1 // SPDX-License-Identifier: GPL-2.0-or-later
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 = info->userhdr;
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);
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;
1214 if (bdev && !bdev_max_discard_sectors(bdev->backing_bdev))
1217 if (connection->cstate >= C_CONNECTED &&
1218 !(connection->agreed_features & DRBD_FF_TRIM)) {
1219 drbd_info(connection,
1220 "peer DRBD too old, does not support TRIM: disabling discards\n");
1225 * We don't care for the granularity, really.
1227 * Stacking limits below should fix it for the local device. Whether or
1228 * not it is a suitable granularity on the remote device is not our
1229 * problem, really. If you care, you need to use devices with similar
1230 * topology on all peers.
1232 blk_queue_discard_granularity(q, 512);
1233 q->limits.max_discard_sectors = drbd_max_discard_sectors(connection);
1234 q->limits.max_write_zeroes_sectors =
1235 drbd_max_discard_sectors(connection);
1239 blk_queue_discard_granularity(q, 0);
1240 q->limits.max_discard_sectors = 0;
1241 q->limits.max_write_zeroes_sectors = 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 drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev,
1260 unsigned int max_bio_size, struct o_qlim *o)
1262 struct request_queue * const q = device->rq_queue;
1263 unsigned int max_hw_sectors = max_bio_size >> 9;
1264 unsigned int max_segments = 0;
1265 struct request_queue *b = NULL;
1266 struct disk_conf *dc;
1269 b = bdev->backing_bdev->bd_disk->queue;
1271 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1273 dc = rcu_dereference(device->ldev->disk_conf);
1274 max_segments = dc->max_bio_bvecs;
1277 blk_set_stacking_limits(&q->limits);
1280 blk_queue_max_hw_sectors(q, max_hw_sectors);
1281 /* This is the workaround for "bio would need to, but cannot, be split" */
1282 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1283 blk_queue_segment_boundary(q, PAGE_SIZE-1);
1284 decide_on_discard_support(device, bdev);
1287 blk_stack_limits(&q->limits, &b->limits, 0);
1288 disk_update_readahead(device->vdisk);
1290 fixup_write_zeroes(device, q);
1293 void drbd_reconsider_queue_parameters(struct drbd_device *device, struct drbd_backing_dev *bdev, struct o_qlim *o)
1295 unsigned int now, new, local, peer;
1297 now = queue_max_hw_sectors(device->rq_queue) << 9;
1298 local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */
1299 peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */
1302 local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9;
1303 device->local_max_bio_size = local;
1305 local = min(local, DRBD_MAX_BIO_SIZE);
1307 /* We may ignore peer limits if the peer is modern enough.
1308 Because new from 8.3.8 onwards the peer can use multiple
1309 BIOs for a single peer_request */
1310 if (device->state.conn >= C_WF_REPORT_PARAMS) {
1311 if (first_peer_device(device)->connection->agreed_pro_version < 94)
1312 peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1313 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1314 else if (first_peer_device(device)->connection->agreed_pro_version == 94)
1315 peer = DRBD_MAX_SIZE_H80_PACKET;
1316 else if (first_peer_device(device)->connection->agreed_pro_version < 100)
1317 peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */
1319 peer = DRBD_MAX_BIO_SIZE;
1321 /* We may later detach and re-attach on a disconnected Primary.
1322 * Avoid this setting to jump back in that case.
1323 * We want to store what we know the peer DRBD can handle,
1324 * not what the peer IO backend can handle. */
1325 if (peer > device->peer_max_bio_size)
1326 device->peer_max_bio_size = peer;
1328 new = min(local, peer);
1330 if (device->state.role == R_PRIMARY && new < now)
1331 drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1334 drbd_info(device, "max BIO size = %u\n", new);
1336 drbd_setup_queue_param(device, bdev, new, o);
1339 /* Starts the worker thread */
1340 static void conn_reconfig_start(struct drbd_connection *connection)
1342 drbd_thread_start(&connection->worker);
1343 drbd_flush_workqueue(&connection->sender_work);
1346 /* if still unconfigured, stops worker again. */
1347 static void conn_reconfig_done(struct drbd_connection *connection)
1350 spin_lock_irq(&connection->resource->req_lock);
1351 stop_threads = conn_all_vols_unconf(connection) &&
1352 connection->cstate == C_STANDALONE;
1353 spin_unlock_irq(&connection->resource->req_lock);
1355 /* ack_receiver thread and ack_sender workqueue are implicitly
1356 * stopped by receiver in conn_disconnect() */
1357 drbd_thread_stop(&connection->receiver);
1358 drbd_thread_stop(&connection->worker);
1362 /* Make sure IO is suspended before calling this function(). */
1363 static void drbd_suspend_al(struct drbd_device *device)
1367 if (!lc_try_lock(device->act_log)) {
1368 drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1372 drbd_al_shrink(device);
1373 spin_lock_irq(&device->resource->req_lock);
1374 if (device->state.conn < C_CONNECTED)
1375 s = !test_and_set_bit(AL_SUSPENDED, &device->flags);
1376 spin_unlock_irq(&device->resource->req_lock);
1377 lc_unlock(device->act_log);
1380 drbd_info(device, "Suspended AL updates\n");
1384 static bool should_set_defaults(struct genl_info *info)
1386 unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1387 return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1390 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1392 /* This is limited by 16 bit "slot" numbers,
1393 * and by available on-disk context storage.
1395 * Also (u16)~0 is special (denotes a "free" extent).
1397 * One transaction occupies one 4kB on-disk block,
1398 * we have n such blocks in the on disk ring buffer,
1399 * the "current" transaction may fail (n-1),
1400 * and there is 919 slot numbers context information per transaction.
1402 * 72 transaction blocks amounts to more than 2**16 context slots,
1403 * so cap there first.
1405 const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1406 const unsigned int sufficient_on_disk =
1407 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1408 /AL_CONTEXT_PER_TRANSACTION;
1410 unsigned int al_size_4k = bdev->md.al_size_4k;
1412 if (al_size_4k > sufficient_on_disk)
1415 return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1418 static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1420 return a->disk_barrier != b->disk_barrier ||
1421 a->disk_flushes != b->disk_flushes ||
1422 a->disk_drain != b->disk_drain;
1425 static void sanitize_disk_conf(struct drbd_device *device, struct disk_conf *disk_conf,
1426 struct drbd_backing_dev *nbc)
1428 struct block_device *bdev = nbc->backing_bdev;
1430 if (disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1431 disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1432 if (disk_conf->al_extents > drbd_al_extents_max(nbc))
1433 disk_conf->al_extents = drbd_al_extents_max(nbc);
1435 if (!bdev_max_discard_sectors(bdev)) {
1436 if (disk_conf->rs_discard_granularity) {
1437 disk_conf->rs_discard_granularity = 0; /* disable feature */
1438 drbd_info(device, "rs_discard_granularity feature disabled\n");
1442 if (disk_conf->rs_discard_granularity) {
1443 int orig_value = disk_conf->rs_discard_granularity;
1444 sector_t discard_size = bdev_max_discard_sectors(bdev) << 9;
1445 unsigned int discard_granularity = bdev_discard_granularity(bdev);
1448 if (discard_granularity > disk_conf->rs_discard_granularity)
1449 disk_conf->rs_discard_granularity = discard_granularity;
1451 remainder = disk_conf->rs_discard_granularity %
1452 discard_granularity;
1453 disk_conf->rs_discard_granularity += remainder;
1455 if (disk_conf->rs_discard_granularity > discard_size)
1456 disk_conf->rs_discard_granularity = discard_size;
1458 if (disk_conf->rs_discard_granularity != orig_value)
1459 drbd_info(device, "rs_discard_granularity changed to %d\n",
1460 disk_conf->rs_discard_granularity);
1464 static int disk_opts_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1468 if (device->act_log &&
1469 device->act_log->nr_elements == dc->al_extents)
1472 drbd_suspend_io(device);
1473 /* If IO completion is currently blocked, we would likely wait
1474 * "forever" for the activity log to become unused. So we don't. */
1475 if (atomic_read(&device->ap_bio_cnt))
1478 wait_event(device->al_wait, lc_try_lock(device->act_log));
1479 drbd_al_shrink(device);
1480 err = drbd_check_al_size(device, dc);
1481 lc_unlock(device->act_log);
1482 wake_up(&device->al_wait);
1484 drbd_resume_io(device);
1488 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1490 struct drbd_config_context adm_ctx;
1491 enum drbd_ret_code retcode;
1492 struct drbd_device *device;
1493 struct disk_conf *new_disk_conf, *old_disk_conf;
1494 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1496 unsigned int fifo_size;
1498 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1499 if (!adm_ctx.reply_skb)
1501 if (retcode != NO_ERROR)
1504 device = adm_ctx.device;
1505 mutex_lock(&adm_ctx.resource->adm_mutex);
1507 /* we also need a disk
1508 * to change the options on */
1509 if (!get_ldev(device)) {
1510 retcode = ERR_NO_DISK;
1514 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1515 if (!new_disk_conf) {
1516 retcode = ERR_NOMEM;
1520 mutex_lock(&device->resource->conf_update);
1521 old_disk_conf = device->ldev->disk_conf;
1522 *new_disk_conf = *old_disk_conf;
1523 if (should_set_defaults(info))
1524 set_disk_conf_defaults(new_disk_conf);
1526 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1527 if (err && err != -ENOMSG) {
1528 retcode = ERR_MANDATORY_TAG;
1529 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1533 if (!expect(new_disk_conf->resync_rate >= 1))
1534 new_disk_conf->resync_rate = 1;
1536 sanitize_disk_conf(device, new_disk_conf, device->ldev);
1538 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1539 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1541 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1542 if (fifo_size != device->rs_plan_s->size) {
1543 new_plan = fifo_alloc(fifo_size);
1545 drbd_err(device, "kmalloc of fifo_buffer failed");
1546 retcode = ERR_NOMEM;
1551 err = disk_opts_check_al_size(device, new_disk_conf);
1553 /* Could be just "busy". Ignore?
1554 * Introduce dedicated error code? */
1555 drbd_msg_put_info(adm_ctx.reply_skb,
1556 "Try again without changing current al-extents setting");
1557 retcode = ERR_NOMEM;
1561 lock_all_resources();
1562 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1563 if (retcode == NO_ERROR) {
1564 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1565 drbd_resync_after_changed(device);
1567 unlock_all_resources();
1569 if (retcode != NO_ERROR)
1573 old_plan = device->rs_plan_s;
1574 rcu_assign_pointer(device->rs_plan_s, new_plan);
1577 mutex_unlock(&device->resource->conf_update);
1579 if (new_disk_conf->al_updates)
1580 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1582 device->ldev->md.flags |= MDF_AL_DISABLED;
1584 if (new_disk_conf->md_flushes)
1585 clear_bit(MD_NO_FUA, &device->flags);
1587 set_bit(MD_NO_FUA, &device->flags);
1589 if (write_ordering_changed(old_disk_conf, new_disk_conf))
1590 drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH);
1592 if (old_disk_conf->discard_zeroes_if_aligned !=
1593 new_disk_conf->discard_zeroes_if_aligned)
1594 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
1596 drbd_md_sync(device);
1598 if (device->state.conn >= C_CONNECTED) {
1599 struct drbd_peer_device *peer_device;
1601 for_each_peer_device(peer_device, device)
1602 drbd_send_sync_param(peer_device);
1605 kvfree_rcu(old_disk_conf);
1607 mod_timer(&device->request_timer, jiffies + HZ);
1611 mutex_unlock(&device->resource->conf_update);
1613 kfree(new_disk_conf);
1618 mutex_unlock(&adm_ctx.resource->adm_mutex);
1620 drbd_adm_finish(&adm_ctx, info, retcode);
1624 static struct block_device *open_backing_dev(struct drbd_device *device,
1625 const char *bdev_path, void *claim_ptr, bool do_bd_link)
1627 struct block_device *bdev;
1630 bdev = blkdev_get_by_path(bdev_path,
1631 FMODE_READ | FMODE_WRITE | FMODE_EXCL, claim_ptr);
1633 drbd_err(device, "open(\"%s\") failed with %ld\n",
1634 bdev_path, PTR_ERR(bdev));
1641 err = bd_link_disk_holder(bdev, device->vdisk);
1643 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1644 drbd_err(device, "bd_link_disk_holder(\"%s\", ...) failed with %d\n",
1646 bdev = ERR_PTR(err);
1651 static int open_backing_devices(struct drbd_device *device,
1652 struct disk_conf *new_disk_conf,
1653 struct drbd_backing_dev *nbc)
1655 struct block_device *bdev;
1657 bdev = open_backing_dev(device, new_disk_conf->backing_dev, device, true);
1659 return ERR_OPEN_DISK;
1660 nbc->backing_bdev = bdev;
1663 * meta_dev_idx >= 0: external fixed size, possibly multiple
1664 * drbd sharing one meta device. TODO in that case, paranoia
1665 * check that [md_bdev, meta_dev_idx] is not yet used by some
1666 * other drbd minor! (if you use drbd.conf + drbdadm, that
1667 * should check it for you already; but if you don't, or
1668 * someone fooled it, we need to double check here)
1670 bdev = open_backing_dev(device, new_disk_conf->meta_dev,
1671 /* claim ptr: device, if claimed exclusively; shared drbd_m_holder,
1672 * if potentially shared with other drbd minors */
1673 (new_disk_conf->meta_dev_idx < 0) ? (void*)device : (void*)drbd_m_holder,
1674 /* avoid double bd_claim_by_disk() for the same (source,target) tuple,
1675 * as would happen with internal metadata. */
1676 (new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_FLEX_INT &&
1677 new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_INTERNAL));
1679 return ERR_OPEN_MD_DISK;
1680 nbc->md_bdev = bdev;
1684 static void close_backing_dev(struct drbd_device *device, struct block_device *bdev,
1690 bd_unlink_disk_holder(bdev, device->vdisk);
1691 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1694 void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev)
1699 close_backing_dev(device, ldev->md_bdev, ldev->md_bdev != ldev->backing_bdev);
1700 close_backing_dev(device, ldev->backing_bdev, true);
1702 kfree(ldev->disk_conf);
1706 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1708 struct drbd_config_context adm_ctx;
1709 struct drbd_device *device;
1710 struct drbd_peer_device *peer_device;
1711 struct drbd_connection *connection;
1713 enum drbd_ret_code retcode;
1714 enum determine_dev_size dd;
1715 sector_t max_possible_sectors;
1716 sector_t min_md_device_sectors;
1717 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1718 struct disk_conf *new_disk_conf = NULL;
1719 struct lru_cache *resync_lru = NULL;
1720 struct fifo_buffer *new_plan = NULL;
1721 union drbd_state ns, os;
1722 enum drbd_state_rv rv;
1723 struct net_conf *nc;
1725 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1726 if (!adm_ctx.reply_skb)
1728 if (retcode != NO_ERROR)
1731 device = adm_ctx.device;
1732 mutex_lock(&adm_ctx.resource->adm_mutex);
1733 peer_device = first_peer_device(device);
1734 connection = peer_device->connection;
1735 conn_reconfig_start(connection);
1737 /* if you want to reconfigure, please tear down first */
1738 if (device->state.disk > D_DISKLESS) {
1739 retcode = ERR_DISK_CONFIGURED;
1742 /* It may just now have detached because of IO error. Make sure
1743 * drbd_ldev_destroy is done already, we may end up here very fast,
1744 * e.g. if someone calls attach from the on-io-error handler,
1745 * to realize a "hot spare" feature (not that I'd recommend that) */
1746 wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1748 /* make sure there is no leftover from previous force-detach attempts */
1749 clear_bit(FORCE_DETACH, &device->flags);
1750 clear_bit(WAS_IO_ERROR, &device->flags);
1751 clear_bit(WAS_READ_ERROR, &device->flags);
1753 /* and no leftover from previously aborted resync or verify, either */
1754 device->rs_total = 0;
1755 device->rs_failed = 0;
1756 atomic_set(&device->rs_pending_cnt, 0);
1758 /* allocation not in the IO path, drbdsetup context */
1759 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1761 retcode = ERR_NOMEM;
1764 spin_lock_init(&nbc->md.uuid_lock);
1766 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1767 if (!new_disk_conf) {
1768 retcode = ERR_NOMEM;
1771 nbc->disk_conf = new_disk_conf;
1773 set_disk_conf_defaults(new_disk_conf);
1774 err = disk_conf_from_attrs(new_disk_conf, info);
1776 retcode = ERR_MANDATORY_TAG;
1777 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1781 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1782 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1784 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1786 retcode = ERR_NOMEM;
1790 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1791 retcode = ERR_MD_IDX_INVALID;
1796 nc = rcu_dereference(connection->net_conf);
1798 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1800 retcode = ERR_STONITH_AND_PROT_A;
1806 retcode = open_backing_devices(device, new_disk_conf, nbc);
1807 if (retcode != NO_ERROR)
1810 if ((nbc->backing_bdev == nbc->md_bdev) !=
1811 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1812 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1813 retcode = ERR_MD_IDX_INVALID;
1817 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1818 1, 61, sizeof(struct bm_extent),
1819 offsetof(struct bm_extent, lce));
1821 retcode = ERR_NOMEM;
1825 /* Read our meta data super block early.
1826 * This also sets other on-disk offsets. */
1827 retcode = drbd_md_read(device, nbc);
1828 if (retcode != NO_ERROR)
1831 sanitize_disk_conf(device, new_disk_conf, nbc);
1833 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1834 drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1835 (unsigned long long) drbd_get_max_capacity(nbc),
1836 (unsigned long long) new_disk_conf->disk_size);
1837 retcode = ERR_DISK_TOO_SMALL;
1841 if (new_disk_conf->meta_dev_idx < 0) {
1842 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1843 /* at least one MB, otherwise it does not make sense */
1844 min_md_device_sectors = (2<<10);
1846 max_possible_sectors = DRBD_MAX_SECTORS;
1847 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1850 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1851 retcode = ERR_MD_DISK_TOO_SMALL;
1852 drbd_warn(device, "refusing attach: md-device too small, "
1853 "at least %llu sectors needed for this meta-disk type\n",
1854 (unsigned long long) min_md_device_sectors);
1858 /* Make sure the new disk is big enough
1859 * (we may currently be R_PRIMARY with no local disk...) */
1860 if (drbd_get_max_capacity(nbc) < get_capacity(device->vdisk)) {
1861 retcode = ERR_DISK_TOO_SMALL;
1865 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1867 if (nbc->known_size > max_possible_sectors) {
1868 drbd_warn(device, "==> truncating very big lower level device "
1869 "to currently maximum possible %llu sectors <==\n",
1870 (unsigned long long) max_possible_sectors);
1871 if (new_disk_conf->meta_dev_idx >= 0)
1872 drbd_warn(device, "==>> using internal or flexible "
1873 "meta data may help <<==\n");
1876 drbd_suspend_io(device);
1877 /* also wait for the last barrier ack. */
1878 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1879 * We need a way to either ignore barrier acks for barriers sent before a device
1880 * was attached, or a way to wait for all pending barrier acks to come in.
1881 * As barriers are counted per resource,
1882 * we'd need to suspend io on all devices of a resource.
1884 wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1885 /* and for any other previously queued work */
1886 drbd_flush_workqueue(&connection->sender_work);
1888 rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1889 retcode = (enum drbd_ret_code)rv;
1890 drbd_resume_io(device);
1891 if (rv < SS_SUCCESS)
1894 if (!get_ldev_if_state(device, D_ATTACHING))
1895 goto force_diskless;
1897 if (!device->bitmap) {
1898 if (drbd_bm_init(device)) {
1899 retcode = ERR_NOMEM;
1900 goto force_diskless_dec;
1904 if (device->state.pdsk != D_UP_TO_DATE && device->ed_uuid &&
1905 (device->state.role == R_PRIMARY || device->state.peer == R_PRIMARY) &&
1906 (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1907 drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1908 (unsigned long long)device->ed_uuid);
1909 retcode = ERR_DATA_NOT_CURRENT;
1910 goto force_diskless_dec;
1913 /* Since we are diskless, fix the activity log first... */
1914 if (drbd_check_al_size(device, new_disk_conf)) {
1915 retcode = ERR_NOMEM;
1916 goto force_diskless_dec;
1919 /* Prevent shrinking of consistent devices ! */
1921 unsigned long long nsz = drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0);
1922 unsigned long long eff = nbc->md.la_size_sect;
1923 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && nsz < eff) {
1924 if (nsz == nbc->disk_conf->disk_size) {
1925 drbd_warn(device, "truncating a consistent device during attach (%llu < %llu)\n", nsz, eff);
1927 drbd_warn(device, "refusing to truncate a consistent device (%llu < %llu)\n", nsz, eff);
1928 drbd_msg_sprintf_info(adm_ctx.reply_skb,
1929 "To-be-attached device has last effective > current size, and is consistent\n"
1930 "(%llu > %llu sectors). Refusing to attach.", eff, nsz);
1931 retcode = ERR_IMPLICIT_SHRINK;
1932 goto force_diskless_dec;
1937 lock_all_resources();
1938 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1939 if (retcode != NO_ERROR) {
1940 unlock_all_resources();
1941 goto force_diskless_dec;
1944 /* Reset the "barriers don't work" bits here, then force meta data to
1945 * be written, to ensure we determine if barriers are supported. */
1946 if (new_disk_conf->md_flushes)
1947 clear_bit(MD_NO_FUA, &device->flags);
1949 set_bit(MD_NO_FUA, &device->flags);
1951 /* Point of no return reached.
1952 * Devices and memory are no longer released by error cleanup below.
1953 * now device takes over responsibility, and the state engine should
1954 * clean it up somewhere. */
1955 D_ASSERT(device, device->ldev == NULL);
1957 device->resync = resync_lru;
1958 device->rs_plan_s = new_plan;
1961 new_disk_conf = NULL;
1964 drbd_resync_after_changed(device);
1965 drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH);
1966 unlock_all_resources();
1968 if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
1969 set_bit(CRASHED_PRIMARY, &device->flags);
1971 clear_bit(CRASHED_PRIMARY, &device->flags);
1973 if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1974 !(device->state.role == R_PRIMARY && device->resource->susp_nod))
1975 set_bit(CRASHED_PRIMARY, &device->flags);
1977 device->send_cnt = 0;
1978 device->recv_cnt = 0;
1979 device->read_cnt = 0;
1980 device->writ_cnt = 0;
1982 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
1984 /* If I am currently not R_PRIMARY,
1985 * but meta data primary indicator is set,
1986 * I just now recover from a hard crash,
1987 * and have been R_PRIMARY before that crash.
1989 * Now, if I had no connection before that crash
1990 * (have been degraded R_PRIMARY), chances are that
1991 * I won't find my peer now either.
1993 * In that case, and _only_ in that case,
1994 * we use the degr-wfc-timeout instead of the default,
1995 * so we can automatically recover from a crash of a
1996 * degraded but active "cluster" after a certain timeout.
1998 clear_bit(USE_DEGR_WFC_T, &device->flags);
1999 if (device->state.role != R_PRIMARY &&
2000 drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
2001 !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
2002 set_bit(USE_DEGR_WFC_T, &device->flags);
2004 dd = drbd_determine_dev_size(device, 0, NULL);
2005 if (dd <= DS_ERROR) {
2006 retcode = ERR_NOMEM_BITMAP;
2007 goto force_diskless_dec;
2008 } else if (dd == DS_GREW)
2009 set_bit(RESYNC_AFTER_NEG, &device->flags);
2011 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
2012 (test_bit(CRASHED_PRIMARY, &device->flags) &&
2013 drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
2014 drbd_info(device, "Assuming that all blocks are out of sync "
2015 "(aka FullSync)\n");
2016 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2017 "set_n_write from attaching", BM_LOCKED_MASK)) {
2018 retcode = ERR_IO_MD_DISK;
2019 goto force_diskless_dec;
2022 if (drbd_bitmap_io(device, &drbd_bm_read,
2023 "read from attaching", BM_LOCKED_MASK)) {
2024 retcode = ERR_IO_MD_DISK;
2025 goto force_diskless_dec;
2029 if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
2030 drbd_suspend_al(device); /* IO is still suspended here... */
2032 spin_lock_irq(&device->resource->req_lock);
2033 os = drbd_read_state(device);
2035 /* If MDF_CONSISTENT is not set go into inconsistent state,
2036 otherwise investigate MDF_WasUpToDate...
2037 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
2038 otherwise into D_CONSISTENT state.
2040 if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
2041 if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
2042 ns.disk = D_CONSISTENT;
2044 ns.disk = D_OUTDATED;
2046 ns.disk = D_INCONSISTENT;
2049 if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
2050 ns.pdsk = D_OUTDATED;
2053 if (ns.disk == D_CONSISTENT &&
2054 (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
2055 ns.disk = D_UP_TO_DATE;
2057 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
2058 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
2059 this point, because drbd_request_state() modifies these
2062 if (rcu_dereference(device->ldev->disk_conf)->al_updates)
2063 device->ldev->md.flags &= ~MDF_AL_DISABLED;
2065 device->ldev->md.flags |= MDF_AL_DISABLED;
2069 /* In case we are C_CONNECTED postpone any decision on the new disk
2070 state after the negotiation phase. */
2071 if (device->state.conn == C_CONNECTED) {
2072 device->new_state_tmp.i = ns.i;
2074 ns.disk = D_NEGOTIATING;
2076 /* We expect to receive up-to-date UUIDs soon.
2077 To avoid a race in receive_state, free p_uuid while
2078 holding req_lock. I.e. atomic with the state change */
2079 kfree(device->p_uuid);
2080 device->p_uuid = NULL;
2083 rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
2084 spin_unlock_irq(&device->resource->req_lock);
2086 if (rv < SS_SUCCESS)
2087 goto force_diskless_dec;
2089 mod_timer(&device->request_timer, jiffies + HZ);
2091 if (device->state.role == R_PRIMARY)
2092 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
2094 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
2096 drbd_md_mark_dirty(device);
2097 drbd_md_sync(device);
2099 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
2101 conn_reconfig_done(connection);
2102 mutex_unlock(&adm_ctx.resource->adm_mutex);
2103 drbd_adm_finish(&adm_ctx, info, retcode);
2109 drbd_force_state(device, NS(disk, D_DISKLESS));
2110 drbd_md_sync(device);
2112 conn_reconfig_done(connection);
2114 close_backing_dev(device, nbc->md_bdev, nbc->md_bdev != nbc->backing_bdev);
2115 close_backing_dev(device, nbc->backing_bdev, true);
2118 kfree(new_disk_conf);
2119 lc_destroy(resync_lru);
2121 mutex_unlock(&adm_ctx.resource->adm_mutex);
2123 drbd_adm_finish(&adm_ctx, info, retcode);
2127 static int adm_detach(struct drbd_device *device, int force)
2130 set_bit(FORCE_DETACH, &device->flags);
2131 drbd_force_state(device, NS(disk, D_FAILED));
2135 return drbd_request_detach_interruptible(device);
2138 /* Detaching the disk is a process in multiple stages. First we need to lock
2139 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
2140 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
2141 * internal references as well.
2142 * Only then we have finally detached. */
2143 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
2145 struct drbd_config_context adm_ctx;
2146 enum drbd_ret_code retcode;
2147 struct detach_parms parms = { };
2150 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2151 if (!adm_ctx.reply_skb)
2153 if (retcode != NO_ERROR)
2156 if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
2157 err = detach_parms_from_attrs(&parms, info);
2159 retcode = ERR_MANDATORY_TAG;
2160 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2165 mutex_lock(&adm_ctx.resource->adm_mutex);
2166 retcode = adm_detach(adm_ctx.device, parms.force_detach);
2167 mutex_unlock(&adm_ctx.resource->adm_mutex);
2169 drbd_adm_finish(&adm_ctx, info, retcode);
2173 static bool conn_resync_running(struct drbd_connection *connection)
2175 struct drbd_peer_device *peer_device;
2180 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2181 struct drbd_device *device = peer_device->device;
2182 if (device->state.conn == C_SYNC_SOURCE ||
2183 device->state.conn == C_SYNC_TARGET ||
2184 device->state.conn == C_PAUSED_SYNC_S ||
2185 device->state.conn == C_PAUSED_SYNC_T) {
2195 static bool conn_ov_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_VERIFY_S ||
2205 device->state.conn == C_VERIFY_T) {
2215 static enum drbd_ret_code
2216 _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2218 struct drbd_peer_device *peer_device;
2221 if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2222 if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2223 return ERR_NEED_APV_100;
2225 if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2226 return ERR_NEED_APV_100;
2228 if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2229 return ERR_NEED_APV_100;
2232 if (!new_net_conf->two_primaries &&
2233 conn_highest_role(connection) == R_PRIMARY &&
2234 conn_highest_peer(connection) == R_PRIMARY)
2235 return ERR_NEED_ALLOW_TWO_PRI;
2237 if (new_net_conf->two_primaries &&
2238 (new_net_conf->wire_protocol != DRBD_PROT_C))
2239 return ERR_NOT_PROTO_C;
2241 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2242 struct drbd_device *device = peer_device->device;
2243 if (get_ldev(device)) {
2244 enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2246 if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2247 return ERR_STONITH_AND_PROT_A;
2249 if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2250 return ERR_DISCARD_IMPOSSIBLE;
2253 if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2254 return ERR_CONG_NOT_PROTO_A;
2259 static enum drbd_ret_code
2260 check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2262 enum drbd_ret_code rv;
2263 struct drbd_peer_device *peer_device;
2267 rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2270 /* connection->peer_devices protected by genl_lock() here */
2271 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2272 struct drbd_device *device = peer_device->device;
2273 if (!device->bitmap) {
2274 if (drbd_bm_init(device))
2283 struct crypto_shash *verify_tfm;
2284 struct crypto_shash *csums_tfm;
2285 struct crypto_shash *cram_hmac_tfm;
2286 struct crypto_shash *integrity_tfm;
2290 alloc_shash(struct crypto_shash **tfm, char *tfm_name, int err_alg)
2295 *tfm = crypto_alloc_shash(tfm_name, 0, 0);
2304 static enum drbd_ret_code
2305 alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2307 char hmac_name[CRYPTO_MAX_ALG_NAME];
2308 enum drbd_ret_code rv;
2310 rv = alloc_shash(&crypto->csums_tfm, new_net_conf->csums_alg,
2314 rv = alloc_shash(&crypto->verify_tfm, new_net_conf->verify_alg,
2318 rv = alloc_shash(&crypto->integrity_tfm, new_net_conf->integrity_alg,
2322 if (new_net_conf->cram_hmac_alg[0] != 0) {
2323 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
2324 new_net_conf->cram_hmac_alg);
2326 rv = alloc_shash(&crypto->cram_hmac_tfm, hmac_name,
2333 static void free_crypto(struct crypto *crypto)
2335 crypto_free_shash(crypto->cram_hmac_tfm);
2336 crypto_free_shash(crypto->integrity_tfm);
2337 crypto_free_shash(crypto->csums_tfm);
2338 crypto_free_shash(crypto->verify_tfm);
2341 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2343 struct drbd_config_context adm_ctx;
2344 enum drbd_ret_code retcode;
2345 struct drbd_connection *connection;
2346 struct net_conf *old_net_conf, *new_net_conf = NULL;
2348 int ovr; /* online verify running */
2349 int rsr; /* re-sync running */
2350 struct crypto crypto = { };
2352 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2353 if (!adm_ctx.reply_skb)
2355 if (retcode != NO_ERROR)
2358 connection = adm_ctx.connection;
2359 mutex_lock(&adm_ctx.resource->adm_mutex);
2361 new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
2362 if (!new_net_conf) {
2363 retcode = ERR_NOMEM;
2367 conn_reconfig_start(connection);
2369 mutex_lock(&connection->data.mutex);
2370 mutex_lock(&connection->resource->conf_update);
2371 old_net_conf = connection->net_conf;
2373 if (!old_net_conf) {
2374 drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect");
2375 retcode = ERR_INVALID_REQUEST;
2379 *new_net_conf = *old_net_conf;
2380 if (should_set_defaults(info))
2381 set_net_conf_defaults(new_net_conf);
2383 err = net_conf_from_attrs_for_change(new_net_conf, info);
2384 if (err && err != -ENOMSG) {
2385 retcode = ERR_MANDATORY_TAG;
2386 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2390 retcode = check_net_options(connection, new_net_conf);
2391 if (retcode != NO_ERROR)
2394 /* re-sync running */
2395 rsr = conn_resync_running(connection);
2396 if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2397 retcode = ERR_CSUMS_RESYNC_RUNNING;
2401 /* online verify running */
2402 ovr = conn_ov_running(connection);
2403 if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2404 retcode = ERR_VERIFY_RUNNING;
2408 retcode = alloc_crypto(&crypto, new_net_conf);
2409 if (retcode != NO_ERROR)
2412 rcu_assign_pointer(connection->net_conf, new_net_conf);
2415 crypto_free_shash(connection->csums_tfm);
2416 connection->csums_tfm = crypto.csums_tfm;
2417 crypto.csums_tfm = NULL;
2420 crypto_free_shash(connection->verify_tfm);
2421 connection->verify_tfm = crypto.verify_tfm;
2422 crypto.verify_tfm = NULL;
2425 crypto_free_shash(connection->integrity_tfm);
2426 connection->integrity_tfm = crypto.integrity_tfm;
2427 if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2428 /* Do this without trying to take connection->data.mutex again. */
2429 __drbd_send_protocol(connection, P_PROTOCOL_UPDATE);
2431 crypto_free_shash(connection->cram_hmac_tfm);
2432 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2434 mutex_unlock(&connection->resource->conf_update);
2435 mutex_unlock(&connection->data.mutex);
2436 kvfree_rcu(old_net_conf);
2438 if (connection->cstate >= C_WF_REPORT_PARAMS) {
2439 struct drbd_peer_device *peer_device;
2442 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2443 drbd_send_sync_param(peer_device);
2449 mutex_unlock(&connection->resource->conf_update);
2450 mutex_unlock(&connection->data.mutex);
2451 free_crypto(&crypto);
2452 kfree(new_net_conf);
2454 conn_reconfig_done(connection);
2456 mutex_unlock(&adm_ctx.resource->adm_mutex);
2458 drbd_adm_finish(&adm_ctx, info, retcode);
2462 static void connection_to_info(struct connection_info *info,
2463 struct drbd_connection *connection)
2465 info->conn_connection_state = connection->cstate;
2466 info->conn_role = conn_highest_peer(connection);
2469 static void peer_device_to_info(struct peer_device_info *info,
2470 struct drbd_peer_device *peer_device)
2472 struct drbd_device *device = peer_device->device;
2474 info->peer_repl_state =
2475 max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn);
2476 info->peer_disk_state = device->state.pdsk;
2477 info->peer_resync_susp_user = device->state.user_isp;
2478 info->peer_resync_susp_peer = device->state.peer_isp;
2479 info->peer_resync_susp_dependency = device->state.aftr_isp;
2482 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2484 struct connection_info connection_info;
2485 enum drbd_notification_type flags;
2486 unsigned int peer_devices = 0;
2487 struct drbd_config_context adm_ctx;
2488 struct drbd_peer_device *peer_device;
2489 struct net_conf *old_net_conf, *new_net_conf = NULL;
2490 struct crypto crypto = { };
2491 struct drbd_resource *resource;
2492 struct drbd_connection *connection;
2493 enum drbd_ret_code retcode;
2494 enum drbd_state_rv rv;
2498 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2500 if (!adm_ctx.reply_skb)
2502 if (retcode != NO_ERROR)
2504 if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2505 drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing");
2506 retcode = ERR_INVALID_REQUEST;
2510 /* No need for _rcu here. All reconfiguration is
2511 * strictly serialized on genl_lock(). We are protected against
2512 * concurrent reconfiguration/addition/deletion */
2513 for_each_resource(resource, &drbd_resources) {
2514 for_each_connection(connection, resource) {
2515 if (nla_len(adm_ctx.my_addr) == connection->my_addr_len &&
2516 !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr,
2517 connection->my_addr_len)) {
2518 retcode = ERR_LOCAL_ADDR;
2522 if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len &&
2523 !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr,
2524 connection->peer_addr_len)) {
2525 retcode = ERR_PEER_ADDR;
2531 mutex_lock(&adm_ctx.resource->adm_mutex);
2532 connection = first_connection(adm_ctx.resource);
2533 conn_reconfig_start(connection);
2535 if (connection->cstate > C_STANDALONE) {
2536 retcode = ERR_NET_CONFIGURED;
2540 /* allocation not in the IO path, drbdsetup / netlink process context */
2541 new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL);
2542 if (!new_net_conf) {
2543 retcode = ERR_NOMEM;
2547 set_net_conf_defaults(new_net_conf);
2549 err = net_conf_from_attrs(new_net_conf, info);
2550 if (err && err != -ENOMSG) {
2551 retcode = ERR_MANDATORY_TAG;
2552 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2556 retcode = check_net_options(connection, new_net_conf);
2557 if (retcode != NO_ERROR)
2560 retcode = alloc_crypto(&crypto, new_net_conf);
2561 if (retcode != NO_ERROR)
2564 ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2566 drbd_flush_workqueue(&connection->sender_work);
2568 mutex_lock(&adm_ctx.resource->conf_update);
2569 old_net_conf = connection->net_conf;
2571 retcode = ERR_NET_CONFIGURED;
2572 mutex_unlock(&adm_ctx.resource->conf_update);
2575 rcu_assign_pointer(connection->net_conf, new_net_conf);
2577 conn_free_crypto(connection);
2578 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2579 connection->integrity_tfm = crypto.integrity_tfm;
2580 connection->csums_tfm = crypto.csums_tfm;
2581 connection->verify_tfm = crypto.verify_tfm;
2583 connection->my_addr_len = nla_len(adm_ctx.my_addr);
2584 memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2585 connection->peer_addr_len = nla_len(adm_ctx.peer_addr);
2586 memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2588 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2592 connection_to_info(&connection_info, connection);
2593 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2594 mutex_lock(¬ification_mutex);
2595 notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags);
2596 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2597 struct peer_device_info peer_device_info;
2599 peer_device_to_info(&peer_device_info, peer_device);
2600 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2601 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags);
2603 mutex_unlock(¬ification_mutex);
2604 mutex_unlock(&adm_ctx.resource->conf_update);
2607 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2608 struct drbd_device *device = peer_device->device;
2609 device->send_cnt = 0;
2610 device->recv_cnt = 0;
2614 rv = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2616 conn_reconfig_done(connection);
2617 mutex_unlock(&adm_ctx.resource->adm_mutex);
2618 drbd_adm_finish(&adm_ctx, info, rv);
2622 free_crypto(&crypto);
2623 kfree(new_net_conf);
2625 conn_reconfig_done(connection);
2626 mutex_unlock(&adm_ctx.resource->adm_mutex);
2628 drbd_adm_finish(&adm_ctx, info, retcode);
2632 static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2634 enum drbd_conns cstate;
2635 enum drbd_state_rv rv;
2638 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2639 force ? CS_HARD : 0);
2642 case SS_NOTHING_TO_DO:
2644 case SS_ALREADY_STANDALONE:
2646 case SS_PRIMARY_NOP:
2647 /* Our state checking code wants to see the peer outdated. */
2648 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2650 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2651 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2654 case SS_CW_FAILED_BY_PEER:
2655 spin_lock_irq(&connection->resource->req_lock);
2656 cstate = connection->cstate;
2657 spin_unlock_irq(&connection->resource->req_lock);
2658 if (cstate <= C_WF_CONNECTION)
2660 /* The peer probably wants to see us outdated. */
2661 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2662 disk, D_OUTDATED), 0);
2663 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2664 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2669 /* no special handling necessary */
2672 if (rv >= SS_SUCCESS) {
2673 enum drbd_state_rv rv2;
2674 /* No one else can reconfigure the network while I am here.
2675 * The state handling only uses drbd_thread_stop_nowait(),
2676 * we want to really wait here until the receiver is no more.
2678 drbd_thread_stop(&connection->receiver);
2680 /* Race breaker. This additional state change request may be
2681 * necessary, if this was a forced disconnect during a receiver
2682 * restart. We may have "killed" the receiver thread just
2683 * after drbd_receiver() returned. Typically, we should be
2684 * C_STANDALONE already, now, and this becomes a no-op.
2686 rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2687 CS_VERBOSE | CS_HARD);
2688 if (rv2 < SS_SUCCESS)
2689 drbd_err(connection,
2690 "unexpected rv2=%d in conn_try_disconnect()\n",
2692 /* Unlike in DRBD 9, the state engine has generated
2693 * NOTIFY_DESTROY events before clearing connection->net_conf. */
2698 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2700 struct drbd_config_context adm_ctx;
2701 struct disconnect_parms parms;
2702 struct drbd_connection *connection;
2703 enum drbd_state_rv rv;
2704 enum drbd_ret_code retcode;
2707 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2708 if (!adm_ctx.reply_skb)
2710 if (retcode != NO_ERROR)
2713 connection = adm_ctx.connection;
2714 memset(&parms, 0, sizeof(parms));
2715 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2716 err = disconnect_parms_from_attrs(&parms, info);
2718 retcode = ERR_MANDATORY_TAG;
2719 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2724 mutex_lock(&adm_ctx.resource->adm_mutex);
2725 rv = conn_try_disconnect(connection, parms.force_disconnect);
2726 mutex_unlock(&adm_ctx.resource->adm_mutex);
2727 if (rv < SS_SUCCESS) {
2728 drbd_adm_finish(&adm_ctx, info, rv);
2733 drbd_adm_finish(&adm_ctx, info, retcode);
2737 void resync_after_online_grow(struct drbd_device *device)
2739 int iass; /* I am sync source */
2741 drbd_info(device, "Resync of new storage after online grow\n");
2742 if (device->state.role != device->state.peer)
2743 iass = (device->state.role == R_PRIMARY);
2745 iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2748 drbd_start_resync(device, C_SYNC_SOURCE);
2750 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2753 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2755 struct drbd_config_context adm_ctx;
2756 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2757 struct resize_parms rs;
2758 struct drbd_device *device;
2759 enum drbd_ret_code retcode;
2760 enum determine_dev_size dd;
2761 bool change_al_layout = false;
2762 enum dds_flags ddsf;
2766 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2767 if (!adm_ctx.reply_skb)
2769 if (retcode != NO_ERROR)
2772 mutex_lock(&adm_ctx.resource->adm_mutex);
2773 device = adm_ctx.device;
2774 if (!get_ldev(device)) {
2775 retcode = ERR_NO_DISK;
2779 memset(&rs, 0, sizeof(struct resize_parms));
2780 rs.al_stripes = device->ldev->md.al_stripes;
2781 rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2782 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2783 err = resize_parms_from_attrs(&rs, info);
2785 retcode = ERR_MANDATORY_TAG;
2786 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2791 if (device->state.conn > C_CONNECTED) {
2792 retcode = ERR_RESIZE_RESYNC;
2796 if (device->state.role == R_SECONDARY &&
2797 device->state.peer == R_SECONDARY) {
2798 retcode = ERR_NO_PRIMARY;
2802 if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2803 retcode = ERR_NEED_APV_93;
2808 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2810 if (u_size != (sector_t)rs.resize_size) {
2811 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2812 if (!new_disk_conf) {
2813 retcode = ERR_NOMEM;
2818 if (device->ldev->md.al_stripes != rs.al_stripes ||
2819 device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2820 u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2822 if (al_size_k > (16 * 1024 * 1024)) {
2823 retcode = ERR_MD_LAYOUT_TOO_BIG;
2827 if (al_size_k < MD_32kB_SECT/2) {
2828 retcode = ERR_MD_LAYOUT_TOO_SMALL;
2832 if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2833 retcode = ERR_MD_LAYOUT_CONNECTED;
2837 change_al_layout = true;
2840 if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev))
2841 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
2843 if (new_disk_conf) {
2844 mutex_lock(&device->resource->conf_update);
2845 old_disk_conf = device->ldev->disk_conf;
2846 *new_disk_conf = *old_disk_conf;
2847 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2848 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2849 mutex_unlock(&device->resource->conf_update);
2850 kvfree_rcu(old_disk_conf);
2851 new_disk_conf = NULL;
2854 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2855 dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL);
2856 drbd_md_sync(device);
2858 if (dd == DS_ERROR) {
2859 retcode = ERR_NOMEM_BITMAP;
2861 } else if (dd == DS_ERROR_SPACE_MD) {
2862 retcode = ERR_MD_LAYOUT_NO_FIT;
2864 } else if (dd == DS_ERROR_SHRINK) {
2865 retcode = ERR_IMPLICIT_SHRINK;
2869 if (device->state.conn == C_CONNECTED) {
2871 set_bit(RESIZE_PENDING, &device->flags);
2873 drbd_send_uuids(first_peer_device(device));
2874 drbd_send_sizes(first_peer_device(device), 1, ddsf);
2878 mutex_unlock(&adm_ctx.resource->adm_mutex);
2880 drbd_adm_finish(&adm_ctx, info, retcode);
2885 kfree(new_disk_conf);
2889 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2891 struct drbd_config_context adm_ctx;
2892 enum drbd_ret_code retcode;
2893 struct res_opts res_opts;
2896 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2897 if (!adm_ctx.reply_skb)
2899 if (retcode != NO_ERROR)
2902 res_opts = adm_ctx.resource->res_opts;
2903 if (should_set_defaults(info))
2904 set_res_opts_defaults(&res_opts);
2906 err = res_opts_from_attrs(&res_opts, info);
2907 if (err && err != -ENOMSG) {
2908 retcode = ERR_MANDATORY_TAG;
2909 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2913 mutex_lock(&adm_ctx.resource->adm_mutex);
2914 err = set_resource_options(adm_ctx.resource, &res_opts);
2916 retcode = ERR_INVALID_REQUEST;
2918 retcode = ERR_NOMEM;
2920 mutex_unlock(&adm_ctx.resource->adm_mutex);
2923 drbd_adm_finish(&adm_ctx, info, retcode);
2927 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2929 struct drbd_config_context adm_ctx;
2930 struct drbd_device *device;
2931 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2933 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2934 if (!adm_ctx.reply_skb)
2936 if (retcode != NO_ERROR)
2939 device = adm_ctx.device;
2940 if (!get_ldev(device)) {
2941 retcode = ERR_NO_DISK;
2945 mutex_lock(&adm_ctx.resource->adm_mutex);
2947 /* If there is still bitmap IO pending, probably because of a previous
2948 * resync just being finished, wait for it before requesting a new resync.
2949 * Also wait for it's after_state_ch(). */
2950 drbd_suspend_io(device);
2951 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2952 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2954 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2955 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2956 * try to start a resync handshake as sync target for full sync.
2958 if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
2959 retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
2960 if (retcode >= SS_SUCCESS) {
2961 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2962 "set_n_write from invalidate", BM_LOCKED_MASK))
2963 retcode = ERR_IO_MD_DISK;
2966 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
2967 drbd_resume_io(device);
2968 mutex_unlock(&adm_ctx.resource->adm_mutex);
2971 drbd_adm_finish(&adm_ctx, info, retcode);
2975 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2976 union drbd_state mask, union drbd_state val)
2978 struct drbd_config_context adm_ctx;
2979 enum drbd_ret_code retcode;
2981 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2982 if (!adm_ctx.reply_skb)
2984 if (retcode != NO_ERROR)
2987 mutex_lock(&adm_ctx.resource->adm_mutex);
2988 retcode = drbd_request_state(adm_ctx.device, mask, val);
2989 mutex_unlock(&adm_ctx.resource->adm_mutex);
2991 drbd_adm_finish(&adm_ctx, info, retcode);
2995 static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local)
2999 rv = drbd_bmio_set_n_write(device);
3000 drbd_suspend_al(device);
3004 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
3006 struct drbd_config_context adm_ctx;
3007 int retcode; /* drbd_ret_code, drbd_state_rv */
3008 struct drbd_device *device;
3010 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3011 if (!adm_ctx.reply_skb)
3013 if (retcode != NO_ERROR)
3016 device = adm_ctx.device;
3017 if (!get_ldev(device)) {
3018 retcode = ERR_NO_DISK;
3022 mutex_lock(&adm_ctx.resource->adm_mutex);
3024 /* If there is still bitmap IO pending, probably because of a previous
3025 * resync just being finished, wait for it before requesting a new resync.
3026 * Also wait for it's after_state_ch(). */
3027 drbd_suspend_io(device);
3028 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
3029 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
3031 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
3032 * in the bitmap. Otherwise, try to start a resync handshake
3033 * as sync source for full sync.
3035 if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
3036 /* The peer will get a resync upon connect anyways. Just make that
3037 into a full resync. */
3038 retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
3039 if (retcode >= SS_SUCCESS) {
3040 if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al,
3041 "set_n_write from invalidate_peer",
3042 BM_LOCKED_SET_ALLOWED))
3043 retcode = ERR_IO_MD_DISK;
3046 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
3047 drbd_resume_io(device);
3048 mutex_unlock(&adm_ctx.resource->adm_mutex);
3051 drbd_adm_finish(&adm_ctx, info, retcode);
3055 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
3057 struct drbd_config_context adm_ctx;
3058 enum drbd_ret_code retcode;
3060 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3061 if (!adm_ctx.reply_skb)
3063 if (retcode != NO_ERROR)
3066 mutex_lock(&adm_ctx.resource->adm_mutex);
3067 if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
3068 retcode = ERR_PAUSE_IS_SET;
3069 mutex_unlock(&adm_ctx.resource->adm_mutex);
3071 drbd_adm_finish(&adm_ctx, info, retcode);
3075 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
3077 struct drbd_config_context adm_ctx;
3078 union drbd_dev_state s;
3079 enum drbd_ret_code retcode;
3081 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3082 if (!adm_ctx.reply_skb)
3084 if (retcode != NO_ERROR)
3087 mutex_lock(&adm_ctx.resource->adm_mutex);
3088 if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
3089 s = adm_ctx.device->state;
3090 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
3091 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
3092 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
3094 retcode = ERR_PAUSE_IS_CLEAR;
3097 mutex_unlock(&adm_ctx.resource->adm_mutex);
3099 drbd_adm_finish(&adm_ctx, info, retcode);
3103 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
3105 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
3108 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
3110 struct drbd_config_context adm_ctx;
3111 struct drbd_device *device;
3112 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3114 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3115 if (!adm_ctx.reply_skb)
3117 if (retcode != NO_ERROR)
3120 mutex_lock(&adm_ctx.resource->adm_mutex);
3121 device = adm_ctx.device;
3122 if (test_bit(NEW_CUR_UUID, &device->flags)) {
3123 if (get_ldev_if_state(device, D_ATTACHING)) {
3124 drbd_uuid_new_current(device);
3127 /* This is effectively a multi-stage "forced down".
3128 * The NEW_CUR_UUID bit is supposedly only set, if we
3129 * lost the replication connection, and are configured
3130 * to freeze IO and wait for some fence-peer handler.
3131 * So we still don't have a replication connection.
3132 * And now we don't have a local disk either. After
3133 * resume, we will fail all pending and new IO, because
3134 * we don't have any data anymore. Which means we will
3135 * eventually be able to terminate all users of this
3136 * device, and then take it down. By bumping the
3137 * "effective" data uuid, we make sure that you really
3138 * need to tear down before you reconfigure, we will
3139 * the refuse to re-connect or re-attach (because no
3140 * matching real data uuid exists).
3143 get_random_bytes(&val, sizeof(u64));
3144 drbd_set_ed_uuid(device, val);
3145 drbd_warn(device, "Resumed without access to data; please tear down before attempting to re-configure.\n");
3147 clear_bit(NEW_CUR_UUID, &device->flags);
3149 drbd_suspend_io(device);
3150 retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
3151 if (retcode == SS_SUCCESS) {
3152 if (device->state.conn < C_CONNECTED)
3153 tl_clear(first_peer_device(device)->connection);
3154 if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
3155 tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO);
3157 drbd_resume_io(device);
3158 mutex_unlock(&adm_ctx.resource->adm_mutex);
3160 drbd_adm_finish(&adm_ctx, info, retcode);
3164 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
3166 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
3169 static int nla_put_drbd_cfg_context(struct sk_buff *skb,
3170 struct drbd_resource *resource,
3171 struct drbd_connection *connection,
3172 struct drbd_device *device)
3175 nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_CONTEXT);
3177 goto nla_put_failure;
3179 nla_put_u32(skb, T_ctx_volume, device->vnr))
3180 goto nla_put_failure;
3181 if (nla_put_string(skb, T_ctx_resource_name, resource->name))
3182 goto nla_put_failure;
3184 if (connection->my_addr_len &&
3185 nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr))
3186 goto nla_put_failure;
3187 if (connection->peer_addr_len &&
3188 nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr))
3189 goto nla_put_failure;
3191 nla_nest_end(skb, nla);
3196 nla_nest_cancel(skb, nla);
3201 * The generic netlink dump callbacks are called outside the genl_lock(), so
3202 * they cannot use the simple attribute parsing code which uses global
3205 static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr)
3207 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3208 const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3211 nla = nla_find(nlmsg_attrdata(nlh, hdrlen), nlmsg_attrlen(nlh, hdrlen),
3212 DRBD_NLA_CFG_CONTEXT);
3215 return drbd_nla_find_nested(maxtype, nla, __nla_type(attr));
3218 static void resource_to_info(struct resource_info *, struct drbd_resource *);
3220 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb)
3222 struct drbd_genlmsghdr *dh;
3223 struct drbd_resource *resource;
3224 struct resource_info resource_info;
3225 struct resource_statistics resource_statistics;
3230 for_each_resource_rcu(resource, &drbd_resources)
3231 if (resource == (struct drbd_resource *)cb->args[0])
3232 goto found_resource;
3233 err = 0; /* resource was probably deleted */
3236 resource = list_entry(&drbd_resources,
3237 struct drbd_resource, resources);
3240 list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) {
3247 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3248 cb->nlh->nlmsg_seq, &drbd_genl_family,
3249 NLM_F_MULTI, DRBD_ADM_GET_RESOURCES);
3254 dh->ret_code = NO_ERROR;
3255 err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL);
3258 err = res_opts_to_skb(skb, &resource->res_opts, !capable(CAP_SYS_ADMIN));
3261 resource_to_info(&resource_info, resource);
3262 err = resource_info_to_skb(skb, &resource_info, !capable(CAP_SYS_ADMIN));
3265 resource_statistics.res_stat_write_ordering = resource->write_ordering;
3266 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
3269 cb->args[0] = (long)resource;
3270 genlmsg_end(skb, dh);
3280 static void device_to_statistics(struct device_statistics *s,
3281 struct drbd_device *device)
3283 memset(s, 0, sizeof(*s));
3284 s->dev_upper_blocked = !may_inc_ap_bio(device);
3285 if (get_ldev(device)) {
3286 struct drbd_md *md = &device->ldev->md;
3287 u64 *history_uuids = (u64 *)s->history_uuids;
3290 spin_lock_irq(&md->uuid_lock);
3291 s->dev_current_uuid = md->uuid[UI_CURRENT];
3292 BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1);
3293 for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++)
3294 history_uuids[n] = md->uuid[UI_HISTORY_START + n];
3295 for (; n < HISTORY_UUIDS; n++)
3296 history_uuids[n] = 0;
3297 s->history_uuids_len = HISTORY_UUIDS;
3298 spin_unlock_irq(&md->uuid_lock);
3300 s->dev_disk_flags = md->flags;
3303 s->dev_size = get_capacity(device->vdisk);
3304 s->dev_read = device->read_cnt;
3305 s->dev_write = device->writ_cnt;
3306 s->dev_al_writes = device->al_writ_cnt;
3307 s->dev_bm_writes = device->bm_writ_cnt;
3308 s->dev_upper_pending = atomic_read(&device->ap_bio_cnt);
3309 s->dev_lower_pending = atomic_read(&device->local_cnt);
3310 s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags);
3311 s->dev_exposed_data_uuid = device->ed_uuid;
3314 static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr)
3317 struct drbd_resource *resource =
3318 (struct drbd_resource *)cb->args[0];
3319 kref_put(&resource->kref, drbd_destroy_resource);
3325 int drbd_adm_dump_devices_done(struct netlink_callback *cb) {
3326 return put_resource_in_arg0(cb, 7);
3329 static void device_to_info(struct device_info *, struct drbd_device *);
3331 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb)
3333 struct nlattr *resource_filter;
3334 struct drbd_resource *resource;
3335 struct drbd_device *device;
3336 int minor, err, retcode;
3337 struct drbd_genlmsghdr *dh;
3338 struct device_info device_info;
3339 struct device_statistics device_statistics;
3340 struct idr *idr_to_search;
3342 resource = (struct drbd_resource *)cb->args[0];
3343 if (!cb->args[0] && !cb->args[1]) {
3344 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3345 if (resource_filter) {
3346 retcode = ERR_RES_NOT_KNOWN;
3347 resource = drbd_find_resource(nla_data(resource_filter));
3350 cb->args[0] = (long)resource;
3355 minor = cb->args[1];
3356 idr_to_search = resource ? &resource->devices : &drbd_devices;
3357 device = idr_get_next(idr_to_search, &minor);
3362 idr_for_each_entry_continue(idr_to_search, device, minor) {
3364 goto put_result; /* only one iteration */
3367 goto out; /* no more devices */
3370 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3371 cb->nlh->nlmsg_seq, &drbd_genl_family,
3372 NLM_F_MULTI, DRBD_ADM_GET_DEVICES);
3376 dh->ret_code = retcode;
3378 if (retcode == NO_ERROR) {
3379 dh->minor = device->minor;
3380 err = nla_put_drbd_cfg_context(skb, device->resource, NULL, device);
3383 if (get_ldev(device)) {
3384 struct disk_conf *disk_conf =
3385 rcu_dereference(device->ldev->disk_conf);
3387 err = disk_conf_to_skb(skb, disk_conf, !capable(CAP_SYS_ADMIN));
3392 device_to_info(&device_info, device);
3393 err = device_info_to_skb(skb, &device_info, !capable(CAP_SYS_ADMIN));
3397 device_to_statistics(&device_statistics, device);
3398 err = device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
3401 cb->args[1] = minor + 1;
3403 genlmsg_end(skb, dh);
3413 int drbd_adm_dump_connections_done(struct netlink_callback *cb)
3415 return put_resource_in_arg0(cb, 6);
3418 enum { SINGLE_RESOURCE, ITERATE_RESOURCES };
3420 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb)
3422 struct nlattr *resource_filter;
3423 struct drbd_resource *resource = NULL, *next_resource;
3424 struct drbd_connection *connection;
3425 int err = 0, retcode;
3426 struct drbd_genlmsghdr *dh;
3427 struct connection_info connection_info;
3428 struct connection_statistics connection_statistics;
3431 resource = (struct drbd_resource *)cb->args[0];
3433 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3434 if (resource_filter) {
3435 retcode = ERR_RES_NOT_KNOWN;
3436 resource = drbd_find_resource(nla_data(resource_filter));
3439 cb->args[0] = (long)resource;
3440 cb->args[1] = SINGLE_RESOURCE;
3444 if (list_empty(&drbd_resources))
3446 resource = list_first_entry(&drbd_resources, struct drbd_resource, resources);
3447 kref_get(&resource->kref);
3448 cb->args[0] = (long)resource;
3449 cb->args[1] = ITERATE_RESOURCES;
3454 mutex_lock(&resource->conf_update);
3457 for_each_connection_rcu(connection, resource)
3458 if (connection == (struct drbd_connection *)cb->args[2])
3459 goto found_connection;
3460 /* connection was probably deleted */
3461 goto no_more_connections;
3463 connection = list_entry(&resource->connections, struct drbd_connection, connections);
3466 list_for_each_entry_continue_rcu(connection, &resource->connections, connections) {
3467 if (!has_net_conf(connection))
3470 goto put_result; /* only one iteration */
3473 no_more_connections:
3474 if (cb->args[1] == ITERATE_RESOURCES) {
3475 for_each_resource_rcu(next_resource, &drbd_resources) {
3476 if (next_resource == resource)
3477 goto found_resource;
3479 /* resource was probably deleted */
3484 list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) {
3485 mutex_unlock(&resource->conf_update);
3486 kref_put(&resource->kref, drbd_destroy_resource);
3487 resource = next_resource;
3488 kref_get(&resource->kref);
3489 cb->args[0] = (long)resource;
3493 goto out; /* no more resources */
3496 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3497 cb->nlh->nlmsg_seq, &drbd_genl_family,
3498 NLM_F_MULTI, DRBD_ADM_GET_CONNECTIONS);
3502 dh->ret_code = retcode;
3504 if (retcode == NO_ERROR) {
3505 struct net_conf *net_conf;
3507 err = nla_put_drbd_cfg_context(skb, resource, connection, NULL);
3510 net_conf = rcu_dereference(connection->net_conf);
3512 err = net_conf_to_skb(skb, net_conf, !capable(CAP_SYS_ADMIN));
3516 connection_to_info(&connection_info, connection);
3517 err = connection_info_to_skb(skb, &connection_info, !capable(CAP_SYS_ADMIN));
3520 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
3521 err = connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
3524 cb->args[2] = (long)connection;
3526 genlmsg_end(skb, dh);
3532 mutex_unlock(&resource->conf_update);
3538 enum mdf_peer_flag {
3539 MDF_PEER_CONNECTED = 1 << 0,
3540 MDF_PEER_OUTDATED = 1 << 1,
3541 MDF_PEER_FENCING = 1 << 2,
3542 MDF_PEER_FULL_SYNC = 1 << 3,
3545 static void peer_device_to_statistics(struct peer_device_statistics *s,
3546 struct drbd_peer_device *peer_device)
3548 struct drbd_device *device = peer_device->device;
3550 memset(s, 0, sizeof(*s));
3551 s->peer_dev_received = device->recv_cnt;
3552 s->peer_dev_sent = device->send_cnt;
3553 s->peer_dev_pending = atomic_read(&device->ap_pending_cnt) +
3554 atomic_read(&device->rs_pending_cnt);
3555 s->peer_dev_unacked = atomic_read(&device->unacked_cnt);
3556 s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9);
3557 s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9);
3558 if (get_ldev(device)) {
3559 struct drbd_md *md = &device->ldev->md;
3561 spin_lock_irq(&md->uuid_lock);
3562 s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP];
3563 spin_unlock_irq(&md->uuid_lock);
3565 (drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ?
3566 MDF_PEER_CONNECTED : 0) +
3567 (drbd_md_test_flag(device->ldev, MDF_CONSISTENT) &&
3568 !drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ?
3569 MDF_PEER_OUTDATED : 0) +
3570 /* FIXME: MDF_PEER_FENCING? */
3571 (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ?
3572 MDF_PEER_FULL_SYNC : 0);
3577 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb)
3579 return put_resource_in_arg0(cb, 9);
3582 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb)
3584 struct nlattr *resource_filter;
3585 struct drbd_resource *resource;
3586 struct drbd_device *device;
3587 struct drbd_peer_device *peer_device = NULL;
3588 int minor, err, retcode;
3589 struct drbd_genlmsghdr *dh;
3590 struct idr *idr_to_search;
3592 resource = (struct drbd_resource *)cb->args[0];
3593 if (!cb->args[0] && !cb->args[1]) {
3594 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3595 if (resource_filter) {
3596 retcode = ERR_RES_NOT_KNOWN;
3597 resource = drbd_find_resource(nla_data(resource_filter));
3601 cb->args[0] = (long)resource;
3605 minor = cb->args[1];
3606 idr_to_search = resource ? &resource->devices : &drbd_devices;
3607 device = idr_find(idr_to_search, minor);
3612 device = idr_get_next(idr_to_search, &minor);
3619 for_each_peer_device(peer_device, device)
3620 if (peer_device == (struct drbd_peer_device *)cb->args[2])
3621 goto found_peer_device;
3622 /* peer device was probably deleted */
3625 /* Make peer_device point to the list head (not the first entry). */
3626 peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices);
3629 list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) {
3630 if (!has_net_conf(peer_device->connection))
3633 goto put_result; /* only one iteration */
3638 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3639 cb->nlh->nlmsg_seq, &drbd_genl_family,
3640 NLM_F_MULTI, DRBD_ADM_GET_PEER_DEVICES);
3644 dh->ret_code = retcode;
3646 if (retcode == NO_ERROR) {
3647 struct peer_device_info peer_device_info;
3648 struct peer_device_statistics peer_device_statistics;
3651 err = nla_put_drbd_cfg_context(skb, device->resource, peer_device->connection, device);
3654 peer_device_to_info(&peer_device_info, peer_device);
3655 err = peer_device_info_to_skb(skb, &peer_device_info, !capable(CAP_SYS_ADMIN));
3658 peer_device_to_statistics(&peer_device_statistics, peer_device);
3659 err = peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
3662 cb->args[1] = minor;
3663 cb->args[2] = (long)peer_device;
3665 genlmsg_end(skb, dh);
3675 * Return the connection of @resource if @resource has exactly one connection.
3677 static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
3679 struct list_head *connections = &resource->connections;
3681 if (list_empty(connections) || connections->next->next != connections)
3683 return list_first_entry(&resource->connections, struct drbd_connection, connections);
3686 static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
3687 const struct sib_info *sib)
3689 struct drbd_resource *resource = device->resource;
3690 struct state_info *si = NULL; /* for sizeof(si->member); */
3694 int exclude_sensitive;
3696 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
3697 * to. So we better exclude_sensitive information.
3699 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
3700 * in the context of the requesting user process. Exclude sensitive
3701 * information, unless current has superuser.
3703 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
3704 * relies on the current implementation of netlink_dump(), which
3705 * executes the dump callback successively from netlink_recvmsg(),
3706 * always in the context of the receiving process */
3707 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
3709 got_ldev = get_ldev(device);
3711 /* We need to add connection name and volume number information still.
3712 * Minor number is in drbd_genlmsghdr. */
3713 if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device))
3714 goto nla_put_failure;
3716 if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive))
3717 goto nla_put_failure;
3721 struct disk_conf *disk_conf;
3723 disk_conf = rcu_dereference(device->ldev->disk_conf);
3724 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
3727 struct net_conf *nc;
3729 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3731 err = net_conf_to_skb(skb, nc, exclude_sensitive);
3735 goto nla_put_failure;
3737 nla = nla_nest_start_noflag(skb, DRBD_NLA_STATE_INFO);
3739 goto nla_put_failure;
3740 if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
3741 nla_put_u32(skb, T_current_state, device->state.i) ||
3742 nla_put_u64_0pad(skb, T_ed_uuid, device->ed_uuid) ||
3743 nla_put_u64_0pad(skb, T_capacity, get_capacity(device->vdisk)) ||
3744 nla_put_u64_0pad(skb, T_send_cnt, device->send_cnt) ||
3745 nla_put_u64_0pad(skb, T_recv_cnt, device->recv_cnt) ||
3746 nla_put_u64_0pad(skb, T_read_cnt, device->read_cnt) ||
3747 nla_put_u64_0pad(skb, T_writ_cnt, device->writ_cnt) ||
3748 nla_put_u64_0pad(skb, T_al_writ_cnt, device->al_writ_cnt) ||
3749 nla_put_u64_0pad(skb, T_bm_writ_cnt, device->bm_writ_cnt) ||
3750 nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) ||
3751 nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) ||
3752 nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt)))
3753 goto nla_put_failure;
3758 spin_lock_irq(&device->ldev->md.uuid_lock);
3759 err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid);
3760 spin_unlock_irq(&device->ldev->md.uuid_lock);
3763 goto nla_put_failure;
3765 if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) ||
3766 nla_put_u64_0pad(skb, T_bits_total, drbd_bm_bits(device)) ||
3767 nla_put_u64_0pad(skb, T_bits_oos,
3768 drbd_bm_total_weight(device)))
3769 goto nla_put_failure;
3770 if (C_SYNC_SOURCE <= device->state.conn &&
3771 C_PAUSED_SYNC_T >= device->state.conn) {
3772 if (nla_put_u64_0pad(skb, T_bits_rs_total,
3773 device->rs_total) ||
3774 nla_put_u64_0pad(skb, T_bits_rs_failed,
3776 goto nla_put_failure;
3781 switch(sib->sib_reason) {
3782 case SIB_SYNC_PROGRESS:
3783 case SIB_GET_STATUS_REPLY:
3785 case SIB_STATE_CHANGE:
3786 if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
3787 nla_put_u32(skb, T_new_state, sib->ns.i))
3788 goto nla_put_failure;
3790 case SIB_HELPER_POST:
3791 if (nla_put_u32(skb, T_helper_exit_code,
3792 sib->helper_exit_code))
3793 goto nla_put_failure;
3795 case SIB_HELPER_PRE:
3796 if (nla_put_string(skb, T_helper, sib->helper_name))
3797 goto nla_put_failure;
3801 nla_nest_end(skb, nla);
3811 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3813 struct drbd_config_context adm_ctx;
3814 enum drbd_ret_code retcode;
3817 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3818 if (!adm_ctx.reply_skb)
3820 if (retcode != NO_ERROR)
3823 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL);
3825 nlmsg_free(adm_ctx.reply_skb);
3829 drbd_adm_finish(&adm_ctx, info, retcode);
3833 static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3835 struct drbd_device *device;
3836 struct drbd_genlmsghdr *dh;
3837 struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3838 struct drbd_resource *resource = NULL;
3839 struct drbd_resource *tmp;
3840 unsigned volume = cb->args[1];
3842 /* Open coded, deferred, iteration:
3843 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3844 * connection = "first connection of resource or undefined";
3845 * idr_for_each_entry(&resource->devices, device, i) {
3849 * where resource is cb->args[0];
3850 * and i is cb->args[1];
3852 * cb->args[2] indicates if we shall loop over all resources,
3853 * or just dump all volumes of a single resource.
3855 * This may miss entries inserted after this dump started,
3856 * or entries deleted before they are reached.
3858 * We need to make sure the device won't disappear while
3859 * we are looking at it, and revalidate our iterators
3860 * on each iteration.
3863 /* synchronize with conn_create()/drbd_destroy_connection() */
3865 /* revalidate iterator position */
3866 for_each_resource_rcu(tmp, &drbd_resources) {
3868 /* first iteration */
3880 device = idr_get_next(&resource->devices, &volume);
3882 /* No more volumes to dump on this resource.
3883 * Advance resource iterator. */
3884 pos = list_entry_rcu(resource->resources.next,
3885 struct drbd_resource, resources);
3886 /* Did we dump any volume of this resource yet? */
3888 /* If we reached the end of the list,
3889 * or only a single resource dump was requested,
3891 if (&pos->resources == &drbd_resources || cb->args[2])
3899 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3900 cb->nlh->nlmsg_seq, &drbd_genl_family,
3901 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
3906 /* This is a connection without a single volume.
3907 * Suprisingly enough, it may have a network
3909 struct drbd_connection *connection;
3912 dh->ret_code = NO_ERROR;
3913 connection = the_only_connection(resource);
3914 if (nla_put_drbd_cfg_context(skb, resource, connection, NULL))
3917 struct net_conf *nc;
3919 nc = rcu_dereference(connection->net_conf);
3920 if (nc && net_conf_to_skb(skb, nc, 1) != 0)
3926 D_ASSERT(device, device->vnr == volume);
3927 D_ASSERT(device, device->resource == resource);
3929 dh->minor = device_to_minor(device);
3930 dh->ret_code = NO_ERROR;
3932 if (nla_put_status_info(skb, device, NULL)) {
3934 genlmsg_cancel(skb, dh);
3938 genlmsg_end(skb, dh);
3943 /* where to start the next iteration */
3944 cb->args[0] = (long)pos;
3945 cb->args[1] = (pos == resource) ? volume + 1 : 0;
3947 /* No more resources/volumes/minors found results in an empty skb.
3948 * Which will terminate the dump. */
3953 * Request status of all resources, or of all volumes within a single resource.
3955 * This is a dump, as the answer may not fit in a single reply skb otherwise.
3956 * Which means we cannot use the family->attrbuf or other such members, because
3957 * dump is NOT protected by the genl_lock(). During dump, we only have access
3958 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3960 * Once things are setup properly, we call into get_one_status().
3962 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
3964 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3966 const char *resource_name;
3967 struct drbd_resource *resource;
3970 /* Is this a followup call? */
3972 /* ... of a single resource dump,
3973 * and the resource iterator has been advanced already? */
3974 if (cb->args[2] && cb->args[2] != cb->args[0])
3975 return 0; /* DONE. */
3979 /* First call (from netlink_dump_start). We need to figure out
3980 * which resource(s) the user wants us to dump. */
3981 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
3982 nlmsg_attrlen(cb->nlh, hdrlen),
3983 DRBD_NLA_CFG_CONTEXT);
3985 /* No explicit context given. Dump all. */
3988 maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3989 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
3991 return PTR_ERR(nla);
3992 /* context given, but no name present? */
3995 resource_name = nla_data(nla);
3996 if (!*resource_name)
3998 resource = drbd_find_resource(resource_name);
4002 kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */
4004 /* prime iterators, and set "filter" mode mark:
4005 * only dump this connection. */
4006 cb->args[0] = (long)resource;
4007 /* cb->args[1] = 0; passed in this way. */
4008 cb->args[2] = (long)resource;
4011 return get_one_status(skb, cb);
4014 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
4016 struct drbd_config_context adm_ctx;
4017 enum drbd_ret_code retcode;
4018 struct timeout_parms tp;
4021 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4022 if (!adm_ctx.reply_skb)
4024 if (retcode != NO_ERROR)
4028 adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
4029 test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED :
4032 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
4034 nlmsg_free(adm_ctx.reply_skb);
4038 drbd_adm_finish(&adm_ctx, info, retcode);
4042 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
4044 struct drbd_config_context adm_ctx;
4045 struct drbd_device *device;
4046 enum drbd_ret_code retcode;
4047 struct start_ov_parms parms;
4049 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4050 if (!adm_ctx.reply_skb)
4052 if (retcode != NO_ERROR)
4055 device = adm_ctx.device;
4057 /* resume from last known position, if possible */
4058 parms.ov_start_sector = device->ov_start_sector;
4059 parms.ov_stop_sector = ULLONG_MAX;
4060 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
4061 int err = start_ov_parms_from_attrs(&parms, info);
4063 retcode = ERR_MANDATORY_TAG;
4064 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4068 mutex_lock(&adm_ctx.resource->adm_mutex);
4070 /* w_make_ov_request expects position to be aligned */
4071 device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
4072 device->ov_stop_sector = parms.ov_stop_sector;
4074 /* If there is still bitmap IO pending, e.g. previous resync or verify
4075 * just being finished, wait for it before requesting a new resync. */
4076 drbd_suspend_io(device);
4077 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
4078 retcode = drbd_request_state(device, NS(conn, C_VERIFY_S));
4079 drbd_resume_io(device);
4081 mutex_unlock(&adm_ctx.resource->adm_mutex);
4083 drbd_adm_finish(&adm_ctx, info, retcode);
4088 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
4090 struct drbd_config_context adm_ctx;
4091 struct drbd_device *device;
4092 enum drbd_ret_code retcode;
4093 int skip_initial_sync = 0;
4095 struct new_c_uuid_parms args;
4097 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4098 if (!adm_ctx.reply_skb)
4100 if (retcode != NO_ERROR)
4103 device = adm_ctx.device;
4104 memset(&args, 0, sizeof(args));
4105 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
4106 err = new_c_uuid_parms_from_attrs(&args, info);
4108 retcode = ERR_MANDATORY_TAG;
4109 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4114 mutex_lock(&adm_ctx.resource->adm_mutex);
4115 mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */
4117 if (!get_ldev(device)) {
4118 retcode = ERR_NO_DISK;
4122 /* this is "skip initial sync", assume to be clean */
4123 if (device->state.conn == C_CONNECTED &&
4124 first_peer_device(device)->connection->agreed_pro_version >= 90 &&
4125 device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
4126 drbd_info(device, "Preparing to skip initial sync\n");
4127 skip_initial_sync = 1;
4128 } else if (device->state.conn != C_STANDALONE) {
4129 retcode = ERR_CONNECTED;
4133 drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
4134 drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
4136 if (args.clear_bm) {
4137 err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
4138 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
4140 drbd_err(device, "Writing bitmap failed with %d\n", err);
4141 retcode = ERR_IO_MD_DISK;
4143 if (skip_initial_sync) {
4144 drbd_send_uuids_skip_initial_sync(first_peer_device(device));
4145 _drbd_uuid_set(device, UI_BITMAP, 0);
4146 drbd_print_uuids(device, "cleared bitmap UUID");
4147 spin_lock_irq(&device->resource->req_lock);
4148 _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
4150 spin_unlock_irq(&device->resource->req_lock);
4154 drbd_md_sync(device);
4158 mutex_unlock(device->state_mutex);
4159 mutex_unlock(&adm_ctx.resource->adm_mutex);
4161 drbd_adm_finish(&adm_ctx, info, retcode);
4165 static enum drbd_ret_code
4166 drbd_check_resource_name(struct drbd_config_context *adm_ctx)
4168 const char *name = adm_ctx->resource_name;
4169 if (!name || !name[0]) {
4170 drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing");
4171 return ERR_MANDATORY_TAG;
4173 /* if we want to use these in sysfs/configfs/debugfs some day,
4174 * we must not allow slashes */
4175 if (strchr(name, '/')) {
4176 drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name");
4177 return ERR_INVALID_REQUEST;
4182 static void resource_to_info(struct resource_info *info,
4183 struct drbd_resource *resource)
4185 info->res_role = conn_highest_role(first_connection(resource));
4186 info->res_susp = resource->susp;
4187 info->res_susp_nod = resource->susp_nod;
4188 info->res_susp_fen = resource->susp_fen;
4191 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
4193 struct drbd_connection *connection;
4194 struct drbd_config_context adm_ctx;
4195 enum drbd_ret_code retcode;
4196 struct res_opts res_opts;
4199 retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0);
4200 if (!adm_ctx.reply_skb)
4202 if (retcode != NO_ERROR)
4205 set_res_opts_defaults(&res_opts);
4206 err = res_opts_from_attrs(&res_opts, info);
4207 if (err && err != -ENOMSG) {
4208 retcode = ERR_MANDATORY_TAG;
4209 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4213 retcode = drbd_check_resource_name(&adm_ctx);
4214 if (retcode != NO_ERROR)
4217 if (adm_ctx.resource) {
4218 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
4219 retcode = ERR_INVALID_REQUEST;
4220 drbd_msg_put_info(adm_ctx.reply_skb, "resource exists");
4222 /* else: still NO_ERROR */
4226 /* not yet safe for genl_family.parallel_ops */
4227 mutex_lock(&resources_mutex);
4228 connection = conn_create(adm_ctx.resource_name, &res_opts);
4229 mutex_unlock(&resources_mutex);
4232 struct resource_info resource_info;
4234 mutex_lock(¬ification_mutex);
4235 resource_to_info(&resource_info, connection->resource);
4236 notify_resource_state(NULL, 0, connection->resource,
4237 &resource_info, NOTIFY_CREATE);
4238 mutex_unlock(¬ification_mutex);
4240 retcode = ERR_NOMEM;
4243 drbd_adm_finish(&adm_ctx, info, retcode);
4247 static void device_to_info(struct device_info *info,
4248 struct drbd_device *device)
4250 info->dev_disk_state = device->state.disk;
4254 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info)
4256 struct drbd_config_context adm_ctx;
4257 struct drbd_genlmsghdr *dh = info->userhdr;
4258 enum drbd_ret_code retcode;
4260 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4261 if (!adm_ctx.reply_skb)
4263 if (retcode != NO_ERROR)
4266 if (dh->minor > MINORMASK) {
4267 drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range");
4268 retcode = ERR_INVALID_REQUEST;
4271 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
4272 drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range");
4273 retcode = ERR_INVALID_REQUEST;
4277 /* drbd_adm_prepare made sure already
4278 * that first_peer_device(device)->connection and device->vnr match the request. */
4279 if (adm_ctx.device) {
4280 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
4281 retcode = ERR_MINOR_OR_VOLUME_EXISTS;
4282 /* else: still NO_ERROR */
4286 mutex_lock(&adm_ctx.resource->adm_mutex);
4287 retcode = drbd_create_device(&adm_ctx, dh->minor);
4288 if (retcode == NO_ERROR) {
4289 struct drbd_device *device;
4290 struct drbd_peer_device *peer_device;
4291 struct device_info info;
4292 unsigned int peer_devices = 0;
4293 enum drbd_notification_type flags;
4295 device = minor_to_device(dh->minor);
4296 for_each_peer_device(peer_device, device) {
4297 if (!has_net_conf(peer_device->connection))
4302 device_to_info(&info, device);
4303 mutex_lock(¬ification_mutex);
4304 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4305 notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags);
4306 for_each_peer_device(peer_device, device) {
4307 struct peer_device_info peer_device_info;
4309 if (!has_net_conf(peer_device->connection))
4311 peer_device_to_info(&peer_device_info, peer_device);
4312 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4313 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info,
4314 NOTIFY_CREATE | flags);
4316 mutex_unlock(¬ification_mutex);
4318 mutex_unlock(&adm_ctx.resource->adm_mutex);
4320 drbd_adm_finish(&adm_ctx, info, retcode);
4324 static enum drbd_ret_code adm_del_minor(struct drbd_device *device)
4326 struct drbd_peer_device *peer_device;
4328 if (device->state.disk == D_DISKLESS &&
4329 /* no need to be device->state.conn == C_STANDALONE &&
4330 * we may want to delete a minor from a live replication group.
4332 device->state.role == R_SECONDARY) {
4333 struct drbd_connection *connection =
4334 first_connection(device->resource);
4336 _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS),
4337 CS_VERBOSE + CS_WAIT_COMPLETE);
4339 /* If the state engine hasn't stopped the sender thread yet, we
4340 * need to flush the sender work queue before generating the
4341 * DESTROY events here. */
4342 if (get_t_state(&connection->worker) == RUNNING)
4343 drbd_flush_workqueue(&connection->sender_work);
4345 mutex_lock(¬ification_mutex);
4346 for_each_peer_device(peer_device, device) {
4347 if (!has_net_conf(peer_device->connection))
4349 notify_peer_device_state(NULL, 0, peer_device, NULL,
4350 NOTIFY_DESTROY | NOTIFY_CONTINUES);
4352 notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY);
4353 mutex_unlock(¬ification_mutex);
4355 drbd_delete_device(device);
4358 return ERR_MINOR_CONFIGURED;
4361 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info)
4363 struct drbd_config_context adm_ctx;
4364 enum drbd_ret_code retcode;
4366 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4367 if (!adm_ctx.reply_skb)
4369 if (retcode != NO_ERROR)
4372 mutex_lock(&adm_ctx.resource->adm_mutex);
4373 retcode = adm_del_minor(adm_ctx.device);
4374 mutex_unlock(&adm_ctx.resource->adm_mutex);
4376 drbd_adm_finish(&adm_ctx, info, retcode);
4380 static int adm_del_resource(struct drbd_resource *resource)
4382 struct drbd_connection *connection;
4384 for_each_connection(connection, resource) {
4385 if (connection->cstate > C_STANDALONE)
4386 return ERR_NET_CONFIGURED;
4388 if (!idr_is_empty(&resource->devices))
4389 return ERR_RES_IN_USE;
4391 /* The state engine has stopped the sender thread, so we don't
4392 * need to flush the sender work queue before generating the
4393 * DESTROY event here. */
4394 mutex_lock(¬ification_mutex);
4395 notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY);
4396 mutex_unlock(¬ification_mutex);
4398 mutex_lock(&resources_mutex);
4399 list_del_rcu(&resource->resources);
4400 mutex_unlock(&resources_mutex);
4401 /* Make sure all threads have actually stopped: state handling only
4402 * does drbd_thread_stop_nowait(). */
4403 list_for_each_entry(connection, &resource->connections, connections)
4404 drbd_thread_stop(&connection->worker);
4406 drbd_free_resource(resource);
4410 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
4412 struct drbd_config_context adm_ctx;
4413 struct drbd_resource *resource;
4414 struct drbd_connection *connection;
4415 struct drbd_device *device;
4416 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
4419 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4420 if (!adm_ctx.reply_skb)
4422 if (retcode != NO_ERROR)
4425 resource = adm_ctx.resource;
4426 mutex_lock(&resource->adm_mutex);
4428 for_each_connection(connection, resource) {
4429 struct drbd_peer_device *peer_device;
4431 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
4432 retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0);
4433 if (retcode < SS_SUCCESS) {
4434 drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote");
4439 retcode = conn_try_disconnect(connection, 0);
4440 if (retcode < SS_SUCCESS) {
4441 drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect");
4447 idr_for_each_entry(&resource->devices, device, i) {
4448 retcode = adm_detach(device, 0);
4449 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
4450 drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach");
4455 /* delete volumes */
4456 idr_for_each_entry(&resource->devices, device, i) {
4457 retcode = adm_del_minor(device);
4458 if (retcode != NO_ERROR) {
4459 /* "can not happen" */
4460 drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume");
4465 retcode = adm_del_resource(resource);
4467 mutex_unlock(&resource->adm_mutex);
4469 drbd_adm_finish(&adm_ctx, info, retcode);
4473 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
4475 struct drbd_config_context adm_ctx;
4476 struct drbd_resource *resource;
4477 enum drbd_ret_code retcode;
4479 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4480 if (!adm_ctx.reply_skb)
4482 if (retcode != NO_ERROR)
4484 resource = adm_ctx.resource;
4486 mutex_lock(&resource->adm_mutex);
4487 retcode = adm_del_resource(resource);
4488 mutex_unlock(&resource->adm_mutex);
4490 drbd_adm_finish(&adm_ctx, info, retcode);
4494 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib)
4496 struct sk_buff *msg;
4497 struct drbd_genlmsghdr *d_out;
4501 seq = atomic_inc_return(&drbd_genl_seq);
4502 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4507 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
4508 if (!d_out) /* cannot happen, but anyways. */
4509 goto nla_put_failure;
4510 d_out->minor = device_to_minor(device);
4511 d_out->ret_code = NO_ERROR;
4513 if (nla_put_status_info(msg, device, sib))
4514 goto nla_put_failure;
4515 genlmsg_end(msg, d_out);
4516 err = drbd_genl_multicast_events(msg, GFP_NOWAIT);
4517 /* msg has been consumed or freed in netlink_broadcast() */
4518 if (err && err != -ESRCH)
4526 drbd_err(device, "Error %d while broadcasting event. "
4527 "Event seq:%u sib_reason:%u\n",
4528 err, seq, sib->sib_reason);
4531 static int nla_put_notification_header(struct sk_buff *msg,
4532 enum drbd_notification_type type)
4534 struct drbd_notification_header nh = {
4538 return drbd_notification_header_to_skb(msg, &nh, true);
4541 int notify_resource_state(struct sk_buff *skb,
4543 struct drbd_resource *resource,
4544 struct resource_info *resource_info,
4545 enum drbd_notification_type type)
4547 struct resource_statistics resource_statistics;
4548 struct drbd_genlmsghdr *dh;
4549 bool multicast = false;
4553 seq = atomic_inc_return(¬ify_genl_seq);
4554 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4562 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_RESOURCE_STATE);
4564 goto nla_put_failure;
4566 dh->ret_code = NO_ERROR;
4567 if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) ||
4568 nla_put_notification_header(skb, type) ||
4569 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4570 resource_info_to_skb(skb, resource_info, true)))
4571 goto nla_put_failure;
4572 resource_statistics.res_stat_write_ordering = resource->write_ordering;
4573 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
4575 goto nla_put_failure;
4576 genlmsg_end(skb, dh);
4578 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4579 /* skb has been consumed or freed in netlink_broadcast() */
4580 if (err && err != -ESRCH)
4588 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4593 int notify_device_state(struct sk_buff *skb,
4595 struct drbd_device *device,
4596 struct device_info *device_info,
4597 enum drbd_notification_type type)
4599 struct device_statistics device_statistics;
4600 struct drbd_genlmsghdr *dh;
4601 bool multicast = false;
4605 seq = atomic_inc_return(¬ify_genl_seq);
4606 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4614 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_DEVICE_STATE);
4616 goto nla_put_failure;
4617 dh->minor = device->minor;
4618 dh->ret_code = NO_ERROR;
4619 if (nla_put_drbd_cfg_context(skb, device->resource, NULL, device) ||
4620 nla_put_notification_header(skb, type) ||
4621 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4622 device_info_to_skb(skb, device_info, true)))
4623 goto nla_put_failure;
4624 device_to_statistics(&device_statistics, device);
4625 device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
4626 genlmsg_end(skb, dh);
4628 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4629 /* skb has been consumed or freed in netlink_broadcast() */
4630 if (err && err != -ESRCH)
4638 drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n",
4643 int notify_connection_state(struct sk_buff *skb,
4645 struct drbd_connection *connection,
4646 struct connection_info *connection_info,
4647 enum drbd_notification_type type)
4649 struct connection_statistics connection_statistics;
4650 struct drbd_genlmsghdr *dh;
4651 bool multicast = false;
4655 seq = atomic_inc_return(¬ify_genl_seq);
4656 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4664 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_CONNECTION_STATE);
4666 goto nla_put_failure;
4668 dh->ret_code = NO_ERROR;
4669 if (nla_put_drbd_cfg_context(skb, connection->resource, connection, NULL) ||
4670 nla_put_notification_header(skb, type) ||
4671 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4672 connection_info_to_skb(skb, connection_info, true)))
4673 goto nla_put_failure;
4674 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
4675 connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
4676 genlmsg_end(skb, dh);
4678 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4679 /* skb has been consumed or freed in netlink_broadcast() */
4680 if (err && err != -ESRCH)
4688 drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n",
4693 int notify_peer_device_state(struct sk_buff *skb,
4695 struct drbd_peer_device *peer_device,
4696 struct peer_device_info *peer_device_info,
4697 enum drbd_notification_type type)
4699 struct peer_device_statistics peer_device_statistics;
4700 struct drbd_resource *resource = peer_device->device->resource;
4701 struct drbd_genlmsghdr *dh;
4702 bool multicast = false;
4706 seq = atomic_inc_return(¬ify_genl_seq);
4707 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4715 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_PEER_DEVICE_STATE);
4717 goto nla_put_failure;
4719 dh->ret_code = NO_ERROR;
4720 if (nla_put_drbd_cfg_context(skb, resource, peer_device->connection, peer_device->device) ||
4721 nla_put_notification_header(skb, type) ||
4722 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4723 peer_device_info_to_skb(skb, peer_device_info, true)))
4724 goto nla_put_failure;
4725 peer_device_to_statistics(&peer_device_statistics, peer_device);
4726 peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
4727 genlmsg_end(skb, dh);
4729 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4730 /* skb has been consumed or freed in netlink_broadcast() */
4731 if (err && err != -ESRCH)
4739 drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n",
4744 void notify_helper(enum drbd_notification_type type,
4745 struct drbd_device *device, struct drbd_connection *connection,
4746 const char *name, int status)
4748 struct drbd_resource *resource = device ? device->resource : connection->resource;
4749 struct drbd_helper_info helper_info;
4750 unsigned int seq = atomic_inc_return(¬ify_genl_seq);
4751 struct sk_buff *skb = NULL;
4752 struct drbd_genlmsghdr *dh;
4755 strlcpy(helper_info.helper_name, name, sizeof(helper_info.helper_name));
4756 helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name));
4757 helper_info.helper_status = status;
4759 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4765 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_HELPER);
4768 dh->minor = device ? device->minor : -1;
4769 dh->ret_code = NO_ERROR;
4770 mutex_lock(¬ification_mutex);
4771 if (nla_put_drbd_cfg_context(skb, resource, connection, device) ||
4772 nla_put_notification_header(skb, type) ||
4773 drbd_helper_info_to_skb(skb, &helper_info, true))
4775 genlmsg_end(skb, dh);
4776 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4778 /* skb has been consumed or freed in netlink_broadcast() */
4779 if (err && err != -ESRCH)
4781 mutex_unlock(¬ification_mutex);
4785 mutex_unlock(¬ification_mutex);
4788 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4792 static int notify_initial_state_done(struct sk_buff *skb, unsigned int seq)
4794 struct drbd_genlmsghdr *dh;
4798 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_INITIAL_STATE_DONE);
4800 goto nla_put_failure;
4802 dh->ret_code = NO_ERROR;
4803 if (nla_put_notification_header(skb, NOTIFY_EXISTS))
4804 goto nla_put_failure;
4805 genlmsg_end(skb, dh);
4810 pr_err("Error %d sending event. Event seq:%u\n", err, seq);
4814 static void free_state_changes(struct list_head *list)
4816 while (!list_empty(list)) {
4817 struct drbd_state_change *state_change =
4818 list_first_entry(list, struct drbd_state_change, list);
4819 list_del(&state_change->list);
4820 forget_state_change(state_change);
4824 static unsigned int notifications_for_state_change(struct drbd_state_change *state_change)
4827 state_change->n_connections +
4828 state_change->n_devices +
4829 state_change->n_devices * state_change->n_connections;
4832 static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4834 struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0];
4835 unsigned int seq = cb->args[2];
4837 enum drbd_notification_type flags = 0;
4840 /* There is no need for taking notification_mutex here: it doesn't
4841 matter if the initial state events mix with later state chage
4842 events; we can always tell the events apart by the NOTIFY_EXISTS
4846 if (cb->args[5] == 1) {
4847 err = notify_initial_state_done(skb, seq);
4851 if (cb->args[4] < cb->args[3])
4852 flags |= NOTIFY_CONTINUES;
4854 err = notify_resource_state_change(skb, seq, state_change->resource,
4855 NOTIFY_EXISTS | flags);
4859 if (n < state_change->n_connections) {
4860 err = notify_connection_state_change(skb, seq, &state_change->connections[n],
4861 NOTIFY_EXISTS | flags);
4864 n -= state_change->n_connections;
4865 if (n < state_change->n_devices) {
4866 err = notify_device_state_change(skb, seq, &state_change->devices[n],
4867 NOTIFY_EXISTS | flags);
4870 n -= state_change->n_devices;
4871 if (n < state_change->n_devices * state_change->n_connections) {
4872 err = notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n],
4873 NOTIFY_EXISTS | flags);
4878 if (cb->args[4] == cb->args[3]) {
4879 struct drbd_state_change *next_state_change =
4880 list_entry(state_change->list.next,
4881 struct drbd_state_change, list);
4882 cb->args[0] = (long)next_state_change;
4883 cb->args[3] = notifications_for_state_change(next_state_change);
4893 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4895 struct drbd_resource *resource;
4898 if (cb->args[5] >= 1) {
4899 if (cb->args[5] > 1)
4900 return get_initial_state(skb, cb);
4902 struct drbd_state_change *state_change =
4903 (struct drbd_state_change *)cb->args[0];
4905 /* connect list to head */
4906 list_add(&head, &state_change->list);
4907 free_state_changes(&head);
4912 cb->args[5] = 2; /* number of iterations */
4913 mutex_lock(&resources_mutex);
4914 for_each_resource(resource, &drbd_resources) {
4915 struct drbd_state_change *state_change;
4917 state_change = remember_old_state(resource, GFP_KERNEL);
4918 if (!state_change) {
4919 if (!list_empty(&head))
4920 free_state_changes(&head);
4921 mutex_unlock(&resources_mutex);
4924 copy_old_to_new_state_change(state_change);
4925 list_add_tail(&state_change->list, &head);
4926 cb->args[5] += notifications_for_state_change(state_change);
4928 mutex_unlock(&resources_mutex);
4930 if (!list_empty(&head)) {
4931 struct drbd_state_change *state_change =
4932 list_entry(head.next, struct drbd_state_change, list);
4933 cb->args[0] = (long)state_change;
4934 cb->args[3] = notifications_for_state_change(state_change);
4935 list_del(&head); /* detach list from head */
4938 cb->args[2] = cb->nlh->nlmsg_seq;
4939 return get_initial_state(skb, cb);