MDF_PEER_CONNECTED : 0) +
3590 (drbd_md_test_flag(device->ldev, MDF_CONSISTENT) &&
3591 !drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ?
3592 MDF_PEER_OUTDATED : 0) +
3593 /* FIXME: MDF_PEER_FENCING? */
3594 (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ?
3595 MDF_PEER_FULL_SYNC : 0);
3600 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb)
3602 return put_resource_in_arg0(cb, 9);
3605 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb)
3607 struct nlattr *resource_filter;
3608 struct drbd_resource *resource;
3609 struct drbd_device *device;
3610 struct drbd_peer_device *peer_device = NULL;
3611 int minor, err, retcode;
3612 struct drbd_genlmsghdr *dh;
3613 struct idr *idr_to_search;
3615 resource = (struct drbd_resource *)cb->args[0];
3616 if (!cb->args[0] && !cb->args[1]) {
3617 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3618 if (resource_filter) {
3619 retcode = ERR_RES_NOT_KNOWN;
3620 resource = drbd_find_resource(nla_data(resource_filter));
3624 cb->args[0] = (long)resource;
3628 minor = cb->args[1];
3629 idr_to_search = resource ? &resource->devices : &drbd_devices;
3630 device = idr_find(idr_to_search, minor);
3635 device = idr_get_next(idr_to_search, &minor);
3642 for_each_peer_device(peer_device, device)
3643 if (peer_device == (struct drbd_peer_device *)cb->args[2])
3644 goto found_peer_device;
3645 /* peer device was probably deleted */
3648 /* Make peer_device point to the list head (not the first entry). */
3649 peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices);
3652 list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) {
3653 if (!has_net_conf(peer_device->connection))
3656 goto put_result; /* only one iteration */
3661 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3662 cb->nlh->nlmsg_seq, &drbd_genl_family,
3663 NLM_F_MULTI, DRBD_ADM_GET_PEER_DEVICES);
3667 dh->ret_code = retcode;
3669 if (retcode == NO_ERROR) {
3670 struct peer_device_info peer_device_info;
3671 struct peer_device_statistics peer_device_statistics;
3674 err = nla_put_drbd_cfg_context(skb, device->resource, peer_device->connection, device);
3677 peer_device_to_info(&peer_device_info, peer_device);
3678 err = peer_device_info_to_skb(skb, &peer_device_info, !capable(CAP_SYS_ADMIN));
3681 peer_device_to_statistics(&peer_device_statistics, peer_device);
3682 err = peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
3685 cb->args[1] = minor;
3686 cb->args[2] = (long)peer_device;
3688 genlmsg_end(skb, dh);
3698 * Return the connection of @resource if @resource has exactly one connection.
3700 static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
3702 struct list_head *connections = &resource->connections;
3704 if (list_empty(connections) || connections->next->next != connections)
3706 return list_first_entry(&resource->connections, struct drbd_connection, connections);
3709 static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
3710 const struct sib_info *sib)
3712 struct drbd_resource *resource = device->resource;
3713 struct state_info *si = NULL; /* for sizeof(si->member); */
3717 int exclude_sensitive;
3719 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
3720 * to. So we better exclude_sensitive information.
3722 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
3723 * in the context of the requesting user process. Exclude sensitive
3724 * information, unless current has superuser.
3726 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
3727 * relies on the current implementation of netlink_dump(), which
3728 * executes the dump callback successively from netlink_recvmsg(),
3729 * always in the context of the receiving process */
3730 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
3732 got_ldev = get_ldev(device);
3734 /* We need to add connection name and volume number information still.
3735 * Minor number is in drbd_genlmsghdr. */
3736 if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device))
3737 goto nla_put_failure;
3739 if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive))
3740 goto nla_put_failure;
3744 struct disk_conf *disk_conf;
3746 disk_conf = rcu_dereference(device->ldev->disk_conf);
3747 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
3750 struct net_conf *nc;
3752 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3754 err = net_conf_to_skb(skb, nc, exclude_sensitive);
3758 goto nla_put_failure;
3760 nla = nla_nest_start_noflag(skb, DRBD_NLA_STATE_INFO);
3762 goto nla_put_failure;
3763 if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
3764 nla_put_u32(skb, T_current_state, device->state.i) ||
3765 nla_put_u64_0pad(skb, T_ed_uuid, device->ed_uuid) ||
3766 nla_put_u64_0pad(skb, T_capacity, get_capacity(device->vdisk)) ||
3767 nla_put_u64_0pad(skb, T_send_cnt, device->send_cnt) ||
3768 nla_put_u64_0pad(skb, T_recv_cnt, device->recv_cnt) ||
3769 nla_put_u64_0pad(skb, T_read_cnt, device->read_cnt) ||
3770 nla_put_u64_0pad(skb, T_writ_cnt, device->writ_cnt) ||
3771 nla_put_u64_0pad(skb, T_al_writ_cnt, device->al_writ_cnt) ||
3772 nla_put_u64_0pad(skb, T_bm_writ_cnt, device->bm_writ_cnt) ||
3773 nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) ||
3774 nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) ||
3775 nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt)))
3776 goto nla_put_failure;
3781 spin_lock_irq(&device->ldev->md.uuid_lock);
3782 err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid);
3783 spin_unlock_irq(&device->ldev->md.uuid_lock);
3786 goto nla_put_failure;
3788 if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) ||
3789 nla_put_u64_0pad(skb, T_bits_total, drbd_bm_bits(device)) ||
3790 nla_put_u64_0pad(skb, T_bits_oos,
3791 drbd_bm_total_weight(device)))
3792 goto nla_put_failure;
3793 if (C_SYNC_SOURCE <= device->state.conn &&
3794 C_PAUSED_SYNC_T >= device->state.conn) {
3795 if (nla_put_u64_0pad(skb, T_bits_rs_total,
3796 device->rs_total) ||
3797 nla_put_u64_0pad(skb, T_bits_rs_failed,
3799 goto nla_put_failure;
3804 switch(sib->sib_reason) {
3805 case SIB_SYNC_PROGRESS:
3806 case SIB_GET_STATUS_REPLY:
3808 case SIB_STATE_CHANGE:
3809 if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
3810 nla_put_u32(skb, T_new_state, sib->ns.i))
3811 goto nla_put_failure;
3813 case SIB_HELPER_POST:
3814 if (nla_put_u32(skb, T_helper_exit_code,
3815 sib->helper_exit_code))
3816 goto nla_put_failure;
3818 case SIB_HELPER_PRE:
3819 if (nla_put_string(skb, T_helper, sib->helper_name))
3820 goto nla_put_failure;
3824 nla_nest_end(skb, nla);
3834 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3836 struct drbd_config_context adm_ctx;
3837 enum drbd_ret_code retcode;
3840 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3841 if (!adm_ctx.reply_skb)
3843 if (retcode != NO_ERROR)
3846 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL);
3848 nlmsg_free(adm_ctx.reply_skb);
3852 drbd_adm_finish(&adm_ctx, info, retcode);
3856 static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3858 struct drbd_device *device;
3859 struct drbd_genlmsghdr *dh;
3860 struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3861 struct drbd_resource *resource = NULL;
3862 struct drbd_resource *tmp;
3863 unsigned volume = cb->args[1];
3865 /* Open coded, deferred, iteration:
3866 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3867 * connection = "first connection of resource or undefined";
3868 * idr_for_each_entry(&resource->devices, device, i) {
3872 * where resource is cb->args[0];
3873 * and i is cb->args[1];
3875 * cb->args[2] indicates if we shall loop over all resources,
3876 * or just dump all volumes of a single resource.
3878 * This may miss entries inserted after this dump started,
3879 * or entries deleted before they are reached.
3881 * We need to make sure the device won't disappear while
3882 * we are looking at it, and revalidate our iterators
3883 * on each iteration.
3886 /* synchronize with conn_create()/drbd_destroy_connection() */
3888 /* revalidate iterator position */
3889 for_each_resource_rcu(tmp, &drbd_resources) {
3891 /* first iteration */
3903 device = idr_get_next(&resource->devices, &volume);
3905 /* No more volumes to dump on this resource.
3906 * Advance resource iterator. */
3907 pos = list_entry_rcu(resource->resources.next,
3908 struct drbd_resource, resources);
3909 /* Did we dump any volume of this resource yet? */
3911 /* If we reached the end of the list,
3912 * or only a single resource dump was requested,
3914 if (&pos->resources == &drbd_resources || cb->args[2])
3922 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3923 cb->nlh->nlmsg_seq, &drbd_genl_family,
3924 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
3929 /* This is a connection without a single volume.
3930 * Suprisingly enough, it may have a network
3932 struct drbd_connection *connection;
3935 dh->ret_code = NO_ERROR;
3936 connection = the_only_connection(resource);
3937 if (nla_put_drbd_cfg_context(skb, resource, connection, NULL))
3940 struct net_conf *nc;
3942 nc = rcu_dereference(connection->net_conf);
3943 if (nc && net_conf_to_skb(skb, nc, 1) != 0)
3949 D_ASSERT(device, device->vnr == volume);
3950 D_ASSERT(device, device->resource == resource);
3952 dh->minor = device_to_minor(device);
3953 dh->ret_code = NO_ERROR;
3955 if (nla_put_status_info(skb, device, NULL)) {
3957 genlmsg_cancel(skb, dh);
3961 genlmsg_end(skb, dh);
3966 /* where to start the next iteration */
3967 cb->args[0] = (long)pos;
3968 cb->args[1] = (pos == resource) ? volume + 1 : 0;
3970 /* No more resources/volumes/minors found results in an empty skb.
3971 * Which will terminate the dump. */
3976 * Request status of all resources, or of all volumes within a single resource.
3978 * This is a dump, as the answer may not fit in a single reply skb otherwise.
3979 * Which means we cannot use the family->attrbuf or other such members, because
3980 * dump is NOT protected by the genl_lock(). During dump, we only have access
3981 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3983 * Once things are setup properly, we call into get_one_status().
3985 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
3987 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3989 const char *resource_name;
3990 struct drbd_resource *resource;
3993 /* Is this a followup call? */
3995 /* ... of a single resource dump,
3996 * and the resource iterator has been advanced already? */
3997 if (cb->args[2] && cb->args[2] != cb->args[0])
3998 return 0; /* DONE. */
4002 /* First call (from netlink_dump_start). We need to figure out
4003 * which resource(s) the user wants us to dump. */
4004 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
4005 nlmsg_attrlen(cb->nlh, hdrlen),
4006 DRBD_NLA_CFG_CONTEXT);
4008 /* No explicit context given. Dump all. */
4011 maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
4012 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
4014 return PTR_ERR(nla);
4015 /* context given, but no name present? */
4018 resource_name = nla_data(nla);
4019 if (!*resource_name)
4021 resource = drbd_find_resource(resource_name);
4025 kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */
4027 /* prime iterators, and set "filter" mode mark:
4028 * only dump this connection. */
4029 cb->args[0] = (long)resource;
4030 /* cb->args[1] = 0; passed in this way. */
4031 cb->args[2] = (long)resource;
4034 return get_one_status(skb, cb);
4037 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
4039 struct drbd_config_context adm_ctx;
4040 enum drbd_ret_code retcode;
4041 struct timeout_parms tp;
4044 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4045 if (!adm_ctx.reply_skb)
4047 if (retcode != NO_ERROR)
4051 adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
4052 test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED :
4055 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
4057 nlmsg_free(adm_ctx.reply_skb);
4061 drbd_adm_finish(&adm_ctx, info, retcode);
4065 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
4067 struct drbd_config_context adm_ctx;
4068 struct drbd_device *device;
4069 enum drbd_ret_code retcode;
4070 struct start_ov_parms parms;
4072 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4073 if (!adm_ctx.reply_skb)
4075 if (retcode != NO_ERROR)
4078 device = adm_ctx.device;
4080 /* resume from last known position, if possible */
4081 parms.ov_start_sector = device->ov_start_sector;
4082 parms.ov_stop_sector = ULLONG_MAX;
4083 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
4084 int err = start_ov_parms_from_attrs(&parms, info);
4086 retcode = ERR_MANDATORY_TAG;
4087 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4091 mutex_lock(&adm_ctx.resource->adm_mutex);
4093 /* w_make_ov_request expects position to be aligned */
4094 device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
4095 device->ov_stop_sector = parms.ov_stop_sector;
4097 /* If there is still bitmap IO pending, e.g. previous resync or verify
4098 * just being finished, wait for it before requesting a new resync. */
4099 drbd_suspend_io(device);
4100 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
4101 retcode = drbd_request_state(device, NS(conn, C_VERIFY_S));
4102 drbd_resume_io(device);
4104 mutex_unlock(&adm_ctx.resource->adm_mutex);
4106 drbd_adm_finish(&adm_ctx, info, retcode);
4111 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
4113 struct drbd_config_context adm_ctx;
4114 struct drbd_device *device;
4115 enum drbd_ret_code retcode;
4116 int skip_initial_sync = 0;
4118 struct new_c_uuid_parms args;
4120 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4121 if (!adm_ctx.reply_skb)
4123 if (retcode != NO_ERROR)
4126 device = adm_ctx.device;
4127 memset(&args, 0, sizeof(args));
4128 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
4129 err = new_c_uuid_parms_from_attrs(&args, info);
4131 retcode = ERR_MANDATORY_TAG;
4132 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4137 mutex_lock(&adm_ctx.resource->adm_mutex);
4138 mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */
4140 if (!get_ldev(device)) {
4141 retcode = ERR_NO_DISK;
4145 /* this is "skip initial sync", assume to be clean */
4146 if (device->state.conn == C_CONNECTED &&
4147 first_peer_device(device)->connection->agreed_pro_version >= 90 &&
4148 device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
4149 drbd_info(device, "Preparing to skip initial sync\n");
4150 skip_initial_sync = 1;
4151 } else if (device->state.conn != C_STANDALONE) {
4152 retcode = ERR_CONNECTED;
4156 drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
4157 drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
4159 if (args.clear_bm) {
4160 err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
4161 "clear_n_write from new_c_uuid", BM_LOCKED_MASK, NULL);
4163 drbd_err(device, "Writing bitmap failed with %d\n", err);
4164 retcode = ERR_IO_MD_DISK;
4166 if (skip_initial_sync) {
4167 drbd_send_uuids_skip_initial_sync(first_peer_device(device));
4168 _drbd_uuid_set(device, UI_BITMAP, 0);
4169 drbd_print_uuids(device, "cleared bitmap UUID");
4170 spin_lock_irq(&device->resource->req_lock);
4171 _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
4173 spin_unlock_irq(&device->resource->req_lock);
4177 drbd_md_sync(device);
4181 mutex_unlock(device->state_mutex);
4182 mutex_unlock(&adm_ctx.resource->adm_mutex);
4184 drbd_adm_finish(&adm_ctx, info, retcode);
4188 static enum drbd_ret_code
4189 drbd_check_resource_name(struct drbd_config_context *adm_ctx)
4191 const char *name = adm_ctx->resource_name;
4192 if (!name || !name[0]) {
4193 drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing");
4194 return ERR_MANDATORY_TAG;
4196 /* if we want to use these in sysfs/configfs/debugfs some day,
4197 * we must not allow slashes */
4198 if (strchr(name, '/')) {
4199 drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name");
4200 return ERR_INVALID_REQUEST;
4205 static void resource_to_info(struct resource_info *info,
4206 struct drbd_resource *resource)
4208 info->res_role = conn_highest_role(first_connection(resource));
4209 info->res_susp = resource->susp;
4210 info->res_susp_nod = resource->susp_nod;
4211 info->res_susp_fen = resource->susp_fen;
4214 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
4216 struct drbd_connection *connection;
4217 struct drbd_config_context adm_ctx;
4218 enum drbd_ret_code retcode;
4219 struct res_opts res_opts;
4222 retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0);
4223 if (!adm_ctx.reply_skb)
4225 if (retcode != NO_ERROR)
4228 set_res_opts_defaults(&res_opts);
4229 err = res_opts_from_attrs(&res_opts, info);
4230 if (err && err != -ENOMSG) {
4231 retcode = ERR_MANDATORY_TAG;
4232 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4236 retcode = drbd_check_resource_name(&adm_ctx);
4237 if (retcode != NO_ERROR)
4240 if (adm_ctx.resource) {
4241 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
4242 retcode = ERR_INVALID_REQUEST;
4243 drbd_msg_put_info(adm_ctx.reply_skb, "resource exists");
4245 /* else: still NO_ERROR */
4249 /* not yet safe for genl_family.parallel_ops */
4250 mutex_lock(&resources_mutex);
4251 connection = conn_create(adm_ctx.resource_name, &res_opts);
4252 mutex_unlock(&resources_mutex);
4255 struct resource_info resource_info;
4257 mutex_lock(¬ification_mutex);
4258 resource_to_info(&resource_info, connection->resource);
4259 notify_resource_state(NULL, 0, connection->resource,
4260 &resource_info, NOTIFY_CREATE);
4261 mutex_unlock(¬ification_mutex);
4263 retcode = ERR_NOMEM;
4266 drbd_adm_finish(&adm_ctx, info, retcode);
4270 static void device_to_info(struct device_info *info,
4271 struct drbd_device *device)
4273 info->dev_disk_state = device->state.disk;
4277 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info)
4279 struct drbd_config_context adm_ctx;
4280 struct drbd_genlmsghdr *dh = genl_info_userhdr(info);
4281 enum drbd_ret_code retcode;
4283 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4284 if (!adm_ctx.reply_skb)
4286 if (retcode != NO_ERROR)
4289 if (dh->minor > MINORMASK) {
4290 drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range");
4291 retcode = ERR_INVALID_REQUEST;
4294 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
4295 drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range");
4296 retcode = ERR_INVALID_REQUEST;
4300 /* drbd_adm_prepare made sure already
4301 * that first_peer_device(device)->connection and device->vnr match the request. */
4302 if (adm_ctx.device) {
4303 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
4304 retcode = ERR_MINOR_OR_VOLUME_EXISTS;
4305 /* else: still NO_ERROR */
4309 mutex_lock(&adm_ctx.resource->adm_mutex);
4310 retcode = drbd_create_device(&adm_ctx, dh->minor);
4311 if (retcode == NO_ERROR) {
4312 struct drbd_device *device;
4313 struct drbd_peer_device *peer_device;
4314 struct device_info info;
4315 unsigned int peer_devices = 0;
4316 enum drbd_notification_type flags;
4318 device = minor_to_device(dh->minor);
4319 for_each_peer_device(peer_device, device) {
4320 if (!has_net_conf(peer_device->connection))
4325 device_to_info(&info, device);
4326 mutex_lock(¬ification_mutex);
4327 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4328 notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags);
4329 for_each_peer_device(peer_device, device) {
4330 struct peer_device_info peer_device_info;
4332 if (!has_net_conf(peer_device->connection))
4334 peer_device_to_info(&peer_device_info, peer_device);
4335 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4336 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info,
4337 NOTIFY_CREATE | flags);
4339 mutex_unlock(¬ification_mutex);
4341 mutex_unlock(&adm_ctx.resource->adm_mutex);
4343 drbd_adm_finish(&adm_ctx, info, retcode);
4347 static enum drbd_ret_code adm_del_minor(struct drbd_device *device)
4349 struct drbd_peer_device *peer_device;
4351 if (device->state.disk == D_DISKLESS &&
4352 /* no need to be device->state.conn == C_STANDALONE &&
4353 * we may want to delete a minor from a live replication group.
4355 device->state.role == R_SECONDARY) {
4356 struct drbd_connection *connection =
4357 first_connection(device->resource);
4359 _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS),
4360 CS_VERBOSE + CS_WAIT_COMPLETE);
4362 /* If the state engine hasn't stopped the sender thread yet, we
4363 * need to flush the sender work queue before generating the
4364 * DESTROY events here. */
4365 if (get_t_state(&connection->worker) == RUNNING)
4366 drbd_flush_workqueue(&connection->sender_work);
4368 mutex_lock(¬ification_mutex);
4369 for_each_peer_device(peer_device, device) {
4370 if (!has_net_conf(peer_device->connection))
4372 notify_peer_device_state(NULL, 0, peer_device, NULL,
4373 NOTIFY_DESTROY | NOTIFY_CONTINUES);
4375 notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY);
4376 mutex_unlock(¬ification_mutex);
4378 drbd_delete_device(device);
4381 return ERR_MINOR_CONFIGURED;
4384 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info)
4386 struct drbd_config_context adm_ctx;
4387 enum drbd_ret_code retcode;
4389 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4390 if (!adm_ctx.reply_skb)
4392 if (retcode != NO_ERROR)
4395 mutex_lock(&adm_ctx.resource->adm_mutex);
4396 retcode = adm_del_minor(adm_ctx.device);
4397 mutex_unlock(&adm_ctx.resource->adm_mutex);
4399 drbd_adm_finish(&adm_ctx, info, retcode);
4403 static int adm_del_resource(struct drbd_resource *resource)
4405 struct drbd_connection *connection;
4407 for_each_connection(connection, resource) {
4408 if (connection->cstate > C_STANDALONE)
4409 return ERR_NET_CONFIGURED;
4411 if (!idr_is_empty(&resource->devices))
4412 return ERR_RES_IN_USE;
4414 /* The state engine has stopped the sender thread, so we don't
4415 * need to flush the sender work queue before generating the
4416 * DESTROY event here. */
4417 mutex_lock(¬ification_mutex);
4418 notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY);
4419 mutex_unlock(¬ification_mutex);
4421 mutex_lock(&resources_mutex);
4422 list_del_rcu(&resource->resources);
4423 mutex_unlock(&resources_mutex);
4424 /* Make sure all threads have actually stopped: state handling only
4425 * does drbd_thread_stop_nowait(). */
4426 list_for_each_entry(connection, &resource->connections, connections)
4427 drbd_thread_stop(&connection->worker);
4429 drbd_free_resource(resource);
4433 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
4435 struct drbd_config_context adm_ctx;
4436 struct drbd_resource *resource;
4437 struct drbd_connection *connection;
4438 struct drbd_device *device;
4439 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
4442 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4443 if (!adm_ctx.reply_skb)
4445 if (retcode != NO_ERROR)
4448 resource = adm_ctx.resource;
4449 mutex_lock(&resource->adm_mutex);
4451 for_each_connection(connection, resource) {
4452 struct drbd_peer_device *peer_device;
4454 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
4455 retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0);
4456 if (retcode < SS_SUCCESS) {
4457 drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote");
4462 retcode = conn_try_disconnect(connection, 0);
4463 if (retcode < SS_SUCCESS) {
4464 drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect");
4470 idr_for_each_entry(&resource->devices, device, i) {
4471 retcode = adm_detach(device, 0);
4472 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
4473 drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach");
4478 /* delete volumes */
4479 idr_for_each_entry(&resource->devices, device, i) {
4480 retcode = adm_del_minor(device);
4481 if (retcode != NO_ERROR) {
4482 /* "can not happen" */
4483 drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume");
4488 retcode = adm_del_resource(resource);
4490 mutex_unlock(&resource->adm_mutex);
4492 drbd_adm_finish(&adm_ctx, info, retcode);
4496 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
4498 struct drbd_config_context adm_ctx;
4499 struct drbd_resource *resource;
4500 enum drbd_ret_code retcode;
4502 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4503 if (!adm_ctx.reply_skb)
4505 if (retcode != NO_ERROR)
4507 resource = adm_ctx.resource;
4509 mutex_lock(&resource->adm_mutex);
4510 retcode = adm_del_resource(resource);
4511 mutex_unlock(&resource->adm_mutex);
4513 drbd_adm_finish(&adm_ctx, info, retcode);
4517 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib)
4519 struct sk_buff *msg;
4520 struct drbd_genlmsghdr *d_out;
4524 seq = atomic_inc_return(&drbd_genl_seq);
4525 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4530 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
4531 if (!d_out) /* cannot happen, but anyways. */
4532 goto nla_put_failure;
4533 d_out->minor = device_to_minor(device);
4534 d_out->ret_code = NO_ERROR;
4536 if (nla_put_status_info(msg, device, sib))
4537 goto nla_put_failure;
4538 genlmsg_end(msg, d_out);
4539 err = drbd_genl_multicast_events(msg, GFP_NOWAIT);
4540 /* msg has been consumed or freed in netlink_broadcast() */
4541 if (err && err != -ESRCH)
4549 drbd_err(device, "Error %d while broadcasting event. "
4550 "Event seq:%u sib_reason:%u\n",
4551 err, seq, sib->sib_reason);
4554 static int nla_put_notification_header(struct sk_buff *msg,
4555 enum drbd_notification_type type)
4557 struct drbd_notification_header nh = {
4561 return drbd_notification_header_to_skb(msg, &nh, true);
4564 int notify_resource_state(struct sk_buff *skb,
4566 struct drbd_resource *resource,
4567 struct resource_info *resource_info,
4568 enum drbd_notification_type type)
4570 struct resource_statistics resource_statistics;
4571 struct drbd_genlmsghdr *dh;
4572 bool multicast = false;
4576 seq = atomic_inc_return(¬ify_genl_seq);
4577 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4585 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_RESOURCE_STATE);
4587 goto nla_put_failure;
4589 dh->ret_code = NO_ERROR;
4590 if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) ||
4591 nla_put_notification_header(skb, type) ||
4592 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4593 resource_info_to_skb(skb, resource_info, true)))
4594 goto nla_put_failure;
4595 resource_statistics.res_stat_write_ordering = resource->write_ordering;
4596 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
4598 goto nla_put_failure;
4599 genlmsg_end(skb, dh);
4601 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4602 /* skb has been consumed or freed in netlink_broadcast() */
4603 if (err && err != -ESRCH)
4611 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4616 int notify_device_state(struct sk_buff *skb,
4618 struct drbd_device *device,
4619 struct device_info *device_info,
4620 enum drbd_notification_type type)
4622 struct device_statistics device_statistics;
4623 struct drbd_genlmsghdr *dh;
4624 bool multicast = false;
4628 seq = atomic_inc_return(¬ify_genl_seq);
4629 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4637 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_DEVICE_STATE);
4639 goto nla_put_failure;
4640 dh->minor = device->minor;
4641 dh->ret_code = NO_ERROR;
4642 if (nla_put_drbd_cfg_context(skb, device->resource, NULL, device) ||
4643 nla_put_notification_header(skb, type) ||
4644 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4645 device_info_to_skb(skb, device_info, true)))
4646 goto nla_put_failure;
4647 device_to_statistics(&device_statistics, device);
4648 device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
4649 genlmsg_end(skb, dh);
4651 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4652 /* skb has been consumed or freed in netlink_broadcast() */
4653 if (err && err != -ESRCH)
4661 drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n",
4666 int notify_connection_state(struct sk_buff *skb,
4668 struct drbd_connection *connection,
4669 struct connection_info *connection_info,
4670 enum drbd_notification_type type)
4672 struct connection_statistics connection_statistics;
4673 struct drbd_genlmsghdr *dh;
4674 bool multicast = false;
4678 seq = atomic_inc_return(¬ify_genl_seq);
4679 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4687 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_CONNECTION_STATE);
4689 goto nla_put_failure;
4691 dh->ret_code = NO_ERROR;
4692 if (nla_put_drbd_cfg_context(skb, connection->resource, connection, NULL) ||
4693 nla_put_notification_header(skb, type) ||
4694 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4695 connection_info_to_skb(skb, connection_info, true)))
4696 goto nla_put_failure;
4697 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
4698 connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
4699 genlmsg_end(skb, dh);
4701 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4702 /* skb has been consumed or freed in netlink_broadcast() */
4703 if (err && err != -ESRCH)
4711 drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n",
4716 int notify_peer_device_state(struct sk_buff *skb,
4718 struct drbd_peer_device *peer_device,
4719 struct peer_device_info *peer_device_info,
4720 enum drbd_notification_type type)
4722 struct peer_device_statistics peer_device_statistics;
4723 struct drbd_resource *resource = peer_device->device->resource;
4724 struct drbd_genlmsghdr *dh;
4725 bool multicast = false;
4729 seq = atomic_inc_return(¬ify_genl_seq);
4730 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4738 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_PEER_DEVICE_STATE);
4740 goto nla_put_failure;
4742 dh->ret_code = NO_ERROR;
4743 if (nla_put_drbd_cfg_context(skb, resource, peer_device->connection, peer_device->device) ||
4744 nla_put_notification_header(skb, type) ||
4745 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4746 peer_device_info_to_skb(skb, peer_device_info, true)))
4747 goto nla_put_failure;
4748 peer_device_to_statistics(&peer_device_statistics, peer_device);
4749 peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
4750 genlmsg_end(skb, dh);
4752 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4753 /* skb has been consumed or freed in netlink_broadcast() */
4754 if (err && err != -ESRCH)
4762 drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n",
4767 void notify_helper(enum drbd_notification_type type,
4768 struct drbd_device *device, struct drbd_connection *connection,
4769 const char *name, int status)
4771 struct drbd_resource *resource = device ? device->resource : connection->resource;
4772 struct drbd_helper_info helper_info;
4773 unsigned int seq = atomic_inc_return(¬ify_genl_seq);
4774 struct sk_buff *skb = NULL;
4775 struct drbd_genlmsghdr *dh;
4778 strscpy(helper_info.helper_name, name, sizeof(helper_info.helper_name));
4779 helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name));
4780 helper_info.helper_status = status;
4782 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4788 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_HELPER);
4791 dh->minor = device ? device->minor : -1;
4792 dh->ret_code = NO_ERROR;
4793 mutex_lock(¬ification_mutex);
4794 if (nla_put_drbd_cfg_context(skb, resource, connection, device) ||
4795 nla_put_notification_header(skb, type) ||
4796 drbd_helper_info_to_skb(skb, &helper_info, true))
4798 genlmsg_end(skb, dh);
4799 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4801 /* skb has been consumed or freed in netlink_broadcast() */
4802 if (err && err != -ESRCH)
4804 mutex_unlock(¬ification_mutex);
4808 mutex_unlock(¬ification_mutex);
4811 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4815 static int notify_initial_state_done(struct sk_buff *skb, unsigned int seq)
4817 struct drbd_genlmsghdr *dh;
4821 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_INITIAL_STATE_DONE);
4823 goto nla_put_failure;
4825 dh->ret_code = NO_ERROR;
4826 if (nla_put_notification_header(skb, NOTIFY_EXISTS))
4827 goto nla_put_failure;
4828 genlmsg_end(skb, dh);
4833 pr_err("Error %d sending event. Event seq:%u\n", err, seq);
4837 static void free_state_changes(struct list_head *list)
4839 while (!list_empty(list)) {
4840 struct drbd_state_change *state_change =
4841 list_first_entry(list, struct drbd_state_change, list);
4842 list_del(&state_change->list);
4843 forget_state_change(state_change);
4847 static unsigned int notifications_for_state_change(struct drbd_state_change *state_change)
4850 state_change->n_connections +
4851 state_change->n_devices +
4852 state_change->n_devices * state_change->n_connections;
4855 static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4857 struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0];
4858 unsigned int seq = cb->args[2];
4860 enum drbd_notification_type flags = 0;
4863 /* There is no need for taking notification_mutex here: it doesn't
4864 matter if the initial state events mix with later state chage
4865 events; we can always tell the events apart by the NOTIFY_EXISTS
4869 if (cb->args[5] == 1) {
4870 err = notify_initial_state_done(skb, seq);
4874 if (cb->args[4] < cb->args[3])
4875 flags |= NOTIFY_CONTINUES;
4877 err = notify_resource_state_change(skb, seq, state_change->resource,
4878 NOTIFY_EXISTS | flags);
4882 if (n < state_change->n_connections) {
4883 err = notify_connection_state_change(skb, seq, &state_change->connections[n],
4884 NOTIFY_EXISTS | flags);
4887 n -= state_change->n_connections;
4888 if (n < state_change->n_devices) {
4889 err = notify_device_state_change(skb, seq, &state_change->devices[n],
4890 NOTIFY_EXISTS | flags);
4893 n -= state_change->n_devices;
4894 if (n < state_change->n_devices * state_change->n_connections) {
4895 err = notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n],
4896 NOTIFY_EXISTS | flags);
4901 if (cb->args[4] == cb->args[3]) {
4902 struct drbd_state_change *next_state_change =
4903 list_entry(state_change->list.next,
4904 struct drbd_state_change, list);
4905 cb->args[0] = (long)next_state_change;
4906 cb->args[3] = notifications_for_state_change(next_state_change);
4916 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4918 struct drbd_resource *resource;
4921 if (cb->args[5] >= 1) {
4922 if (cb->args[5] > 1)
4923 return get_initial_state(skb, cb);
4925 struct drbd_state_change *state_change =
4926 (struct drbd_state_change *)cb->args[0];
4928 /* connect list to head */
4929 list_add(&head, &state_change->list);
4930 free_state_changes(&head);
4935 cb->args[5] = 2; /* number of iterations */
4936 mutex_lock(&resources_mutex);
4937 for_each_resource(resource, &drbd_resources) {
4938 struct drbd_state_change *state_change;
4940 state_change = remember_old_state(resource, GFP_KERNEL);
4941 if (!state_change) {
4942 if (!list_empty(&head))
4943 free_state_changes(&head);
4944 mutex_unlock(&resources_mutex);
4947 copy_old_to_new_state_change(state_change);
4948 list_add_tail(&state_change->list, &head);
4949 cb->args[5] += notifications_for_state_change(state_change);
4951 mutex_unlock(&resources_mutex);
4953 if (!list_empty(&head)) {
4954 struct drbd_state_change *state_change =
4955 list_entry(head.next, struct drbd_state_change, list);
4956 cb->args[0] = (long)state_change;
4957 cb->args[3] = notifications_for_state_change(state_change);
4958 list_del(&head); /* detach list from head */
4961 cb->args[2] = cb->nlh->nlmsg_seq;
4962 return get_initial_state(skb, cb);