4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10 drbd is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
15 drbd is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with drbd; see the file COPYING. If not, write to
22 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
26 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
28 #include <linux/module.h>
29 #include <linux/drbd.h>
32 #include <linux/file.h>
33 #include <linux/slab.h>
34 #include <linux/blkpg.h>
35 #include <linux/cpumask.h>
37 #include "drbd_protocol.h"
39 #include "drbd_state_change.h"
40 #include <asm/unaligned.h>
41 #include <linux/drbd_limits.h>
42 #include <linux/kthread.h>
44 #include <net/genetlink.h>
47 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
48 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
50 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info);
51 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info);
53 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info);
54 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info);
55 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
57 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
58 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
59 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
60 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
61 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
62 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
63 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
64 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
65 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
66 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
67 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
68 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
69 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
70 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
71 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
72 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
73 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
74 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
75 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
76 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
78 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
79 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb);
80 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb);
81 int drbd_adm_dump_devices_done(struct netlink_callback *cb);
82 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb);
83 int drbd_adm_dump_connections_done(struct netlink_callback *cb);
84 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb);
85 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb);
86 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb);
88 #include <linux/drbd_genl_api.h>
90 #include <linux/genl_magic_func.h>
92 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
93 static atomic_t notify_genl_seq = ATOMIC_INIT(2); /* two. */
95 DEFINE_MUTEX(notification_mutex);
97 /* used blkdev_get_by_path, to claim our meta data device(s) */
98 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
100 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
102 genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
103 if (genlmsg_reply(skb, info))
104 pr_err("error sending genl reply\n");
107 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
108 * reason it could fail was no space in skb, and there are 4k available. */
109 static int drbd_msg_put_info(struct sk_buff *skb, const char *info)
114 if (!info || !info[0])
117 nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
121 err = nla_put_string(skb, T_info_text, info);
123 nla_nest_cancel(skb, nla);
126 nla_nest_end(skb, nla);
130 /* This would be a good candidate for a "pre_doit" hook,
131 * and per-family private info->pointers.
132 * But we need to stay compatible with older kernels.
133 * If it returns successfully, adm_ctx members are valid.
135 * At this point, we still rely on the global genl_lock().
136 * If we want to avoid that, and allow "genl_family.parallel_ops", we may need
137 * to add additional synchronization against object destruction/modification.
139 #define DRBD_ADM_NEED_MINOR 1
140 #define DRBD_ADM_NEED_RESOURCE 2
141 #define DRBD_ADM_NEED_CONNECTION 4
142 static int drbd_adm_prepare(struct drbd_config_context *adm_ctx,
143 struct sk_buff *skb, struct genl_info *info, unsigned flags)
145 struct drbd_genlmsghdr *d_in = info->userhdr;
146 const u8 cmd = info->genlhdr->cmd;
149 memset(adm_ctx, 0, sizeof(*adm_ctx));
151 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
152 if (cmd != DRBD_ADM_GET_STATUS && !capable(CAP_NET_ADMIN))
155 adm_ctx->reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
156 if (!adm_ctx->reply_skb) {
161 adm_ctx->reply_dh = genlmsg_put_reply(adm_ctx->reply_skb,
162 info, &drbd_genl_family, 0, cmd);
163 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
165 if (!adm_ctx->reply_dh) {
170 adm_ctx->reply_dh->minor = d_in->minor;
171 adm_ctx->reply_dh->ret_code = NO_ERROR;
173 adm_ctx->volume = VOLUME_UNSPECIFIED;
174 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
176 /* parse and validate only */
177 err = drbd_cfg_context_from_attrs(NULL, info);
181 /* It was present, and valid,
182 * copy it over to the reply skb. */
183 err = nla_put_nohdr(adm_ctx->reply_skb,
184 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
185 info->attrs[DRBD_NLA_CFG_CONTEXT]);
189 /* and assign stuff to the adm_ctx */
190 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
192 adm_ctx->volume = nla_get_u32(nla);
193 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
195 adm_ctx->resource_name = nla_data(nla);
196 adm_ctx->my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)];
197 adm_ctx->peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)];
198 if ((adm_ctx->my_addr &&
199 nla_len(adm_ctx->my_addr) > sizeof(adm_ctx->connection->my_addr)) ||
200 (adm_ctx->peer_addr &&
201 nla_len(adm_ctx->peer_addr) > sizeof(adm_ctx->connection->peer_addr))) {
207 adm_ctx->minor = d_in->minor;
208 adm_ctx->device = minor_to_device(d_in->minor);
210 /* We are protected by the global genl_lock().
211 * But we may explicitly drop it/retake it in drbd_adm_set_role(),
212 * so make sure this object stays around. */
214 kref_get(&adm_ctx->device->kref);
216 if (adm_ctx->resource_name) {
217 adm_ctx->resource = drbd_find_resource(adm_ctx->resource_name);
220 if (!adm_ctx->device && (flags & DRBD_ADM_NEED_MINOR)) {
221 drbd_msg_put_info(adm_ctx->reply_skb, "unknown minor");
222 return ERR_MINOR_INVALID;
224 if (!adm_ctx->resource && (flags & DRBD_ADM_NEED_RESOURCE)) {
225 drbd_msg_put_info(adm_ctx->reply_skb, "unknown resource");
226 if (adm_ctx->resource_name)
227 return ERR_RES_NOT_KNOWN;
228 return ERR_INVALID_REQUEST;
231 if (flags & DRBD_ADM_NEED_CONNECTION) {
232 if (adm_ctx->resource) {
233 drbd_msg_put_info(adm_ctx->reply_skb, "no resource name expected");
234 return ERR_INVALID_REQUEST;
236 if (adm_ctx->device) {
237 drbd_msg_put_info(adm_ctx->reply_skb, "no minor number expected");
238 return ERR_INVALID_REQUEST;
240 if (adm_ctx->my_addr && adm_ctx->peer_addr)
241 adm_ctx->connection = conn_get_by_addrs(nla_data(adm_ctx->my_addr),
242 nla_len(adm_ctx->my_addr),
243 nla_data(adm_ctx->peer_addr),
244 nla_len(adm_ctx->peer_addr));
245 if (!adm_ctx->connection) {
246 drbd_msg_put_info(adm_ctx->reply_skb, "unknown connection");
247 return ERR_INVALID_REQUEST;
251 /* some more paranoia, if the request was over-determined */
252 if (adm_ctx->device && adm_ctx->resource &&
253 adm_ctx->device->resource != adm_ctx->resource) {
254 pr_warning("request: minor=%u, resource=%s; but that minor belongs to resource %s\n",
255 adm_ctx->minor, adm_ctx->resource->name,
256 adm_ctx->device->resource->name);
257 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists in different resource");
258 return ERR_INVALID_REQUEST;
260 if (adm_ctx->device &&
261 adm_ctx->volume != VOLUME_UNSPECIFIED &&
262 adm_ctx->volume != adm_ctx->device->vnr) {
263 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
264 adm_ctx->minor, adm_ctx->volume,
265 adm_ctx->device->vnr,
266 adm_ctx->device->resource->name);
267 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists as different volume");
268 return ERR_INVALID_REQUEST;
271 /* still, provide adm_ctx->resource always, if possible. */
272 if (!adm_ctx->resource) {
273 adm_ctx->resource = adm_ctx->device ? adm_ctx->device->resource
274 : adm_ctx->connection ? adm_ctx->connection->resource : NULL;
275 if (adm_ctx->resource)
276 kref_get(&adm_ctx->resource->kref);
282 nlmsg_free(adm_ctx->reply_skb);
283 adm_ctx->reply_skb = NULL;
287 static int drbd_adm_finish(struct drbd_config_context *adm_ctx,
288 struct genl_info *info, int retcode)
290 if (adm_ctx->device) {
291 kref_put(&adm_ctx->device->kref, drbd_destroy_device);
292 adm_ctx->device = NULL;
294 if (adm_ctx->connection) {
295 kref_put(&adm_ctx->connection->kref, &drbd_destroy_connection);
296 adm_ctx->connection = NULL;
298 if (adm_ctx->resource) {
299 kref_put(&adm_ctx->resource->kref, drbd_destroy_resource);
300 adm_ctx->resource = NULL;
303 if (!adm_ctx->reply_skb)
306 adm_ctx->reply_dh->ret_code = retcode;
307 drbd_adm_send_reply(adm_ctx->reply_skb, info);
311 static void setup_khelper_env(struct drbd_connection *connection, char **envp)
315 /* FIXME: A future version will not allow this case. */
316 if (connection->my_addr_len == 0 || connection->peer_addr_len == 0)
319 switch (((struct sockaddr *)&connection->peer_addr)->sa_family) {
322 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
323 &((struct sockaddr_in6 *)&connection->peer_addr)->sin6_addr);
327 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
328 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
332 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
333 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
335 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
338 int drbd_khelper(struct drbd_device *device, char *cmd)
340 char *envp[] = { "HOME=/",
342 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
343 (char[20]) { }, /* address family */
344 (char[60]) { }, /* address */
347 char *argv[] = {drbd_usermode_helper, cmd, mb, NULL };
348 struct drbd_connection *connection = first_peer_device(device)->connection;
352 if (current == connection->worker.task)
353 set_bit(CALLBACK_PENDING, &connection->flags);
355 snprintf(mb, 14, "minor-%d", device_to_minor(device));
356 setup_khelper_env(connection, envp);
358 /* The helper may take some time.
359 * write out any unsynced meta data changes now */
360 drbd_md_sync(device);
362 drbd_info(device, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, mb);
363 sib.sib_reason = SIB_HELPER_PRE;
364 sib.helper_name = cmd;
365 drbd_bcast_event(device, &sib);
366 notify_helper(NOTIFY_CALL, device, connection, cmd, 0);
367 ret = call_usermodehelper(drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
369 drbd_warn(device, "helper command: %s %s %s exit code %u (0x%x)\n",
370 drbd_usermode_helper, cmd, mb,
371 (ret >> 8) & 0xff, ret);
373 drbd_info(device, "helper command: %s %s %s exit code %u (0x%x)\n",
374 drbd_usermode_helper, cmd, mb,
375 (ret >> 8) & 0xff, ret);
376 sib.sib_reason = SIB_HELPER_POST;
377 sib.helper_exit_code = ret;
378 drbd_bcast_event(device, &sib);
379 notify_helper(NOTIFY_RESPONSE, device, connection, cmd, ret);
381 if (current == connection->worker.task)
382 clear_bit(CALLBACK_PENDING, &connection->flags);
384 if (ret < 0) /* Ignore any ERRNOs we got. */
390 enum drbd_peer_state conn_khelper(struct drbd_connection *connection, char *cmd)
392 char *envp[] = { "HOME=/",
394 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
395 (char[20]) { }, /* address family */
396 (char[60]) { }, /* address */
398 char *resource_name = connection->resource->name;
399 char *argv[] = {drbd_usermode_helper, cmd, resource_name, NULL };
402 setup_khelper_env(connection, envp);
403 conn_md_sync(connection);
405 drbd_info(connection, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, resource_name);
406 /* TODO: conn_bcast_event() ?? */
407 notify_helper(NOTIFY_CALL, NULL, connection, cmd, 0);
409 ret = call_usermodehelper(drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
411 drbd_warn(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
412 drbd_usermode_helper, cmd, resource_name,
413 (ret >> 8) & 0xff, ret);
415 drbd_info(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
416 drbd_usermode_helper, cmd, resource_name,
417 (ret >> 8) & 0xff, ret);
418 /* TODO: conn_bcast_event() ?? */
419 notify_helper(NOTIFY_RESPONSE, NULL, connection, cmd, ret);
421 if (ret < 0) /* Ignore any ERRNOs we got. */
427 static enum drbd_fencing_p highest_fencing_policy(struct drbd_connection *connection)
429 enum drbd_fencing_p fp = FP_NOT_AVAIL;
430 struct drbd_peer_device *peer_device;
434 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
435 struct drbd_device *device = peer_device->device;
436 if (get_ldev_if_state(device, D_CONSISTENT)) {
437 struct disk_conf *disk_conf =
438 rcu_dereference(peer_device->device->ldev->disk_conf);
439 fp = max_t(enum drbd_fencing_p, fp, disk_conf->fencing);
448 static bool resource_is_supended(struct drbd_resource *resource)
450 return resource->susp || resource->susp_fen || resource->susp_nod;
453 bool conn_try_outdate_peer(struct drbd_connection *connection)
455 struct drbd_resource * const resource = connection->resource;
456 unsigned int connect_cnt;
457 union drbd_state mask = { };
458 union drbd_state val = { };
459 enum drbd_fencing_p fp;
463 spin_lock_irq(&resource->req_lock);
464 if (connection->cstate >= C_WF_REPORT_PARAMS) {
465 drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n");
466 spin_unlock_irq(&resource->req_lock);
470 connect_cnt = connection->connect_cnt;
471 spin_unlock_irq(&resource->req_lock);
473 fp = highest_fencing_policy(connection);
476 drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n");
477 spin_lock_irq(&resource->req_lock);
478 if (connection->cstate < C_WF_REPORT_PARAMS) {
479 _conn_request_state(connection,
480 (union drbd_state) { { .susp_fen = 1 } },
481 (union drbd_state) { { .susp_fen = 0 } },
482 CS_VERBOSE | CS_HARD | CS_DC_SUSP);
483 /* We are no longer suspended due to the fencing policy.
484 * We may still be suspended due to the on-no-data-accessible policy.
485 * If that was OND_IO_ERROR, fail pending requests. */
486 if (!resource_is_supended(resource))
487 _tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
489 /* Else: in case we raced with a connection handshake,
490 * let the handshake figure out if we maybe can RESEND,
491 * and do not resume/fail pending requests here.
492 * Worst case is we stay suspended for now, which may be
493 * resolved by either re-establishing the replication link, or
494 * the next link failure, or eventually the administrator. */
495 spin_unlock_irq(&resource->req_lock);
503 r = conn_khelper(connection, "fence-peer");
505 switch ((r>>8) & 0xff) {
506 case P_INCONSISTENT: /* peer is inconsistent */
507 ex_to_string = "peer is inconsistent or worse";
509 val.pdsk = D_INCONSISTENT;
511 case P_OUTDATED: /* peer got outdated, or was already outdated */
512 ex_to_string = "peer was fenced";
514 val.pdsk = D_OUTDATED;
516 case P_DOWN: /* peer was down */
517 if (conn_highest_disk(connection) == D_UP_TO_DATE) {
518 /* we will(have) create(d) a new UUID anyways... */
519 ex_to_string = "peer is unreachable, assumed to be dead";
521 val.pdsk = D_OUTDATED;
523 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
526 case P_PRIMARY: /* Peer is primary, voluntarily outdate myself.
527 * This is useful when an unconnected R_SECONDARY is asked to
528 * become R_PRIMARY, but finds the other peer being active. */
529 ex_to_string = "peer is active";
530 drbd_warn(connection, "Peer is primary, outdating myself.\n");
532 val.disk = D_OUTDATED;
535 /* THINK: do we need to handle this
536 * like case 4, or more like case 5? */
537 if (fp != FP_STONITH)
538 drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n");
539 ex_to_string = "peer was stonithed";
541 val.pdsk = D_OUTDATED;
544 /* The script is broken ... */
545 drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
546 return false; /* Eventually leave IO frozen */
549 drbd_info(connection, "fence-peer helper returned %d (%s)\n",
550 (r>>8) & 0xff, ex_to_string);
553 conn_request_state(connection, mask, val, CS_VERBOSE);
554 here, because we might were able to re-establish the connection in the
556 spin_lock_irq(&resource->req_lock);
557 if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) {
558 if (connection->connect_cnt != connect_cnt)
559 /* In case the connection was established and droped
560 while the fence-peer handler was running, ignore it */
561 drbd_info(connection, "Ignoring fence-peer exit code\n");
563 _conn_request_state(connection, mask, val, CS_VERBOSE);
565 spin_unlock_irq(&resource->req_lock);
567 return conn_highest_pdsk(connection) <= D_OUTDATED;
570 static int _try_outdate_peer_async(void *data)
572 struct drbd_connection *connection = (struct drbd_connection *)data;
574 conn_try_outdate_peer(connection);
576 kref_put(&connection->kref, drbd_destroy_connection);
580 void conn_try_outdate_peer_async(struct drbd_connection *connection)
582 struct task_struct *opa;
584 kref_get(&connection->kref);
585 /* We may just have force_sig()'ed this thread
586 * to get it out of some blocking network function.
587 * Clear signals; otherwise kthread_run(), which internally uses
588 * wait_on_completion_killable(), will mistake our pending signal
589 * for a new fatal signal and fail. */
590 flush_signals(current);
591 opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h");
593 drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n");
594 kref_put(&connection->kref, drbd_destroy_connection);
599 drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force)
601 struct drbd_peer_device *const peer_device = first_peer_device(device);
602 struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL;
603 const int max_tries = 4;
604 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
608 union drbd_state mask, val;
610 if (new_role == R_PRIMARY) {
611 struct drbd_connection *connection;
613 /* Detect dead peers as soon as possible. */
616 for_each_connection(connection, device->resource)
617 request_ping(connection);
621 mutex_lock(device->state_mutex);
623 mask.i = 0; mask.role = R_MASK;
624 val.i = 0; val.role = new_role;
626 while (try++ < max_tries) {
627 rv = _drbd_request_state_holding_state_mutex(device, mask, val, CS_WAIT_COMPLETE);
629 /* in case we first succeeded to outdate,
630 * but now suddenly could establish a connection */
631 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
637 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
638 (device->state.disk < D_UP_TO_DATE &&
639 device->state.disk >= D_INCONSISTENT)) {
641 val.disk = D_UP_TO_DATE;
646 if (rv == SS_NO_UP_TO_DATE_DISK &&
647 device->state.disk == D_CONSISTENT && mask.pdsk == 0) {
648 D_ASSERT(device, device->state.pdsk == D_UNKNOWN);
650 if (conn_try_outdate_peer(connection)) {
651 val.disk = D_UP_TO_DATE;
657 if (rv == SS_NOTHING_TO_DO)
659 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
660 if (!conn_try_outdate_peer(connection) && force) {
661 drbd_warn(device, "Forced into split brain situation!\n");
663 val.pdsk = D_OUTDATED;
668 if (rv == SS_TWO_PRIMARIES) {
669 /* Maybe the peer is detected as dead very soon...
670 retry at most once more in this case. */
673 nc = rcu_dereference(connection->net_conf);
674 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
676 schedule_timeout_interruptible(timeo);
681 if (rv < SS_SUCCESS) {
682 rv = _drbd_request_state(device, mask, val,
683 CS_VERBOSE + CS_WAIT_COMPLETE);
694 drbd_warn(device, "Forced to consider local data as UpToDate!\n");
696 /* Wait until nothing is on the fly :) */
697 wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0);
699 /* FIXME also wait for all pending P_BARRIER_ACK? */
701 if (new_role == R_SECONDARY) {
702 if (get_ldev(device)) {
703 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
707 mutex_lock(&device->resource->conf_update);
708 nc = connection->net_conf;
710 nc->discard_my_data = 0; /* without copy; single bit op is atomic */
711 mutex_unlock(&device->resource->conf_update);
713 if (get_ldev(device)) {
714 if (((device->state.conn < C_CONNECTED ||
715 device->state.pdsk <= D_FAILED)
716 && device->ldev->md.uuid[UI_BITMAP] == 0) || forced)
717 drbd_uuid_new_current(device);
719 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
724 /* writeout of activity log covered areas of the bitmap
725 * to stable storage done in after state change already */
727 if (device->state.conn >= C_WF_REPORT_PARAMS) {
728 /* if this was forced, we should consider sync */
730 drbd_send_uuids(peer_device);
731 drbd_send_current_state(peer_device);
734 drbd_md_sync(device);
735 set_disk_ro(device->vdisk, new_role == R_SECONDARY);
736 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
738 mutex_unlock(device->state_mutex);
742 static const char *from_attrs_err_to_txt(int err)
744 return err == -ENOMSG ? "required attribute missing" :
745 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
746 err == -EEXIST ? "can not change invariant setting" :
747 "invalid attribute value";
750 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
752 struct drbd_config_context adm_ctx;
753 struct set_role_parms parms;
755 enum drbd_ret_code retcode;
757 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
758 if (!adm_ctx.reply_skb)
760 if (retcode != NO_ERROR)
763 memset(&parms, 0, sizeof(parms));
764 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
765 err = set_role_parms_from_attrs(&parms, info);
767 retcode = ERR_MANDATORY_TAG;
768 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
773 mutex_lock(&adm_ctx.resource->adm_mutex);
775 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
776 retcode = drbd_set_role(adm_ctx.device, R_PRIMARY, parms.assume_uptodate);
778 retcode = drbd_set_role(adm_ctx.device, R_SECONDARY, 0);
780 mutex_unlock(&adm_ctx.resource->adm_mutex);
783 drbd_adm_finish(&adm_ctx, info, retcode);
787 /* Initializes the md.*_offset members, so we are able to find
788 * the on disk meta data.
790 * We currently have two possible layouts:
792 * |----------- md_size_sect ------------------|
793 * [ 4k superblock ][ activity log ][ Bitmap ]
795 * | bm_offset = al_offset + X |
796 * ==> bitmap sectors = md_size_sect - bm_offset
799 * |----------- md_size_sect ------------------|
800 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
802 * | bm_offset = al_offset - Y |
803 * ==> bitmap sectors = Y = al_offset - bm_offset
805 * Activity log size used to be fixed 32kB,
806 * but is about to become configurable.
808 static void drbd_md_set_sector_offsets(struct drbd_device *device,
809 struct drbd_backing_dev *bdev)
811 sector_t md_size_sect = 0;
812 unsigned int al_size_sect = bdev->md.al_size_4k * 8;
814 bdev->md.md_offset = drbd_md_ss(bdev);
816 switch (bdev->md.meta_dev_idx) {
818 /* v07 style fixed size indexed meta data */
819 bdev->md.md_size_sect = MD_128MB_SECT;
820 bdev->md.al_offset = MD_4kB_SECT;
821 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
823 case DRBD_MD_INDEX_FLEX_EXT:
824 /* just occupy the full device; unit: sectors */
825 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
826 bdev->md.al_offset = MD_4kB_SECT;
827 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
829 case DRBD_MD_INDEX_INTERNAL:
830 case DRBD_MD_INDEX_FLEX_INT:
831 /* al size is still fixed */
832 bdev->md.al_offset = -al_size_sect;
833 /* we need (slightly less than) ~ this much bitmap sectors: */
834 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
835 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
836 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
837 md_size_sect = ALIGN(md_size_sect, 8);
839 /* plus the "drbd meta data super block",
840 * and the activity log; */
841 md_size_sect += MD_4kB_SECT + al_size_sect;
843 bdev->md.md_size_sect = md_size_sect;
844 /* bitmap offset is adjusted by 'super' block size */
845 bdev->md.bm_offset = -md_size_sect + MD_4kB_SECT;
850 /* input size is expected to be in KB */
851 char *ppsize(char *buf, unsigned long long size)
853 /* Needs 9 bytes at max including trailing NUL:
854 * -1ULL ==> "16384 EB" */
855 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
857 while (size >= 10000 && base < sizeof(units)-1) {
859 size = (size >> 10) + !!(size & (1<<9));
862 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
867 /* there is still a theoretical deadlock when called from receiver
868 * on an D_INCONSISTENT R_PRIMARY:
869 * remote READ does inc_ap_bio, receiver would need to receive answer
870 * packet from remote to dec_ap_bio again.
871 * receiver receive_sizes(), comes here,
872 * waits for ap_bio_cnt == 0. -> deadlock.
873 * but this cannot happen, actually, because:
874 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
875 * (not connected, or bad/no disk on peer):
876 * see drbd_fail_request_early, ap_bio_cnt is zero.
877 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
878 * peer may not initiate a resize.
880 /* Note these are not to be confused with
881 * drbd_adm_suspend_io/drbd_adm_resume_io,
882 * which are (sub) state changes triggered by admin (drbdsetup),
883 * and can be long lived.
884 * This changes an device->flag, is triggered by drbd internals,
885 * and should be short-lived. */
886 /* It needs to be a counter, since multiple threads might
887 independently suspend and resume IO. */
888 void drbd_suspend_io(struct drbd_device *device)
890 atomic_inc(&device->suspend_cnt);
891 if (drbd_suspended(device))
893 wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt));
896 void drbd_resume_io(struct drbd_device *device)
898 if (atomic_dec_and_test(&device->suspend_cnt))
899 wake_up(&device->misc_wait);
903 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
904 * @device: DRBD device.
906 * Returns 0 on success, negative return values indicate errors.
907 * You should call drbd_md_sync() after calling this function.
909 enum determine_dev_size
910 drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local)
912 struct md_offsets_and_sizes {
913 u64 last_agreed_sect;
920 u32 al_stripe_size_4k;
922 sector_t u_size, size;
923 struct drbd_md *md = &device->ldev->md;
927 int md_moved, la_size_changed;
928 enum determine_dev_size rv = DS_UNCHANGED;
930 /* We may change the on-disk offsets of our meta data below. Lock out
931 * anything that may cause meta data IO, to avoid acting on incomplete
932 * layout changes or scribbling over meta data that is in the process
935 * Move is not exactly correct, btw, currently we have all our meta
936 * data in core memory, to "move" it we just write it all out, there
938 drbd_suspend_io(device);
939 buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */
941 drbd_resume_io(device);
945 /* remember current offset and sizes */
946 prev.last_agreed_sect = md->la_size_sect;
947 prev.md_offset = md->md_offset;
948 prev.al_offset = md->al_offset;
949 prev.bm_offset = md->bm_offset;
950 prev.md_size_sect = md->md_size_sect;
951 prev.al_stripes = md->al_stripes;
952 prev.al_stripe_size_4k = md->al_stripe_size_4k;
955 /* rs is non NULL if we should change the AL layout only */
956 md->al_stripes = rs->al_stripes;
957 md->al_stripe_size_4k = rs->al_stripe_size / 4;
958 md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4;
961 drbd_md_set_sector_offsets(device, device->ldev);
964 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
966 size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED);
968 if (size < prev.last_agreed_sect) {
969 if (rs && u_size == 0) {
970 /* Remove "rs &&" later. This check should always be active, but
971 right now the receiver expects the permissive behavior */
972 drbd_warn(device, "Implicit shrink not allowed. "
973 "Use --size=%llus for explicit shrink.\n",
974 (unsigned long long)size);
975 rv = DS_ERROR_SHRINK;
978 rv = DS_ERROR_SPACE_MD;
979 if (rv != DS_UNCHANGED)
983 if (drbd_get_capacity(device->this_bdev) != size ||
984 drbd_bm_capacity(device) != size) {
986 err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC));
988 /* currently there is only one error: ENOMEM! */
989 size = drbd_bm_capacity(device);
991 drbd_err(device, "OUT OF MEMORY! "
992 "Could not allocate bitmap!\n");
994 drbd_err(device, "BM resizing failed. "
995 "Leaving size unchanged\n");
999 /* racy, see comments above. */
1000 drbd_set_my_capacity(device, size);
1001 md->la_size_sect = size;
1002 drbd_info(device, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
1003 (unsigned long long)size>>1);
1008 la_size_changed = (prev.last_agreed_sect != md->la_size_sect);
1010 md_moved = prev.md_offset != md->md_offset
1011 || prev.md_size_sect != md->md_size_sect;
1013 if (la_size_changed || md_moved || rs) {
1016 /* We do some synchronous IO below, which may take some time.
1017 * Clear the timer, to avoid scary "timer expired!" messages,
1018 * "Superblock" is written out at least twice below, anyways. */
1019 del_timer(&device->md_sync_timer);
1021 /* We won't change the "al-extents" setting, we just may need
1022 * to move the on-disk location of the activity log ringbuffer.
1023 * Lock for transaction is good enough, it may well be "dirty"
1024 * or even "starving". */
1025 wait_event(device->al_wait, lc_try_lock_for_transaction(device->act_log));
1027 /* mark current on-disk bitmap and activity log as unreliable */
1028 prev_flags = md->flags;
1029 md->flags |= MDF_FULL_SYNC | MDF_AL_DISABLED;
1030 drbd_md_write(device, buffer);
1032 drbd_al_initialize(device, buffer);
1034 drbd_info(device, "Writing the whole bitmap, %s\n",
1035 la_size_changed && md_moved ? "size changed and md moved" :
1036 la_size_changed ? "size changed" : "md moved");
1037 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
1038 drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
1039 "size changed", BM_LOCKED_MASK);
1041 /* on-disk bitmap and activity log is authoritative again
1042 * (unless there was an IO error meanwhile...) */
1043 md->flags = prev_flags;
1044 drbd_md_write(device, buffer);
1047 drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
1048 md->al_stripes, md->al_stripe_size_4k * 4);
1051 if (size > prev.last_agreed_sect)
1052 rv = prev.last_agreed_sect ? DS_GREW : DS_GREW_FROM_ZERO;
1053 if (size < prev.last_agreed_sect)
1058 /* restore previous offset and sizes */
1059 md->la_size_sect = prev.last_agreed_sect;
1060 md->md_offset = prev.md_offset;
1061 md->al_offset = prev.al_offset;
1062 md->bm_offset = prev.bm_offset;
1063 md->md_size_sect = prev.md_size_sect;
1064 md->al_stripes = prev.al_stripes;
1065 md->al_stripe_size_4k = prev.al_stripe_size_4k;
1066 md->al_size_4k = (u64)prev.al_stripes * prev.al_stripe_size_4k;
1068 lc_unlock(device->act_log);
1069 wake_up(&device->al_wait);
1070 drbd_md_put_buffer(device);
1071 drbd_resume_io(device);
1077 drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev,
1078 sector_t u_size, int assume_peer_has_space)
1080 sector_t p_size = device->p_size; /* partner's disk size. */
1081 sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
1082 sector_t m_size; /* my size */
1085 m_size = drbd_get_max_capacity(bdev);
1087 if (device->state.conn < C_CONNECTED && assume_peer_has_space) {
1088 drbd_warn(device, "Resize while not connected was forced by the user!\n");
1092 if (p_size && m_size) {
1093 size = min_t(sector_t, p_size, m_size);
1096 size = la_size_sect;
1097 if (m_size && m_size < size)
1099 if (p_size && p_size < size)
1110 drbd_err(device, "Both nodes diskless!\n");
1114 drbd_err(device, "Requested disk size is too big (%lu > %lu)\n",
1115 (unsigned long)u_size>>1, (unsigned long)size>>1);
1124 * drbd_check_al_size() - Ensures that the AL is of the right size
1125 * @device: DRBD device.
1127 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1128 * failed, and 0 on success. You should call drbd_md_sync() after you called
1131 static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1133 struct lru_cache *n, *t;
1134 struct lc_element *e;
1135 unsigned int in_use;
1138 if (device->act_log &&
1139 device->act_log->nr_elements == dc->al_extents)
1143 t = device->act_log;
1144 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
1145 dc->al_extents, sizeof(struct lc_element), 0);
1148 drbd_err(device, "Cannot allocate act_log lru!\n");
1151 spin_lock_irq(&device->al_lock);
1153 for (i = 0; i < t->nr_elements; i++) {
1154 e = lc_element_by_index(t, i);
1156 drbd_err(device, "refcnt(%d)==%d\n",
1157 e->lc_number, e->refcnt);
1158 in_use += e->refcnt;
1162 device->act_log = n;
1163 spin_unlock_irq(&device->al_lock);
1165 drbd_err(device, "Activity log still in use!\n");
1171 drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */
1175 static void blk_queue_discard_granularity(struct request_queue *q, unsigned int granularity)
1177 q->limits.discard_granularity = granularity;
1180 static unsigned int drbd_max_discard_sectors(struct drbd_connection *connection)
1182 /* when we introduced REQ_WRITE_SAME support, we also bumped
1183 * our maximum supported batch bio size used for discards. */
1184 if (connection->agreed_features & DRBD_FF_WSAME)
1185 return DRBD_MAX_BBIO_SECTORS;
1186 /* before, with DRBD <= 8.4.6, we only allowed up to one AL_EXTENT_SIZE. */
1187 return AL_EXTENT_SIZE >> 9;
1190 static void decide_on_discard_support(struct drbd_device *device,
1191 struct request_queue *q,
1192 struct request_queue *b,
1193 bool discard_zeroes_if_aligned)
1195 /* q = drbd device queue (device->rq_queue)
1196 * b = backing device queue (device->ldev->backing_bdev->bd_disk->queue),
1197 * or NULL if diskless
1199 struct drbd_connection *connection = first_peer_device(device)->connection;
1200 bool can_do = b ? blk_queue_discard(b) : true;
1202 if (can_do && connection->cstate >= C_CONNECTED && !(connection->agreed_features & DRBD_FF_TRIM)) {
1204 drbd_info(connection, "peer DRBD too old, does not support TRIM: disabling discards\n");
1207 /* We don't care for the granularity, really.
1208 * Stacking limits below should fix it for the local
1209 * device. Whether or not it is a suitable granularity
1210 * on the remote device is not our problem, really. If
1211 * you care, you need to use devices with similar
1212 * topology on all peers. */
1213 blk_queue_discard_granularity(q, 512);
1214 q->limits.max_discard_sectors = drbd_max_discard_sectors(connection);
1215 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1216 q->limits.max_write_zeroes_sectors = drbd_max_discard_sectors(connection);
1218 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1219 blk_queue_discard_granularity(q, 0);
1220 q->limits.max_discard_sectors = 0;
1221 q->limits.max_write_zeroes_sectors = 0;
1225 static void fixup_discard_if_not_supported(struct request_queue *q)
1227 /* To avoid confusion, if this queue does not support discard, clear
1228 * max_discard_sectors, which is what lsblk -D reports to the user.
1229 * Older kernels got this wrong in "stack limits".
1231 if (!blk_queue_discard(q)) {
1232 blk_queue_max_discard_sectors(q, 0);
1233 blk_queue_discard_granularity(q, 0);
1237 static void decide_on_write_same_support(struct drbd_device *device,
1238 struct request_queue *q,
1239 struct request_queue *b, struct o_qlim *o,
1240 bool disable_write_same)
1242 struct drbd_peer_device *peer_device = first_peer_device(device);
1243 struct drbd_connection *connection = peer_device->connection;
1244 bool can_do = b ? b->limits.max_write_same_sectors : true;
1246 if (can_do && disable_write_same) {
1248 drbd_info(peer_device, "WRITE_SAME disabled by config\n");
1251 if (can_do && connection->cstate >= C_CONNECTED && !(connection->agreed_features & DRBD_FF_WSAME)) {
1253 drbd_info(peer_device, "peer does not support WRITE_SAME\n");
1257 /* logical block size; queue_logical_block_size(NULL) is 512 */
1258 unsigned int peer_lbs = be32_to_cpu(o->logical_block_size);
1259 unsigned int me_lbs_b = queue_logical_block_size(b);
1260 unsigned int me_lbs = queue_logical_block_size(q);
1262 if (me_lbs_b != me_lbs) {
1264 "logical block size of local backend does not match (drbd:%u, backend:%u); was this a late attach?\n",
1266 /* rather disable write same than trigger some BUG_ON later in the scsi layer. */
1269 if (me_lbs_b != peer_lbs) {
1270 drbd_warn(peer_device, "logical block sizes do not match (me:%u, peer:%u); this may cause problems.\n",
1273 drbd_dbg(peer_device, "logical block size mismatch: WRITE_SAME disabled.\n");
1276 me_lbs = max(me_lbs, me_lbs_b);
1277 /* We cannot change the logical block size of an in-use queue.
1278 * We can only hope that access happens to be properly aligned.
1279 * If not, the peer will likely produce an IO error, and detach. */
1280 if (peer_lbs > me_lbs) {
1281 if (device->state.role != R_PRIMARY) {
1282 blk_queue_logical_block_size(q, peer_lbs);
1283 drbd_warn(peer_device, "logical block size set to %u\n", peer_lbs);
1285 drbd_warn(peer_device,
1286 "current Primary must NOT adjust logical block size (%u -> %u); hope for the best.\n",
1291 if (can_do && !o->write_same_capable) {
1292 /* If we introduce an open-coded write-same loop on the receiving side,
1293 * the peer would present itself as "capable". */
1294 drbd_dbg(peer_device, "WRITE_SAME disabled (peer device not capable)\n");
1299 blk_queue_max_write_same_sectors(q, can_do ? DRBD_MAX_BBIO_SECTORS : 0);
1302 static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev,
1303 unsigned int max_bio_size, struct o_qlim *o)
1305 struct request_queue * const q = device->rq_queue;
1306 unsigned int max_hw_sectors = max_bio_size >> 9;
1307 unsigned int max_segments = 0;
1308 struct request_queue *b = NULL;
1309 struct disk_conf *dc;
1310 bool discard_zeroes_if_aligned = true;
1311 bool disable_write_same = false;
1314 b = bdev->backing_bdev->bd_disk->queue;
1316 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1318 dc = rcu_dereference(device->ldev->disk_conf);
1319 max_segments = dc->max_bio_bvecs;
1320 discard_zeroes_if_aligned = dc->discard_zeroes_if_aligned;
1321 disable_write_same = dc->disable_write_same;
1324 blk_set_stacking_limits(&q->limits);
1327 blk_queue_max_hw_sectors(q, max_hw_sectors);
1328 /* This is the workaround for "bio would need to, but cannot, be split" */
1329 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1330 blk_queue_segment_boundary(q, PAGE_SIZE-1);
1331 decide_on_discard_support(device, q, b, discard_zeroes_if_aligned);
1332 decide_on_write_same_support(device, q, b, o, disable_write_same);
1335 blk_queue_stack_limits(q, b);
1337 if (q->backing_dev_info->ra_pages !=
1338 b->backing_dev_info->ra_pages) {
1339 drbd_info(device, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1340 q->backing_dev_info->ra_pages,
1341 b->backing_dev_info->ra_pages);
1342 q->backing_dev_info->ra_pages =
1343 b->backing_dev_info->ra_pages;
1346 fixup_discard_if_not_supported(q);
1349 void drbd_reconsider_queue_parameters(struct drbd_device *device, struct drbd_backing_dev *bdev, struct o_qlim *o)
1351 unsigned int now, new, local, peer;
1353 now = queue_max_hw_sectors(device->rq_queue) << 9;
1354 local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */
1355 peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */
1358 local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9;
1359 device->local_max_bio_size = local;
1361 local = min(local, DRBD_MAX_BIO_SIZE);
1363 /* We may ignore peer limits if the peer is modern enough.
1364 Because new from 8.3.8 onwards the peer can use multiple
1365 BIOs for a single peer_request */
1366 if (device->state.conn >= C_WF_REPORT_PARAMS) {
1367 if (first_peer_device(device)->connection->agreed_pro_version < 94)
1368 peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1369 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1370 else if (first_peer_device(device)->connection->agreed_pro_version == 94)
1371 peer = DRBD_MAX_SIZE_H80_PACKET;
1372 else if (first_peer_device(device)->connection->agreed_pro_version < 100)
1373 peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */
1375 peer = DRBD_MAX_BIO_SIZE;
1377 /* We may later detach and re-attach on a disconnected Primary.
1378 * Avoid this setting to jump back in that case.
1379 * We want to store what we know the peer DRBD can handle,
1380 * not what the peer IO backend can handle. */
1381 if (peer > device->peer_max_bio_size)
1382 device->peer_max_bio_size = peer;
1384 new = min(local, peer);
1386 if (device->state.role == R_PRIMARY && new < now)
1387 drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1390 drbd_info(device, "max BIO size = %u\n", new);
1392 drbd_setup_queue_param(device, bdev, new, o);
1395 /* Starts the worker thread */
1396 static void conn_reconfig_start(struct drbd_connection *connection)
1398 drbd_thread_start(&connection->worker);
1399 drbd_flush_workqueue(&connection->sender_work);
1402 /* if still unconfigured, stops worker again. */
1403 static void conn_reconfig_done(struct drbd_connection *connection)
1406 spin_lock_irq(&connection->resource->req_lock);
1407 stop_threads = conn_all_vols_unconf(connection) &&
1408 connection->cstate == C_STANDALONE;
1409 spin_unlock_irq(&connection->resource->req_lock);
1411 /* ack_receiver thread and ack_sender workqueue are implicitly
1412 * stopped by receiver in conn_disconnect() */
1413 drbd_thread_stop(&connection->receiver);
1414 drbd_thread_stop(&connection->worker);
1418 /* Make sure IO is suspended before calling this function(). */
1419 static void drbd_suspend_al(struct drbd_device *device)
1423 if (!lc_try_lock(device->act_log)) {
1424 drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1428 drbd_al_shrink(device);
1429 spin_lock_irq(&device->resource->req_lock);
1430 if (device->state.conn < C_CONNECTED)
1431 s = !test_and_set_bit(AL_SUSPENDED, &device->flags);
1432 spin_unlock_irq(&device->resource->req_lock);
1433 lc_unlock(device->act_log);
1436 drbd_info(device, "Suspended AL updates\n");
1440 static bool should_set_defaults(struct genl_info *info)
1442 unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1443 return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1446 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1448 /* This is limited by 16 bit "slot" numbers,
1449 * and by available on-disk context storage.
1451 * Also (u16)~0 is special (denotes a "free" extent).
1453 * One transaction occupies one 4kB on-disk block,
1454 * we have n such blocks in the on disk ring buffer,
1455 * the "current" transaction may fail (n-1),
1456 * and there is 919 slot numbers context information per transaction.
1458 * 72 transaction blocks amounts to more than 2**16 context slots,
1459 * so cap there first.
1461 const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1462 const unsigned int sufficient_on_disk =
1463 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1464 /AL_CONTEXT_PER_TRANSACTION;
1466 unsigned int al_size_4k = bdev->md.al_size_4k;
1468 if (al_size_4k > sufficient_on_disk)
1471 return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1474 static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1476 return a->disk_barrier != b->disk_barrier ||
1477 a->disk_flushes != b->disk_flushes ||
1478 a->disk_drain != b->disk_drain;
1481 static void sanitize_disk_conf(struct drbd_device *device, struct disk_conf *disk_conf,
1482 struct drbd_backing_dev *nbc)
1484 struct request_queue * const q = nbc->backing_bdev->bd_disk->queue;
1486 if (disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1487 disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1488 if (disk_conf->al_extents > drbd_al_extents_max(nbc))
1489 disk_conf->al_extents = drbd_al_extents_max(nbc);
1491 if (!blk_queue_discard(q)) {
1492 if (disk_conf->rs_discard_granularity) {
1493 disk_conf->rs_discard_granularity = 0; /* disable feature */
1494 drbd_info(device, "rs_discard_granularity feature disabled\n");
1498 if (disk_conf->rs_discard_granularity) {
1499 int orig_value = disk_conf->rs_discard_granularity;
1502 if (q->limits.discard_granularity > disk_conf->rs_discard_granularity)
1503 disk_conf->rs_discard_granularity = q->limits.discard_granularity;
1505 remainder = disk_conf->rs_discard_granularity % q->limits.discard_granularity;
1506 disk_conf->rs_discard_granularity += remainder;
1508 if (disk_conf->rs_discard_granularity > q->limits.max_discard_sectors << 9)
1509 disk_conf->rs_discard_granularity = q->limits.max_discard_sectors << 9;
1511 if (disk_conf->rs_discard_granularity != orig_value)
1512 drbd_info(device, "rs_discard_granularity changed to %d\n",
1513 disk_conf->rs_discard_granularity);
1517 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1519 struct drbd_config_context adm_ctx;
1520 enum drbd_ret_code retcode;
1521 struct drbd_device *device;
1522 struct disk_conf *new_disk_conf, *old_disk_conf;
1523 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1526 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1527 if (!adm_ctx.reply_skb)
1529 if (retcode != NO_ERROR)
1532 device = adm_ctx.device;
1533 mutex_lock(&adm_ctx.resource->adm_mutex);
1535 /* we also need a disk
1536 * to change the options on */
1537 if (!get_ldev(device)) {
1538 retcode = ERR_NO_DISK;
1542 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1543 if (!new_disk_conf) {
1544 retcode = ERR_NOMEM;
1548 mutex_lock(&device->resource->conf_update);
1549 old_disk_conf = device->ldev->disk_conf;
1550 *new_disk_conf = *old_disk_conf;
1551 if (should_set_defaults(info))
1552 set_disk_conf_defaults(new_disk_conf);
1554 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1555 if (err && err != -ENOMSG) {
1556 retcode = ERR_MANDATORY_TAG;
1557 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1561 if (!expect(new_disk_conf->resync_rate >= 1))
1562 new_disk_conf->resync_rate = 1;
1564 sanitize_disk_conf(device, new_disk_conf, device->ldev);
1566 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1567 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1569 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1570 if (fifo_size != device->rs_plan_s->size) {
1571 new_plan = fifo_alloc(fifo_size);
1573 drbd_err(device, "kmalloc of fifo_buffer failed");
1574 retcode = ERR_NOMEM;
1579 drbd_suspend_io(device);
1580 wait_event(device->al_wait, lc_try_lock(device->act_log));
1581 drbd_al_shrink(device);
1582 err = drbd_check_al_size(device, new_disk_conf);
1583 lc_unlock(device->act_log);
1584 wake_up(&device->al_wait);
1585 drbd_resume_io(device);
1588 retcode = ERR_NOMEM;
1592 lock_all_resources();
1593 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1594 if (retcode == NO_ERROR) {
1595 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1596 drbd_resync_after_changed(device);
1598 unlock_all_resources();
1600 if (retcode != NO_ERROR)
1604 old_plan = device->rs_plan_s;
1605 rcu_assign_pointer(device->rs_plan_s, new_plan);
1608 mutex_unlock(&device->resource->conf_update);
1610 if (new_disk_conf->al_updates)
1611 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1613 device->ldev->md.flags |= MDF_AL_DISABLED;
1615 if (new_disk_conf->md_flushes)
1616 clear_bit(MD_NO_FUA, &device->flags);
1618 set_bit(MD_NO_FUA, &device->flags);
1620 if (write_ordering_changed(old_disk_conf, new_disk_conf))
1621 drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH);
1623 if (old_disk_conf->discard_zeroes_if_aligned != new_disk_conf->discard_zeroes_if_aligned
1624 || old_disk_conf->disable_write_same != new_disk_conf->disable_write_same)
1625 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
1627 drbd_md_sync(device);
1629 if (device->state.conn >= C_CONNECTED) {
1630 struct drbd_peer_device *peer_device;
1632 for_each_peer_device(peer_device, device)
1633 drbd_send_sync_param(peer_device);
1637 kfree(old_disk_conf);
1639 mod_timer(&device->request_timer, jiffies + HZ);
1643 mutex_unlock(&device->resource->conf_update);
1645 kfree(new_disk_conf);
1650 mutex_unlock(&adm_ctx.resource->adm_mutex);
1652 drbd_adm_finish(&adm_ctx, info, retcode);
1656 static struct block_device *open_backing_dev(struct drbd_device *device,
1657 const char *bdev_path, void *claim_ptr, bool do_bd_link)
1659 struct block_device *bdev;
1662 bdev = blkdev_get_by_path(bdev_path,
1663 FMODE_READ | FMODE_WRITE | FMODE_EXCL, claim_ptr);
1665 drbd_err(device, "open(\"%s\") failed with %ld\n",
1666 bdev_path, PTR_ERR(bdev));
1673 err = bd_link_disk_holder(bdev, device->vdisk);
1675 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1676 drbd_err(device, "bd_link_disk_holder(\"%s\", ...) failed with %d\n",
1678 bdev = ERR_PTR(err);
1683 static int open_backing_devices(struct drbd_device *device,
1684 struct disk_conf *new_disk_conf,
1685 struct drbd_backing_dev *nbc)
1687 struct block_device *bdev;
1689 bdev = open_backing_dev(device, new_disk_conf->backing_dev, device, true);
1691 return ERR_OPEN_DISK;
1692 nbc->backing_bdev = bdev;
1695 * meta_dev_idx >= 0: external fixed size, possibly multiple
1696 * drbd sharing one meta device. TODO in that case, paranoia
1697 * check that [md_bdev, meta_dev_idx] is not yet used by some
1698 * other drbd minor! (if you use drbd.conf + drbdadm, that
1699 * should check it for you already; but if you don't, or
1700 * someone fooled it, we need to double check here)
1702 bdev = open_backing_dev(device, new_disk_conf->meta_dev,
1703 /* claim ptr: device, if claimed exclusively; shared drbd_m_holder,
1704 * if potentially shared with other drbd minors */
1705 (new_disk_conf->meta_dev_idx < 0) ? (void*)device : (void*)drbd_m_holder,
1706 /* avoid double bd_claim_by_disk() for the same (source,target) tuple,
1707 * as would happen with internal metadata. */
1708 (new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_FLEX_INT &&
1709 new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_INTERNAL));
1711 return ERR_OPEN_MD_DISK;
1712 nbc->md_bdev = bdev;
1716 static void close_backing_dev(struct drbd_device *device, struct block_device *bdev,
1722 bd_unlink_disk_holder(bdev, device->vdisk);
1723 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1726 void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev)
1731 close_backing_dev(device, ldev->md_bdev, ldev->md_bdev != ldev->backing_bdev);
1732 close_backing_dev(device, ldev->backing_bdev, true);
1734 kfree(ldev->disk_conf);
1738 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1740 struct drbd_config_context adm_ctx;
1741 struct drbd_device *device;
1742 struct drbd_peer_device *peer_device;
1743 struct drbd_connection *connection;
1745 enum drbd_ret_code retcode;
1746 enum determine_dev_size dd;
1747 sector_t max_possible_sectors;
1748 sector_t min_md_device_sectors;
1749 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1750 struct disk_conf *new_disk_conf = NULL;
1751 struct lru_cache *resync_lru = NULL;
1752 struct fifo_buffer *new_plan = NULL;
1753 union drbd_state ns, os;
1754 enum drbd_state_rv rv;
1755 struct net_conf *nc;
1757 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1758 if (!adm_ctx.reply_skb)
1760 if (retcode != NO_ERROR)
1763 device = adm_ctx.device;
1764 mutex_lock(&adm_ctx.resource->adm_mutex);
1765 peer_device = first_peer_device(device);
1766 connection = peer_device->connection;
1767 conn_reconfig_start(connection);
1769 /* if you want to reconfigure, please tear down first */
1770 if (device->state.disk > D_DISKLESS) {
1771 retcode = ERR_DISK_CONFIGURED;
1774 /* It may just now have detached because of IO error. Make sure
1775 * drbd_ldev_destroy is done already, we may end up here very fast,
1776 * e.g. if someone calls attach from the on-io-error handler,
1777 * to realize a "hot spare" feature (not that I'd recommend that) */
1778 wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1780 /* make sure there is no leftover from previous force-detach attempts */
1781 clear_bit(FORCE_DETACH, &device->flags);
1782 clear_bit(WAS_IO_ERROR, &device->flags);
1783 clear_bit(WAS_READ_ERROR, &device->flags);
1785 /* and no leftover from previously aborted resync or verify, either */
1786 device->rs_total = 0;
1787 device->rs_failed = 0;
1788 atomic_set(&device->rs_pending_cnt, 0);
1790 /* allocation not in the IO path, drbdsetup context */
1791 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1793 retcode = ERR_NOMEM;
1796 spin_lock_init(&nbc->md.uuid_lock);
1798 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1799 if (!new_disk_conf) {
1800 retcode = ERR_NOMEM;
1803 nbc->disk_conf = new_disk_conf;
1805 set_disk_conf_defaults(new_disk_conf);
1806 err = disk_conf_from_attrs(new_disk_conf, info);
1808 retcode = ERR_MANDATORY_TAG;
1809 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1813 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1814 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1816 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1818 retcode = ERR_NOMEM;
1822 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1823 retcode = ERR_MD_IDX_INVALID;
1828 nc = rcu_dereference(connection->net_conf);
1830 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1832 retcode = ERR_STONITH_AND_PROT_A;
1838 retcode = open_backing_devices(device, new_disk_conf, nbc);
1839 if (retcode != NO_ERROR)
1842 if ((nbc->backing_bdev == nbc->md_bdev) !=
1843 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1844 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1845 retcode = ERR_MD_IDX_INVALID;
1849 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1850 1, 61, sizeof(struct bm_extent),
1851 offsetof(struct bm_extent, lce));
1853 retcode = ERR_NOMEM;
1857 /* Read our meta data super block early.
1858 * This also sets other on-disk offsets. */
1859 retcode = drbd_md_read(device, nbc);
1860 if (retcode != NO_ERROR)
1863 sanitize_disk_conf(device, new_disk_conf, nbc);
1865 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1866 drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1867 (unsigned long long) drbd_get_max_capacity(nbc),
1868 (unsigned long long) new_disk_conf->disk_size);
1869 retcode = ERR_DISK_TOO_SMALL;
1873 if (new_disk_conf->meta_dev_idx < 0) {
1874 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1875 /* at least one MB, otherwise it does not make sense */
1876 min_md_device_sectors = (2<<10);
1878 max_possible_sectors = DRBD_MAX_SECTORS;
1879 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1882 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1883 retcode = ERR_MD_DISK_TOO_SMALL;
1884 drbd_warn(device, "refusing attach: md-device too small, "
1885 "at least %llu sectors needed for this meta-disk type\n",
1886 (unsigned long long) min_md_device_sectors);
1890 /* Make sure the new disk is big enough
1891 * (we may currently be R_PRIMARY with no local disk...) */
1892 if (drbd_get_max_capacity(nbc) <
1893 drbd_get_capacity(device->this_bdev)) {
1894 retcode = ERR_DISK_TOO_SMALL;
1898 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1900 if (nbc->known_size > max_possible_sectors) {
1901 drbd_warn(device, "==> truncating very big lower level device "
1902 "to currently maximum possible %llu sectors <==\n",
1903 (unsigned long long) max_possible_sectors);
1904 if (new_disk_conf->meta_dev_idx >= 0)
1905 drbd_warn(device, "==>> using internal or flexible "
1906 "meta data may help <<==\n");
1909 drbd_suspend_io(device);
1910 /* also wait for the last barrier ack. */
1911 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1912 * We need a way to either ignore barrier acks for barriers sent before a device
1913 * was attached, or a way to wait for all pending barrier acks to come in.
1914 * As barriers are counted per resource,
1915 * we'd need to suspend io on all devices of a resource.
1917 wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1918 /* and for any other previously queued work */
1919 drbd_flush_workqueue(&connection->sender_work);
1921 rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1922 retcode = rv; /* FIXME: Type mismatch. */
1923 drbd_resume_io(device);
1924 if (rv < SS_SUCCESS)
1927 if (!get_ldev_if_state(device, D_ATTACHING))
1928 goto force_diskless;
1930 if (!device->bitmap) {
1931 if (drbd_bm_init(device)) {
1932 retcode = ERR_NOMEM;
1933 goto force_diskless_dec;
1937 if (device->state.conn < C_CONNECTED &&
1938 device->state.role == R_PRIMARY && device->ed_uuid &&
1939 (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1940 drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1941 (unsigned long long)device->ed_uuid);
1942 retcode = ERR_DATA_NOT_CURRENT;
1943 goto force_diskless_dec;
1946 /* Since we are diskless, fix the activity log first... */
1947 if (drbd_check_al_size(device, new_disk_conf)) {
1948 retcode = ERR_NOMEM;
1949 goto force_diskless_dec;
1952 /* Prevent shrinking of consistent devices ! */
1953 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1954 drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1955 drbd_warn(device, "refusing to truncate a consistent device\n");
1956 retcode = ERR_DISK_TOO_SMALL;
1957 goto force_diskless_dec;
1960 lock_all_resources();
1961 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1962 if (retcode != NO_ERROR) {
1963 unlock_all_resources();
1964 goto force_diskless_dec;
1967 /* Reset the "barriers don't work" bits here, then force meta data to
1968 * be written, to ensure we determine if barriers are supported. */
1969 if (new_disk_conf->md_flushes)
1970 clear_bit(MD_NO_FUA, &device->flags);
1972 set_bit(MD_NO_FUA, &device->flags);
1974 /* Point of no return reached.
1975 * Devices and memory are no longer released by error cleanup below.
1976 * now device takes over responsibility, and the state engine should
1977 * clean it up somewhere. */
1978 D_ASSERT(device, device->ldev == NULL);
1980 device->resync = resync_lru;
1981 device->rs_plan_s = new_plan;
1984 new_disk_conf = NULL;
1987 drbd_resync_after_changed(device);
1988 drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH);
1989 unlock_all_resources();
1991 if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
1992 set_bit(CRASHED_PRIMARY, &device->flags);
1994 clear_bit(CRASHED_PRIMARY, &device->flags);
1996 if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1997 !(device->state.role == R_PRIMARY && device->resource->susp_nod))
1998 set_bit(CRASHED_PRIMARY, &device->flags);
2000 device->send_cnt = 0;
2001 device->recv_cnt = 0;
2002 device->read_cnt = 0;
2003 device->writ_cnt = 0;
2005 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
2007 /* If I am currently not R_PRIMARY,
2008 * but meta data primary indicator is set,
2009 * I just now recover from a hard crash,
2010 * and have been R_PRIMARY before that crash.
2012 * Now, if I had no connection before that crash
2013 * (have been degraded R_PRIMARY), chances are that
2014 * I won't find my peer now either.
2016 * In that case, and _only_ in that case,
2017 * we use the degr-wfc-timeout instead of the default,
2018 * so we can automatically recover from a crash of a
2019 * degraded but active "cluster" after a certain timeout.
2021 clear_bit(USE_DEGR_WFC_T, &device->flags);
2022 if (device->state.role != R_PRIMARY &&
2023 drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
2024 !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
2025 set_bit(USE_DEGR_WFC_T, &device->flags);
2027 dd = drbd_determine_dev_size(device, 0, NULL);
2028 if (dd <= DS_ERROR) {
2029 retcode = ERR_NOMEM_BITMAP;
2030 goto force_diskless_dec;
2031 } else if (dd == DS_GREW)
2032 set_bit(RESYNC_AFTER_NEG, &device->flags);
2034 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
2035 (test_bit(CRASHED_PRIMARY, &device->flags) &&
2036 drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
2037 drbd_info(device, "Assuming that all blocks are out of sync "
2038 "(aka FullSync)\n");
2039 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2040 "set_n_write from attaching", BM_LOCKED_MASK)) {
2041 retcode = ERR_IO_MD_DISK;
2042 goto force_diskless_dec;
2045 if (drbd_bitmap_io(device, &drbd_bm_read,
2046 "read from attaching", BM_LOCKED_MASK)) {
2047 retcode = ERR_IO_MD_DISK;
2048 goto force_diskless_dec;
2052 if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
2053 drbd_suspend_al(device); /* IO is still suspended here... */
2055 spin_lock_irq(&device->resource->req_lock);
2056 os = drbd_read_state(device);
2058 /* If MDF_CONSISTENT is not set go into inconsistent state,
2059 otherwise investigate MDF_WasUpToDate...
2060 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
2061 otherwise into D_CONSISTENT state.
2063 if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
2064 if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
2065 ns.disk = D_CONSISTENT;
2067 ns.disk = D_OUTDATED;
2069 ns.disk = D_INCONSISTENT;
2072 if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
2073 ns.pdsk = D_OUTDATED;
2076 if (ns.disk == D_CONSISTENT &&
2077 (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
2078 ns.disk = D_UP_TO_DATE;
2080 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
2081 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
2082 this point, because drbd_request_state() modifies these
2085 if (rcu_dereference(device->ldev->disk_conf)->al_updates)
2086 device->ldev->md.flags &= ~MDF_AL_DISABLED;
2088 device->ldev->md.flags |= MDF_AL_DISABLED;
2092 /* In case we are C_CONNECTED postpone any decision on the new disk
2093 state after the negotiation phase. */
2094 if (device->state.conn == C_CONNECTED) {
2095 device->new_state_tmp.i = ns.i;
2097 ns.disk = D_NEGOTIATING;
2099 /* We expect to receive up-to-date UUIDs soon.
2100 To avoid a race in receive_state, free p_uuid while
2101 holding req_lock. I.e. atomic with the state change */
2102 kfree(device->p_uuid);
2103 device->p_uuid = NULL;
2106 rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
2107 spin_unlock_irq(&device->resource->req_lock);
2109 if (rv < SS_SUCCESS)
2110 goto force_diskless_dec;
2112 mod_timer(&device->request_timer, jiffies + HZ);
2114 if (device->state.role == R_PRIMARY)
2115 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
2117 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
2119 drbd_md_mark_dirty(device);
2120 drbd_md_sync(device);
2122 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
2124 conn_reconfig_done(connection);
2125 mutex_unlock(&adm_ctx.resource->adm_mutex);
2126 drbd_adm_finish(&adm_ctx, info, retcode);
2132 drbd_force_state(device, NS(disk, D_DISKLESS));
2133 drbd_md_sync(device);
2135 conn_reconfig_done(connection);
2137 close_backing_dev(device, nbc->md_bdev, nbc->md_bdev != nbc->backing_bdev);
2138 close_backing_dev(device, nbc->backing_bdev, true);
2141 kfree(new_disk_conf);
2142 lc_destroy(resync_lru);
2144 mutex_unlock(&adm_ctx.resource->adm_mutex);
2146 drbd_adm_finish(&adm_ctx, info, retcode);
2150 static int adm_detach(struct drbd_device *device, int force)
2153 set_bit(FORCE_DETACH, &device->flags);
2154 drbd_force_state(device, NS(disk, D_FAILED));
2158 return drbd_request_detach_interruptible(device);
2161 /* Detaching the disk is a process in multiple stages. First we need to lock
2162 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
2163 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
2164 * internal references as well.
2165 * Only then we have finally detached. */
2166 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
2168 struct drbd_config_context adm_ctx;
2169 enum drbd_ret_code retcode;
2170 struct detach_parms parms = { };
2173 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2174 if (!adm_ctx.reply_skb)
2176 if (retcode != NO_ERROR)
2179 if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
2180 err = detach_parms_from_attrs(&parms, info);
2182 retcode = ERR_MANDATORY_TAG;
2183 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2188 mutex_lock(&adm_ctx.resource->adm_mutex);
2189 retcode = adm_detach(adm_ctx.device, parms.force_detach);
2190 mutex_unlock(&adm_ctx.resource->adm_mutex);
2192 drbd_adm_finish(&adm_ctx, info, retcode);
2196 static bool conn_resync_running(struct drbd_connection *connection)
2198 struct drbd_peer_device *peer_device;
2203 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2204 struct drbd_device *device = peer_device->device;
2205 if (device->state.conn == C_SYNC_SOURCE ||
2206 device->state.conn == C_SYNC_TARGET ||
2207 device->state.conn == C_PAUSED_SYNC_S ||
2208 device->state.conn == C_PAUSED_SYNC_T) {
2218 static bool conn_ov_running(struct drbd_connection *connection)
2220 struct drbd_peer_device *peer_device;
2225 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2226 struct drbd_device *device = peer_device->device;
2227 if (device->state.conn == C_VERIFY_S ||
2228 device->state.conn == C_VERIFY_T) {
2238 static enum drbd_ret_code
2239 _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2241 struct drbd_peer_device *peer_device;
2244 if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2245 if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2246 return ERR_NEED_APV_100;
2248 if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2249 return ERR_NEED_APV_100;
2251 if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2252 return ERR_NEED_APV_100;
2255 if (!new_net_conf->two_primaries &&
2256 conn_highest_role(connection) == R_PRIMARY &&
2257 conn_highest_peer(connection) == R_PRIMARY)
2258 return ERR_NEED_ALLOW_TWO_PRI;
2260 if (new_net_conf->two_primaries &&
2261 (new_net_conf->wire_protocol != DRBD_PROT_C))
2262 return ERR_NOT_PROTO_C;
2264 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2265 struct drbd_device *device = peer_device->device;
2266 if (get_ldev(device)) {
2267 enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2269 if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2270 return ERR_STONITH_AND_PROT_A;
2272 if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2273 return ERR_DISCARD_IMPOSSIBLE;
2276 if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2277 return ERR_CONG_NOT_PROTO_A;
2282 static enum drbd_ret_code
2283 check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2285 enum drbd_ret_code rv;
2286 struct drbd_peer_device *peer_device;
2290 rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2293 /* connection->peer_devices protected by genl_lock() here */
2294 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2295 struct drbd_device *device = peer_device->device;
2296 if (!device->bitmap) {
2297 if (drbd_bm_init(device))
2306 struct crypto_shash *verify_tfm;
2307 struct crypto_shash *csums_tfm;
2308 struct crypto_shash *cram_hmac_tfm;
2309 struct crypto_shash *integrity_tfm;
2313 alloc_shash(struct crypto_shash **tfm, char *tfm_name, int err_alg)
2318 *tfm = crypto_alloc_shash(tfm_name, 0, 0);
2327 static enum drbd_ret_code
2328 alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2330 char hmac_name[CRYPTO_MAX_ALG_NAME];
2331 enum drbd_ret_code rv;
2333 rv = alloc_shash(&crypto->csums_tfm, new_net_conf->csums_alg,
2337 rv = alloc_shash(&crypto->verify_tfm, new_net_conf->verify_alg,
2341 rv = alloc_shash(&crypto->integrity_tfm, new_net_conf->integrity_alg,
2345 if (new_net_conf->cram_hmac_alg[0] != 0) {
2346 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
2347 new_net_conf->cram_hmac_alg);
2349 rv = alloc_shash(&crypto->cram_hmac_tfm, hmac_name,
2356 static void free_crypto(struct crypto *crypto)
2358 crypto_free_shash(crypto->cram_hmac_tfm);
2359 crypto_free_shash(crypto->integrity_tfm);
2360 crypto_free_shash(crypto->csums_tfm);
2361 crypto_free_shash(crypto->verify_tfm);
2364 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2366 struct drbd_config_context adm_ctx;
2367 enum drbd_ret_code retcode;
2368 struct drbd_connection *connection;
2369 struct net_conf *old_net_conf, *new_net_conf = NULL;
2371 int ovr; /* online verify running */
2372 int rsr; /* re-sync running */
2373 struct crypto crypto = { };
2375 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2376 if (!adm_ctx.reply_skb)
2378 if (retcode != NO_ERROR)
2381 connection = adm_ctx.connection;
2382 mutex_lock(&adm_ctx.resource->adm_mutex);
2384 new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
2385 if (!new_net_conf) {
2386 retcode = ERR_NOMEM;
2390 conn_reconfig_start(connection);
2392 mutex_lock(&connection->data.mutex);
2393 mutex_lock(&connection->resource->conf_update);
2394 old_net_conf = connection->net_conf;
2396 if (!old_net_conf) {
2397 drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect");
2398 retcode = ERR_INVALID_REQUEST;
2402 *new_net_conf = *old_net_conf;
2403 if (should_set_defaults(info))
2404 set_net_conf_defaults(new_net_conf);
2406 err = net_conf_from_attrs_for_change(new_net_conf, info);
2407 if (err && err != -ENOMSG) {
2408 retcode = ERR_MANDATORY_TAG;
2409 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2413 retcode = check_net_options(connection, new_net_conf);
2414 if (retcode != NO_ERROR)
2417 /* re-sync running */
2418 rsr = conn_resync_running(connection);
2419 if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2420 retcode = ERR_CSUMS_RESYNC_RUNNING;
2424 /* online verify running */
2425 ovr = conn_ov_running(connection);
2426 if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2427 retcode = ERR_VERIFY_RUNNING;
2431 retcode = alloc_crypto(&crypto, new_net_conf);
2432 if (retcode != NO_ERROR)
2435 rcu_assign_pointer(connection->net_conf, new_net_conf);
2438 crypto_free_shash(connection->csums_tfm);
2439 connection->csums_tfm = crypto.csums_tfm;
2440 crypto.csums_tfm = NULL;
2443 crypto_free_shash(connection->verify_tfm);
2444 connection->verify_tfm = crypto.verify_tfm;
2445 crypto.verify_tfm = NULL;
2448 crypto_free_shash(connection->integrity_tfm);
2449 connection->integrity_tfm = crypto.integrity_tfm;
2450 if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2451 /* Do this without trying to take connection->data.mutex again. */
2452 __drbd_send_protocol(connection, P_PROTOCOL_UPDATE);
2454 crypto_free_shash(connection->cram_hmac_tfm);
2455 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2457 mutex_unlock(&connection->resource->conf_update);
2458 mutex_unlock(&connection->data.mutex);
2460 kfree(old_net_conf);
2462 if (connection->cstate >= C_WF_REPORT_PARAMS) {
2463 struct drbd_peer_device *peer_device;
2466 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2467 drbd_send_sync_param(peer_device);
2473 mutex_unlock(&connection->resource->conf_update);
2474 mutex_unlock(&connection->data.mutex);
2475 free_crypto(&crypto);
2476 kfree(new_net_conf);
2478 conn_reconfig_done(connection);
2480 mutex_unlock(&adm_ctx.resource->adm_mutex);
2482 drbd_adm_finish(&adm_ctx, info, retcode);
2486 static void connection_to_info(struct connection_info *info,
2487 struct drbd_connection *connection)
2489 info->conn_connection_state = connection->cstate;
2490 info->conn_role = conn_highest_peer(connection);
2493 static void peer_device_to_info(struct peer_device_info *info,
2494 struct drbd_peer_device *peer_device)
2496 struct drbd_device *device = peer_device->device;
2498 info->peer_repl_state =
2499 max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn);
2500 info->peer_disk_state = device->state.pdsk;
2501 info->peer_resync_susp_user = device->state.user_isp;
2502 info->peer_resync_susp_peer = device->state.peer_isp;
2503 info->peer_resync_susp_dependency = device->state.aftr_isp;
2506 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2508 struct connection_info connection_info;
2509 enum drbd_notification_type flags;
2510 unsigned int peer_devices = 0;
2511 struct drbd_config_context adm_ctx;
2512 struct drbd_peer_device *peer_device;
2513 struct net_conf *old_net_conf, *new_net_conf = NULL;
2514 struct crypto crypto = { };
2515 struct drbd_resource *resource;
2516 struct drbd_connection *connection;
2517 enum drbd_ret_code retcode;
2521 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2523 if (!adm_ctx.reply_skb)
2525 if (retcode != NO_ERROR)
2527 if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2528 drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing");
2529 retcode = ERR_INVALID_REQUEST;
2533 /* No need for _rcu here. All reconfiguration is
2534 * strictly serialized on genl_lock(). We are protected against
2535 * concurrent reconfiguration/addition/deletion */
2536 for_each_resource(resource, &drbd_resources) {
2537 for_each_connection(connection, resource) {
2538 if (nla_len(adm_ctx.my_addr) == connection->my_addr_len &&
2539 !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr,
2540 connection->my_addr_len)) {
2541 retcode = ERR_LOCAL_ADDR;
2545 if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len &&
2546 !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr,
2547 connection->peer_addr_len)) {
2548 retcode = ERR_PEER_ADDR;
2554 mutex_lock(&adm_ctx.resource->adm_mutex);
2555 connection = first_connection(adm_ctx.resource);
2556 conn_reconfig_start(connection);
2558 if (connection->cstate > C_STANDALONE) {
2559 retcode = ERR_NET_CONFIGURED;
2563 /* allocation not in the IO path, drbdsetup / netlink process context */
2564 new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL);
2565 if (!new_net_conf) {
2566 retcode = ERR_NOMEM;
2570 set_net_conf_defaults(new_net_conf);
2572 err = net_conf_from_attrs(new_net_conf, info);
2573 if (err && err != -ENOMSG) {
2574 retcode = ERR_MANDATORY_TAG;
2575 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2579 retcode = check_net_options(connection, new_net_conf);
2580 if (retcode != NO_ERROR)
2583 retcode = alloc_crypto(&crypto, new_net_conf);
2584 if (retcode != NO_ERROR)
2587 ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2589 drbd_flush_workqueue(&connection->sender_work);
2591 mutex_lock(&adm_ctx.resource->conf_update);
2592 old_net_conf = connection->net_conf;
2594 retcode = ERR_NET_CONFIGURED;
2595 mutex_unlock(&adm_ctx.resource->conf_update);
2598 rcu_assign_pointer(connection->net_conf, new_net_conf);
2600 conn_free_crypto(connection);
2601 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2602 connection->integrity_tfm = crypto.integrity_tfm;
2603 connection->csums_tfm = crypto.csums_tfm;
2604 connection->verify_tfm = crypto.verify_tfm;
2606 connection->my_addr_len = nla_len(adm_ctx.my_addr);
2607 memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2608 connection->peer_addr_len = nla_len(adm_ctx.peer_addr);
2609 memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2611 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2615 connection_to_info(&connection_info, connection);
2616 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2617 mutex_lock(¬ification_mutex);
2618 notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags);
2619 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2620 struct peer_device_info peer_device_info;
2622 peer_device_to_info(&peer_device_info, peer_device);
2623 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2624 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags);
2626 mutex_unlock(¬ification_mutex);
2627 mutex_unlock(&adm_ctx.resource->conf_update);
2630 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2631 struct drbd_device *device = peer_device->device;
2632 device->send_cnt = 0;
2633 device->recv_cnt = 0;
2637 retcode = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2639 conn_reconfig_done(connection);
2640 mutex_unlock(&adm_ctx.resource->adm_mutex);
2641 drbd_adm_finish(&adm_ctx, info, retcode);
2645 free_crypto(&crypto);
2646 kfree(new_net_conf);
2648 conn_reconfig_done(connection);
2649 mutex_unlock(&adm_ctx.resource->adm_mutex);
2651 drbd_adm_finish(&adm_ctx, info, retcode);
2655 static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2657 enum drbd_state_rv rv;
2659 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2660 force ? CS_HARD : 0);
2663 case SS_NOTHING_TO_DO:
2665 case SS_ALREADY_STANDALONE:
2667 case SS_PRIMARY_NOP:
2668 /* Our state checking code wants to see the peer outdated. */
2669 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2671 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2672 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2675 case SS_CW_FAILED_BY_PEER:
2676 /* The peer probably wants to see us outdated. */
2677 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2678 disk, D_OUTDATED), 0);
2679 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2680 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2685 /* no special handling necessary */
2688 if (rv >= SS_SUCCESS) {
2689 enum drbd_state_rv rv2;
2690 /* No one else can reconfigure the network while I am here.
2691 * The state handling only uses drbd_thread_stop_nowait(),
2692 * we want to really wait here until the receiver is no more.
2694 drbd_thread_stop(&connection->receiver);
2696 /* Race breaker. This additional state change request may be
2697 * necessary, if this was a forced disconnect during a receiver
2698 * restart. We may have "killed" the receiver thread just
2699 * after drbd_receiver() returned. Typically, we should be
2700 * C_STANDALONE already, now, and this becomes a no-op.
2702 rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2703 CS_VERBOSE | CS_HARD);
2704 if (rv2 < SS_SUCCESS)
2705 drbd_err(connection,
2706 "unexpected rv2=%d in conn_try_disconnect()\n",
2708 /* Unlike in DRBD 9, the state engine has generated
2709 * NOTIFY_DESTROY events before clearing connection->net_conf. */
2714 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2716 struct drbd_config_context adm_ctx;
2717 struct disconnect_parms parms;
2718 struct drbd_connection *connection;
2719 enum drbd_state_rv rv;
2720 enum drbd_ret_code retcode;
2723 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2724 if (!adm_ctx.reply_skb)
2726 if (retcode != NO_ERROR)
2729 connection = adm_ctx.connection;
2730 memset(&parms, 0, sizeof(parms));
2731 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2732 err = disconnect_parms_from_attrs(&parms, info);
2734 retcode = ERR_MANDATORY_TAG;
2735 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2740 mutex_lock(&adm_ctx.resource->adm_mutex);
2741 rv = conn_try_disconnect(connection, parms.force_disconnect);
2742 if (rv < SS_SUCCESS)
2743 retcode = rv; /* FIXME: Type mismatch. */
2746 mutex_unlock(&adm_ctx.resource->adm_mutex);
2748 drbd_adm_finish(&adm_ctx, info, retcode);
2752 void resync_after_online_grow(struct drbd_device *device)
2754 int iass; /* I am sync source */
2756 drbd_info(device, "Resync of new storage after online grow\n");
2757 if (device->state.role != device->state.peer)
2758 iass = (device->state.role == R_PRIMARY);
2760 iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2763 drbd_start_resync(device, C_SYNC_SOURCE);
2765 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2768 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2770 struct drbd_config_context adm_ctx;
2771 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2772 struct resize_parms rs;
2773 struct drbd_device *device;
2774 enum drbd_ret_code retcode;
2775 enum determine_dev_size dd;
2776 bool change_al_layout = false;
2777 enum dds_flags ddsf;
2781 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2782 if (!adm_ctx.reply_skb)
2784 if (retcode != NO_ERROR)
2787 mutex_lock(&adm_ctx.resource->adm_mutex);
2788 device = adm_ctx.device;
2789 if (!get_ldev(device)) {
2790 retcode = ERR_NO_DISK;
2794 memset(&rs, 0, sizeof(struct resize_parms));
2795 rs.al_stripes = device->ldev->md.al_stripes;
2796 rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2797 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2798 err = resize_parms_from_attrs(&rs, info);
2800 retcode = ERR_MANDATORY_TAG;
2801 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2806 if (device->state.conn > C_CONNECTED) {
2807 retcode = ERR_RESIZE_RESYNC;
2811 if (device->state.role == R_SECONDARY &&
2812 device->state.peer == R_SECONDARY) {
2813 retcode = ERR_NO_PRIMARY;
2817 if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2818 retcode = ERR_NEED_APV_93;
2823 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2825 if (u_size != (sector_t)rs.resize_size) {
2826 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2827 if (!new_disk_conf) {
2828 retcode = ERR_NOMEM;
2833 if (device->ldev->md.al_stripes != rs.al_stripes ||
2834 device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2835 u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2837 if (al_size_k > (16 * 1024 * 1024)) {
2838 retcode = ERR_MD_LAYOUT_TOO_BIG;
2842 if (al_size_k < MD_32kB_SECT/2) {
2843 retcode = ERR_MD_LAYOUT_TOO_SMALL;
2847 if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2848 retcode = ERR_MD_LAYOUT_CONNECTED;
2852 change_al_layout = true;
2855 if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev))
2856 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
2858 if (new_disk_conf) {
2859 mutex_lock(&device->resource->conf_update);
2860 old_disk_conf = device->ldev->disk_conf;
2861 *new_disk_conf = *old_disk_conf;
2862 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2863 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2864 mutex_unlock(&device->resource->conf_update);
2866 kfree(old_disk_conf);
2867 new_disk_conf = NULL;
2870 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2871 dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL);
2872 drbd_md_sync(device);
2874 if (dd == DS_ERROR) {
2875 retcode = ERR_NOMEM_BITMAP;
2877 } else if (dd == DS_ERROR_SPACE_MD) {
2878 retcode = ERR_MD_LAYOUT_NO_FIT;
2880 } else if (dd == DS_ERROR_SHRINK) {
2881 retcode = ERR_IMPLICIT_SHRINK;
2885 if (device->state.conn == C_CONNECTED) {
2887 set_bit(RESIZE_PENDING, &device->flags);
2889 drbd_send_uuids(first_peer_device(device));
2890 drbd_send_sizes(first_peer_device(device), 1, ddsf);
2894 mutex_unlock(&adm_ctx.resource->adm_mutex);
2896 drbd_adm_finish(&adm_ctx, info, retcode);
2901 kfree(new_disk_conf);
2905 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2907 struct drbd_config_context adm_ctx;
2908 enum drbd_ret_code retcode;
2909 struct res_opts res_opts;
2912 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2913 if (!adm_ctx.reply_skb)
2915 if (retcode != NO_ERROR)
2918 res_opts = adm_ctx.resource->res_opts;
2919 if (should_set_defaults(info))
2920 set_res_opts_defaults(&res_opts);
2922 err = res_opts_from_attrs(&res_opts, info);
2923 if (err && err != -ENOMSG) {
2924 retcode = ERR_MANDATORY_TAG;
2925 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2929 mutex_lock(&adm_ctx.resource->adm_mutex);
2930 err = set_resource_options(adm_ctx.resource, &res_opts);
2932 retcode = ERR_INVALID_REQUEST;
2934 retcode = ERR_NOMEM;
2936 mutex_unlock(&adm_ctx.resource->adm_mutex);
2939 drbd_adm_finish(&adm_ctx, info, retcode);
2943 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2945 struct drbd_config_context adm_ctx;
2946 struct drbd_device *device;
2947 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2949 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2950 if (!adm_ctx.reply_skb)
2952 if (retcode != NO_ERROR)
2955 device = adm_ctx.device;
2956 if (!get_ldev(device)) {
2957 retcode = ERR_NO_DISK;
2961 mutex_lock(&adm_ctx.resource->adm_mutex);
2963 /* If there is still bitmap IO pending, probably because of a previous
2964 * resync just being finished, wait for it before requesting a new resync.
2965 * Also wait for it's after_state_ch(). */
2966 drbd_suspend_io(device);
2967 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2968 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2970 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2971 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2972 * try to start a resync handshake as sync target for full sync.
2974 if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
2975 retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
2976 if (retcode >= SS_SUCCESS) {
2977 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2978 "set_n_write from invalidate", BM_LOCKED_MASK))
2979 retcode = ERR_IO_MD_DISK;
2982 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
2983 drbd_resume_io(device);
2984 mutex_unlock(&adm_ctx.resource->adm_mutex);
2987 drbd_adm_finish(&adm_ctx, info, retcode);
2991 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2992 union drbd_state mask, union drbd_state val)
2994 struct drbd_config_context adm_ctx;
2995 enum drbd_ret_code retcode;
2997 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2998 if (!adm_ctx.reply_skb)
3000 if (retcode != NO_ERROR)
3003 mutex_lock(&adm_ctx.resource->adm_mutex);
3004 retcode = drbd_request_state(adm_ctx.device, mask, val);
3005 mutex_unlock(&adm_ctx.resource->adm_mutex);
3007 drbd_adm_finish(&adm_ctx, info, retcode);
3011 static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local)
3015 rv = drbd_bmio_set_n_write(device);
3016 drbd_suspend_al(device);
3020 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
3022 struct drbd_config_context adm_ctx;
3023 int retcode; /* drbd_ret_code, drbd_state_rv */
3024 struct drbd_device *device;
3026 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3027 if (!adm_ctx.reply_skb)
3029 if (retcode != NO_ERROR)
3032 device = adm_ctx.device;
3033 if (!get_ldev(device)) {
3034 retcode = ERR_NO_DISK;
3038 mutex_lock(&adm_ctx.resource->adm_mutex);
3040 /* If there is still bitmap IO pending, probably because of a previous
3041 * resync just being finished, wait for it before requesting a new resync.
3042 * Also wait for it's after_state_ch(). */
3043 drbd_suspend_io(device);
3044 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
3045 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
3047 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
3048 * in the bitmap. Otherwise, try to start a resync handshake
3049 * as sync source for full sync.
3051 if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
3052 /* The peer will get a resync upon connect anyways. Just make that
3053 into a full resync. */
3054 retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
3055 if (retcode >= SS_SUCCESS) {
3056 if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al,
3057 "set_n_write from invalidate_peer",
3058 BM_LOCKED_SET_ALLOWED))
3059 retcode = ERR_IO_MD_DISK;
3062 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
3063 drbd_resume_io(device);
3064 mutex_unlock(&adm_ctx.resource->adm_mutex);
3067 drbd_adm_finish(&adm_ctx, info, retcode);
3071 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
3073 struct drbd_config_context adm_ctx;
3074 enum drbd_ret_code retcode;
3076 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3077 if (!adm_ctx.reply_skb)
3079 if (retcode != NO_ERROR)
3082 mutex_lock(&adm_ctx.resource->adm_mutex);
3083 if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
3084 retcode = ERR_PAUSE_IS_SET;
3085 mutex_unlock(&adm_ctx.resource->adm_mutex);
3087 drbd_adm_finish(&adm_ctx, info, retcode);
3091 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
3093 struct drbd_config_context adm_ctx;
3094 union drbd_dev_state s;
3095 enum drbd_ret_code retcode;
3097 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3098 if (!adm_ctx.reply_skb)
3100 if (retcode != NO_ERROR)
3103 mutex_lock(&adm_ctx.resource->adm_mutex);
3104 if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
3105 s = adm_ctx.device->state;
3106 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
3107 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
3108 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
3110 retcode = ERR_PAUSE_IS_CLEAR;
3113 mutex_unlock(&adm_ctx.resource->adm_mutex);
3115 drbd_adm_finish(&adm_ctx, info, retcode);
3119 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
3121 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
3124 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
3126 struct drbd_config_context adm_ctx;
3127 struct drbd_device *device;
3128 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3130 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3131 if (!adm_ctx.reply_skb)
3133 if (retcode != NO_ERROR)
3136 mutex_lock(&adm_ctx.resource->adm_mutex);
3137 device = adm_ctx.device;
3138 if (test_bit(NEW_CUR_UUID, &device->flags)) {
3139 if (get_ldev_if_state(device, D_ATTACHING)) {
3140 drbd_uuid_new_current(device);
3143 /* This is effectively a multi-stage "forced down".
3144 * The NEW_CUR_UUID bit is supposedly only set, if we
3145 * lost the replication connection, and are configured
3146 * to freeze IO and wait for some fence-peer handler.
3147 * So we still don't have a replication connection.
3148 * And now we don't have a local disk either. After
3149 * resume, we will fail all pending and new IO, because
3150 * we don't have any data anymore. Which means we will
3151 * eventually be able to terminate all users of this
3152 * device, and then take it down. By bumping the
3153 * "effective" data uuid, we make sure that you really
3154 * need to tear down before you reconfigure, we will
3155 * the refuse to re-connect or re-attach (because no
3156 * matching real data uuid exists).
3159 get_random_bytes(&val, sizeof(u64));
3160 drbd_set_ed_uuid(device, val);
3161 drbd_warn(device, "Resumed without access to data; please tear down before attempting to re-configure.\n");
3163 clear_bit(NEW_CUR_UUID, &device->flags);
3165 drbd_suspend_io(device);
3166 retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
3167 if (retcode == SS_SUCCESS) {
3168 if (device->state.conn < C_CONNECTED)
3169 tl_clear(first_peer_device(device)->connection);
3170 if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
3171 tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO);
3173 drbd_resume_io(device);
3174 mutex_unlock(&adm_ctx.resource->adm_mutex);
3176 drbd_adm_finish(&adm_ctx, info, retcode);
3180 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
3182 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
3185 static int nla_put_drbd_cfg_context(struct sk_buff *skb,
3186 struct drbd_resource *resource,
3187 struct drbd_connection *connection,
3188 struct drbd_device *device)
3191 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
3193 goto nla_put_failure;
3195 nla_put_u32(skb, T_ctx_volume, device->vnr))
3196 goto nla_put_failure;
3197 if (nla_put_string(skb, T_ctx_resource_name, resource->name))
3198 goto nla_put_failure;
3200 if (connection->my_addr_len &&
3201 nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr))
3202 goto nla_put_failure;
3203 if (connection->peer_addr_len &&
3204 nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr))
3205 goto nla_put_failure;
3207 nla_nest_end(skb, nla);
3212 nla_nest_cancel(skb, nla);
3217 * The generic netlink dump callbacks are called outside the genl_lock(), so
3218 * they cannot use the simple attribute parsing code which uses global
3221 static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr)
3223 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3224 const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3227 nla = nla_find(nlmsg_attrdata(nlh, hdrlen), nlmsg_attrlen(nlh, hdrlen),
3228 DRBD_NLA_CFG_CONTEXT);
3231 return drbd_nla_find_nested(maxtype, nla, __nla_type(attr));
3234 static void resource_to_info(struct resource_info *, struct drbd_resource *);
3236 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb)
3238 struct drbd_genlmsghdr *dh;
3239 struct drbd_resource *resource;
3240 struct resource_info resource_info;
3241 struct resource_statistics resource_statistics;
3246 for_each_resource_rcu(resource, &drbd_resources)
3247 if (resource == (struct drbd_resource *)cb->args[0])
3248 goto found_resource;
3249 err = 0; /* resource was probably deleted */
3252 resource = list_entry(&drbd_resources,
3253 struct drbd_resource, resources);
3256 list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) {
3263 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3264 cb->nlh->nlmsg_seq, &drbd_genl_family,
3265 NLM_F_MULTI, DRBD_ADM_GET_RESOURCES);
3270 dh->ret_code = NO_ERROR;
3271 err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL);
3274 err = res_opts_to_skb(skb, &resource->res_opts, !capable(CAP_SYS_ADMIN));
3277 resource_to_info(&resource_info, resource);
3278 err = resource_info_to_skb(skb, &resource_info, !capable(CAP_SYS_ADMIN));
3281 resource_statistics.res_stat_write_ordering = resource->write_ordering;
3282 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
3285 cb->args[0] = (long)resource;
3286 genlmsg_end(skb, dh);
3296 static void device_to_statistics(struct device_statistics *s,
3297 struct drbd_device *device)
3299 memset(s, 0, sizeof(*s));
3300 s->dev_upper_blocked = !may_inc_ap_bio(device);
3301 if (get_ldev(device)) {
3302 struct drbd_md *md = &device->ldev->md;
3303 u64 *history_uuids = (u64 *)s->history_uuids;
3304 struct request_queue *q;
3307 spin_lock_irq(&md->uuid_lock);
3308 s->dev_current_uuid = md->uuid[UI_CURRENT];
3309 BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1);
3310 for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++)
3311 history_uuids[n] = md->uuid[UI_HISTORY_START + n];
3312 for (; n < HISTORY_UUIDS; n++)
3313 history_uuids[n] = 0;
3314 s->history_uuids_len = HISTORY_UUIDS;
3315 spin_unlock_irq(&md->uuid_lock);
3317 s->dev_disk_flags = md->flags;
3318 q = bdev_get_queue(device->ldev->backing_bdev);
3319 s->dev_lower_blocked =
3320 bdi_congested(q->backing_dev_info,
3321 (1 << WB_async_congested) |
3322 (1 << WB_sync_congested));
3325 s->dev_size = drbd_get_capacity(device->this_bdev);
3326 s->dev_read = device->read_cnt;
3327 s->dev_write = device->writ_cnt;
3328 s->dev_al_writes = device->al_writ_cnt;
3329 s->dev_bm_writes = device->bm_writ_cnt;
3330 s->dev_upper_pending = atomic_read(&device->ap_bio_cnt);
3331 s->dev_lower_pending = atomic_read(&device->local_cnt);
3332 s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags);
3333 s->dev_exposed_data_uuid = device->ed_uuid;
3336 static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr)
3339 struct drbd_resource *resource =
3340 (struct drbd_resource *)cb->args[0];
3341 kref_put(&resource->kref, drbd_destroy_resource);
3347 int drbd_adm_dump_devices_done(struct netlink_callback *cb) {
3348 return put_resource_in_arg0(cb, 7);
3351 static void device_to_info(struct device_info *, struct drbd_device *);
3353 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb)
3355 struct nlattr *resource_filter;
3356 struct drbd_resource *resource;
3357 struct drbd_device *uninitialized_var(device);
3358 int minor, err, retcode;
3359 struct drbd_genlmsghdr *dh;
3360 struct device_info device_info;
3361 struct device_statistics device_statistics;
3362 struct idr *idr_to_search;
3364 resource = (struct drbd_resource *)cb->args[0];
3365 if (!cb->args[0] && !cb->args[1]) {
3366 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3367 if (resource_filter) {
3368 retcode = ERR_RES_NOT_KNOWN;
3369 resource = drbd_find_resource(nla_data(resource_filter));
3372 cb->args[0] = (long)resource;
3377 minor = cb->args[1];
3378 idr_to_search = resource ? &resource->devices : &drbd_devices;
3379 device = idr_get_next(idr_to_search, &minor);
3384 idr_for_each_entry_continue(idr_to_search, device, minor) {
3386 goto put_result; /* only one iteration */
3389 goto out; /* no more devices */
3392 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3393 cb->nlh->nlmsg_seq, &drbd_genl_family,
3394 NLM_F_MULTI, DRBD_ADM_GET_DEVICES);
3398 dh->ret_code = retcode;
3400 if (retcode == NO_ERROR) {
3401 dh->minor = device->minor;
3402 err = nla_put_drbd_cfg_context(skb, device->resource, NULL, device);
3405 if (get_ldev(device)) {
3406 struct disk_conf *disk_conf =
3407 rcu_dereference(device->ldev->disk_conf);
3409 err = disk_conf_to_skb(skb, disk_conf, !capable(CAP_SYS_ADMIN));
3414 device_to_info(&device_info, device);
3415 err = device_info_to_skb(skb, &device_info, !capable(CAP_SYS_ADMIN));
3419 device_to_statistics(&device_statistics, device);
3420 err = device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
3423 cb->args[1] = minor + 1;
3425 genlmsg_end(skb, dh);
3435 int drbd_adm_dump_connections_done(struct netlink_callback *cb)
3437 return put_resource_in_arg0(cb, 6);
3440 enum { SINGLE_RESOURCE, ITERATE_RESOURCES };
3442 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb)
3444 struct nlattr *resource_filter;
3445 struct drbd_resource *resource = NULL, *next_resource;
3446 struct drbd_connection *uninitialized_var(connection);
3447 int err = 0, retcode;
3448 struct drbd_genlmsghdr *dh;
3449 struct connection_info connection_info;
3450 struct connection_statistics connection_statistics;
3453 resource = (struct drbd_resource *)cb->args[0];
3455 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3456 if (resource_filter) {
3457 retcode = ERR_RES_NOT_KNOWN;
3458 resource = drbd_find_resource(nla_data(resource_filter));
3461 cb->args[0] = (long)resource;
3462 cb->args[1] = SINGLE_RESOURCE;
3466 if (list_empty(&drbd_resources))
3468 resource = list_first_entry(&drbd_resources, struct drbd_resource, resources);
3469 kref_get(&resource->kref);
3470 cb->args[0] = (long)resource;
3471 cb->args[1] = ITERATE_RESOURCES;
3476 mutex_lock(&resource->conf_update);
3479 for_each_connection_rcu(connection, resource)
3480 if (connection == (struct drbd_connection *)cb->args[2])
3481 goto found_connection;
3482 /* connection was probably deleted */
3483 goto no_more_connections;
3485 connection = list_entry(&resource->connections, struct drbd_connection, connections);
3488 list_for_each_entry_continue_rcu(connection, &resource->connections, connections) {
3489 if (!has_net_conf(connection))
3492 goto put_result; /* only one iteration */
3495 no_more_connections:
3496 if (cb->args[1] == ITERATE_RESOURCES) {
3497 for_each_resource_rcu(next_resource, &drbd_resources) {
3498 if (next_resource == resource)
3499 goto found_resource;
3501 /* resource was probably deleted */
3506 list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) {
3507 mutex_unlock(&resource->conf_update);
3508 kref_put(&resource->kref, drbd_destroy_resource);
3509 resource = next_resource;
3510 kref_get(&resource->kref);
3511 cb->args[0] = (long)resource;
3515 goto out; /* no more resources */
3518 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3519 cb->nlh->nlmsg_seq, &drbd_genl_family,
3520 NLM_F_MULTI, DRBD_ADM_GET_CONNECTIONS);
3524 dh->ret_code = retcode;
3526 if (retcode == NO_ERROR) {
3527 struct net_conf *net_conf;
3529 err = nla_put_drbd_cfg_context(skb, resource, connection, NULL);
3532 net_conf = rcu_dereference(connection->net_conf);
3534 err = net_conf_to_skb(skb, net_conf, !capable(CAP_SYS_ADMIN));
3538 connection_to_info(&connection_info, connection);
3539 err = connection_info_to_skb(skb, &connection_info, !capable(CAP_SYS_ADMIN));
3542 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
3543 err = connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
3546 cb->args[2] = (long)connection;
3548 genlmsg_end(skb, dh);
3554 mutex_unlock(&resource->conf_update);
3560 enum mdf_peer_flag {
3561 MDF_PEER_CONNECTED = 1 << 0,
3562 MDF_PEER_OUTDATED = 1 << 1,
3563 MDF_PEER_FENCING = 1 << 2,
3564 MDF_PEER_FULL_SYNC = 1 << 3,
3567 static void peer_device_to_statistics(struct peer_device_statistics *s,
3568 struct drbd_peer_device *peer_device)
3570 struct drbd_device *device = peer_device->device;
3572 memset(s, 0, sizeof(*s));
3573 s->peer_dev_received = device->recv_cnt;
3574 s->peer_dev_sent = device->send_cnt;
3575 s->peer_dev_pending = atomic_read(&device->ap_pending_cnt) +
3576 atomic_read(&device->rs_pending_cnt);
3577 s->peer_dev_unacked = atomic_read(&device->unacked_cnt);
3578 s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9);
3579 s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9);
3580 if (get_ldev(device)) {
3581 struct drbd_md *md = &device->ldev->md;
3583 spin_lock_irq(&md->uuid_lock);
3584 s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP];
3585 spin_unlock_irq(&md->uuid_lock);
3587 (drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ?
3588 MDF_PEER_CONNECTED : 0) +
3589 (drbd_md_test_flag(device->ldev, MDF_CONSISTENT) &&
3590 !drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ?
3591 MDF_PEER_OUTDATED : 0) +
3592 /* FIXME: MDF_PEER_FENCING? */
3593 (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ?
3594 MDF_PEER_FULL_SYNC : 0);
3599 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb)
3601 return put_resource_in_arg0(cb, 9);
3604 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb)
3606 struct nlattr *resource_filter;
3607 struct drbd_resource *resource;
3608 struct drbd_device *uninitialized_var(device);
3609 struct drbd_peer_device *peer_device = NULL;
3610 int minor, err, retcode;
3611 struct drbd_genlmsghdr *dh;
3612 struct idr *idr_to_search;
3614 resource = (struct drbd_resource *)cb->args[0];
3615 if (!cb->args[0] && !cb->args[1]) {
3616 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3617 if (resource_filter) {
3618 retcode = ERR_RES_NOT_KNOWN;
3619 resource = drbd_find_resource(nla_data(resource_filter));
3623 cb->args[0] = (long)resource;
3627 minor = cb->args[1];
3628 idr_to_search = resource ? &resource->devices : &drbd_devices;
3629 device = idr_find(idr_to_search, minor);
3634 device = idr_get_next(idr_to_search, &minor);
3641 for_each_peer_device(peer_device, device)
3642 if (peer_device == (struct drbd_peer_device *)cb->args[2])
3643 goto found_peer_device;
3644 /* peer device was probably deleted */
3647 /* Make peer_device point to the list head (not the first entry). */
3648 peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices);
3651 list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) {
3652 if (!has_net_conf(peer_device->connection))
3655 goto put_result; /* only one iteration */
3660 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3661 cb->nlh->nlmsg_seq, &drbd_genl_family,
3662 NLM_F_MULTI, DRBD_ADM_GET_PEER_DEVICES);
3666 dh->ret_code = retcode;
3668 if (retcode == NO_ERROR) {
3669 struct peer_device_info peer_device_info;
3670 struct peer_device_statistics peer_device_statistics;
3673 err = nla_put_drbd_cfg_context(skb, device->resource, peer_device->connection, device);
3676 peer_device_to_info(&peer_device_info, peer_device);
3677 err = peer_device_info_to_skb(skb, &peer_device_info, !capable(CAP_SYS_ADMIN));
3680 peer_device_to_statistics(&peer_device_statistics, peer_device);
3681 err = peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
3684 cb->args[1] = minor;
3685 cb->args[2] = (long)peer_device;
3687 genlmsg_end(skb, dh);
3697 * Return the connection of @resource if @resource has exactly one connection.
3699 static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
3701 struct list_head *connections = &resource->connections;
3703 if (list_empty(connections) || connections->next->next != connections)
3705 return list_first_entry(&resource->connections, struct drbd_connection, connections);
3708 static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
3709 const struct sib_info *sib)
3711 struct drbd_resource *resource = device->resource;
3712 struct state_info *si = NULL; /* for sizeof(si->member); */
3716 int exclude_sensitive;
3718 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
3719 * to. So we better exclude_sensitive information.
3721 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
3722 * in the context of the requesting user process. Exclude sensitive
3723 * information, unless current has superuser.
3725 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
3726 * relies on the current implementation of netlink_dump(), which
3727 * executes the dump callback successively from netlink_recvmsg(),
3728 * always in the context of the receiving process */
3729 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
3731 got_ldev = get_ldev(device);
3733 /* We need to add connection name and volume number information still.
3734 * Minor number is in drbd_genlmsghdr. */
3735 if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device))
3736 goto nla_put_failure;
3738 if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive))
3739 goto nla_put_failure;
3743 struct disk_conf *disk_conf;
3745 disk_conf = rcu_dereference(device->ldev->disk_conf);
3746 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
3749 struct net_conf *nc;
3751 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3753 err = net_conf_to_skb(skb, nc, exclude_sensitive);
3757 goto nla_put_failure;
3759 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
3761 goto nla_put_failure;
3762 if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
3763 nla_put_u32(skb, T_current_state, device->state.i) ||
3764 nla_put_u64_0pad(skb, T_ed_uuid, device->ed_uuid) ||
3765 nla_put_u64_0pad(skb, T_capacity,
3766 drbd_get_capacity(device->this_bdev)) ||
3767 nla_put_u64_0pad(skb, T_send_cnt, device->send_cnt) ||
3768 nla_put_u64_0pad(skb, T_recv_cnt, device->recv_cnt) ||
3769 nla_put_u64_0pad(skb, T_read_cnt, device->read_cnt) ||
3770 nla_put_u64_0pad(skb, T_writ_cnt, device->writ_cnt) ||
3771 nla_put_u64_0pad(skb, T_al_writ_cnt, device->al_writ_cnt) ||
3772 nla_put_u64_0pad(skb, T_bm_writ_cnt, device->bm_writ_cnt) ||
3773 nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) ||
3774 nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) ||
3775 nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt)))
3776 goto nla_put_failure;
3781 spin_lock_irq(&device->ldev->md.uuid_lock);
3782 err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid);
3783 spin_unlock_irq(&device->ldev->md.uuid_lock);
3786 goto nla_put_failure;
3788 if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) ||
3789 nla_put_u64_0pad(skb, T_bits_total, drbd_bm_bits(device)) ||
3790 nla_put_u64_0pad(skb, T_bits_oos,
3791 drbd_bm_total_weight(device)))
3792 goto nla_put_failure;
3793 if (C_SYNC_SOURCE <= device->state.conn &&
3794 C_PAUSED_SYNC_T >= device->state.conn) {
3795 if (nla_put_u64_0pad(skb, T_bits_rs_total,
3796 device->rs_total) ||
3797 nla_put_u64_0pad(skb, T_bits_rs_failed,
3799 goto nla_put_failure;
3804 switch(sib->sib_reason) {
3805 case SIB_SYNC_PROGRESS:
3806 case SIB_GET_STATUS_REPLY:
3808 case SIB_STATE_CHANGE:
3809 if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
3810 nla_put_u32(skb, T_new_state, sib->ns.i))
3811 goto nla_put_failure;
3813 case SIB_HELPER_POST:
3814 if (nla_put_u32(skb, T_helper_exit_code,
3815 sib->helper_exit_code))
3816 goto nla_put_failure;
3818 case SIB_HELPER_PRE:
3819 if (nla_put_string(skb, T_helper, sib->helper_name))
3820 goto nla_put_failure;
3824 nla_nest_end(skb, nla);
3834 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3836 struct drbd_config_context adm_ctx;
3837 enum drbd_ret_code retcode;
3840 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3841 if (!adm_ctx.reply_skb)
3843 if (retcode != NO_ERROR)
3846 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL);
3848 nlmsg_free(adm_ctx.reply_skb);
3852 drbd_adm_finish(&adm_ctx, info, retcode);
3856 static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3858 struct drbd_device *device;
3859 struct drbd_genlmsghdr *dh;
3860 struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3861 struct drbd_resource *resource = NULL;
3862 struct drbd_resource *tmp;
3863 unsigned volume = cb->args[1];
3865 /* Open coded, deferred, iteration:
3866 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3867 * connection = "first connection of resource or undefined";
3868 * idr_for_each_entry(&resource->devices, device, i) {
3872 * where resource is cb->args[0];
3873 * and i is cb->args[1];
3875 * cb->args[2] indicates if we shall loop over all resources,
3876 * or just dump all volumes of a single resource.
3878 * This may miss entries inserted after this dump started,
3879 * or entries deleted before they are reached.
3881 * We need to make sure the device won't disappear while
3882 * we are looking at it, and revalidate our iterators
3883 * on each iteration.
3886 /* synchronize with conn_create()/drbd_destroy_connection() */
3888 /* revalidate iterator position */
3889 for_each_resource_rcu(tmp, &drbd_resources) {
3891 /* first iteration */
3903 device = idr_get_next(&resource->devices, &volume);
3905 /* No more volumes to dump on this resource.
3906 * Advance resource iterator. */
3907 pos = list_entry_rcu(resource->resources.next,
3908 struct drbd_resource, resources);
3909 /* Did we dump any volume of this resource yet? */
3911 /* If we reached the end of the list,
3912 * or only a single resource dump was requested,
3914 if (&pos->resources == &drbd_resources || cb->args[2])
3922 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3923 cb->nlh->nlmsg_seq, &drbd_genl_family,
3924 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
3929 /* This is a connection without a single volume.
3930 * Suprisingly enough, it may have a network
3932 struct drbd_connection *connection;
3935 dh->ret_code = NO_ERROR;
3936 connection = the_only_connection(resource);
3937 if (nla_put_drbd_cfg_context(skb, resource, connection, NULL))
3940 struct net_conf *nc;
3942 nc = rcu_dereference(connection->net_conf);
3943 if (nc && net_conf_to_skb(skb, nc, 1) != 0)
3949 D_ASSERT(device, device->vnr == volume);
3950 D_ASSERT(device, device->resource == resource);
3952 dh->minor = device_to_minor(device);
3953 dh->ret_code = NO_ERROR;
3955 if (nla_put_status_info(skb, device, NULL)) {
3957 genlmsg_cancel(skb, dh);
3961 genlmsg_end(skb, dh);
3966 /* where to start the next iteration */
3967 cb->args[0] = (long)pos;
3968 cb->args[1] = (pos == resource) ? volume + 1 : 0;
3970 /* No more resources/volumes/minors found results in an empty skb.
3971 * Which will terminate the dump. */
3976 * Request status of all resources, or of all volumes within a single resource.
3978 * This is a dump, as the answer may not fit in a single reply skb otherwise.
3979 * Which means we cannot use the family->attrbuf or other such members, because
3980 * dump is NOT protected by the genl_lock(). During dump, we only have access
3981 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3983 * Once things are setup properly, we call into get_one_status().
3985 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
3987 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3989 const char *resource_name;
3990 struct drbd_resource *resource;
3993 /* Is this a followup call? */
3995 /* ... of a single resource dump,
3996 * and the resource iterator has been advanced already? */
3997 if (cb->args[2] && cb->args[2] != cb->args[0])
3998 return 0; /* DONE. */
4002 /* First call (from netlink_dump_start). We need to figure out
4003 * which resource(s) the user wants us to dump. */
4004 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
4005 nlmsg_attrlen(cb->nlh, hdrlen),
4006 DRBD_NLA_CFG_CONTEXT);
4008 /* No explicit context given. Dump all. */
4011 maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
4012 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
4014 return PTR_ERR(nla);
4015 /* context given, but no name present? */
4018 resource_name = nla_data(nla);
4019 if (!*resource_name)
4021 resource = drbd_find_resource(resource_name);
4025 kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */
4027 /* prime iterators, and set "filter" mode mark:
4028 * only dump this connection. */
4029 cb->args[0] = (long)resource;
4030 /* cb->args[1] = 0; passed in this way. */
4031 cb->args[2] = (long)resource;
4034 return get_one_status(skb, cb);
4037 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
4039 struct drbd_config_context adm_ctx;
4040 enum drbd_ret_code retcode;
4041 struct timeout_parms tp;
4044 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4045 if (!adm_ctx.reply_skb)
4047 if (retcode != NO_ERROR)
4051 adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
4052 test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED :
4055 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
4057 nlmsg_free(adm_ctx.reply_skb);
4061 drbd_adm_finish(&adm_ctx, info, retcode);
4065 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
4067 struct drbd_config_context adm_ctx;
4068 struct drbd_device *device;
4069 enum drbd_ret_code retcode;
4070 struct start_ov_parms parms;
4072 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4073 if (!adm_ctx.reply_skb)
4075 if (retcode != NO_ERROR)
4078 device = adm_ctx.device;
4080 /* resume from last known position, if possible */
4081 parms.ov_start_sector = device->ov_start_sector;
4082 parms.ov_stop_sector = ULLONG_MAX;
4083 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
4084 int err = start_ov_parms_from_attrs(&parms, info);
4086 retcode = ERR_MANDATORY_TAG;
4087 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4091 mutex_lock(&adm_ctx.resource->adm_mutex);
4093 /* w_make_ov_request expects position to be aligned */
4094 device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
4095 device->ov_stop_sector = parms.ov_stop_sector;
4097 /* If there is still bitmap IO pending, e.g. previous resync or verify
4098 * just being finished, wait for it before requesting a new resync. */
4099 drbd_suspend_io(device);
4100 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
4101 retcode = drbd_request_state(device, NS(conn, C_VERIFY_S));
4102 drbd_resume_io(device);
4104 mutex_unlock(&adm_ctx.resource->adm_mutex);
4106 drbd_adm_finish(&adm_ctx, info, retcode);
4111 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
4113 struct drbd_config_context adm_ctx;
4114 struct drbd_device *device;
4115 enum drbd_ret_code retcode;
4116 int skip_initial_sync = 0;
4118 struct new_c_uuid_parms args;
4120 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4121 if (!adm_ctx.reply_skb)
4123 if (retcode != NO_ERROR)
4126 device = adm_ctx.device;
4127 memset(&args, 0, sizeof(args));
4128 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
4129 err = new_c_uuid_parms_from_attrs(&args, info);
4131 retcode = ERR_MANDATORY_TAG;
4132 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4137 mutex_lock(&adm_ctx.resource->adm_mutex);
4138 mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */
4140 if (!get_ldev(device)) {
4141 retcode = ERR_NO_DISK;
4145 /* this is "skip initial sync", assume to be clean */
4146 if (device->state.conn == C_CONNECTED &&
4147 first_peer_device(device)->connection->agreed_pro_version >= 90 &&
4148 device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
4149 drbd_info(device, "Preparing to skip initial sync\n");
4150 skip_initial_sync = 1;
4151 } else if (device->state.conn != C_STANDALONE) {
4152 retcode = ERR_CONNECTED;
4156 drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
4157 drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
4159 if (args.clear_bm) {
4160 err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
4161 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
4163 drbd_err(device, "Writing bitmap failed with %d\n", err);
4164 retcode = ERR_IO_MD_DISK;
4166 if (skip_initial_sync) {
4167 drbd_send_uuids_skip_initial_sync(first_peer_device(device));
4168 _drbd_uuid_set(device, UI_BITMAP, 0);
4169 drbd_print_uuids(device, "cleared bitmap UUID");
4170 spin_lock_irq(&device->resource->req_lock);
4171 _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
4173 spin_unlock_irq(&device->resource->req_lock);
4177 drbd_md_sync(device);
4181 mutex_unlock(device->state_mutex);
4182 mutex_unlock(&adm_ctx.resource->adm_mutex);
4184 drbd_adm_finish(&adm_ctx, info, retcode);
4188 static enum drbd_ret_code
4189 drbd_check_resource_name(struct drbd_config_context *adm_ctx)
4191 const char *name = adm_ctx->resource_name;
4192 if (!name || !name[0]) {
4193 drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing");
4194 return ERR_MANDATORY_TAG;
4196 /* if we want to use these in sysfs/configfs/debugfs some day,
4197 * we must not allow slashes */
4198 if (strchr(name, '/')) {
4199 drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name");
4200 return ERR_INVALID_REQUEST;
4205 static void resource_to_info(struct resource_info *info,
4206 struct drbd_resource *resource)
4208 info->res_role = conn_highest_role(first_connection(resource));
4209 info->res_susp = resource->susp;
4210 info->res_susp_nod = resource->susp_nod;
4211 info->res_susp_fen = resource->susp_fen;
4214 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
4216 struct drbd_connection *connection;
4217 struct drbd_config_context adm_ctx;
4218 enum drbd_ret_code retcode;
4219 struct res_opts res_opts;
4222 retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0);
4223 if (!adm_ctx.reply_skb)
4225 if (retcode != NO_ERROR)
4228 set_res_opts_defaults(&res_opts);
4229 err = res_opts_from_attrs(&res_opts, info);
4230 if (err && err != -ENOMSG) {
4231 retcode = ERR_MANDATORY_TAG;
4232 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4236 retcode = drbd_check_resource_name(&adm_ctx);
4237 if (retcode != NO_ERROR)
4240 if (adm_ctx.resource) {
4241 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
4242 retcode = ERR_INVALID_REQUEST;
4243 drbd_msg_put_info(adm_ctx.reply_skb, "resource exists");
4245 /* else: still NO_ERROR */
4249 /* not yet safe for genl_family.parallel_ops */
4250 mutex_lock(&resources_mutex);
4251 connection = conn_create(adm_ctx.resource_name, &res_opts);
4252 mutex_unlock(&resources_mutex);
4255 struct resource_info resource_info;
4257 mutex_lock(¬ification_mutex);
4258 resource_to_info(&resource_info, connection->resource);
4259 notify_resource_state(NULL, 0, connection->resource,
4260 &resource_info, NOTIFY_CREATE);
4261 mutex_unlock(¬ification_mutex);
4263 retcode = ERR_NOMEM;
4266 drbd_adm_finish(&adm_ctx, info, retcode);
4270 static void device_to_info(struct device_info *info,
4271 struct drbd_device *device)
4273 info->dev_disk_state = device->state.disk;
4277 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info)
4279 struct drbd_config_context adm_ctx;
4280 struct drbd_genlmsghdr *dh = info->userhdr;
4281 enum drbd_ret_code retcode;
4283 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4284 if (!adm_ctx.reply_skb)
4286 if (retcode != NO_ERROR)
4289 if (dh->minor > MINORMASK) {
4290 drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range");
4291 retcode = ERR_INVALID_REQUEST;
4294 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
4295 drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range");
4296 retcode = ERR_INVALID_REQUEST;
4300 /* drbd_adm_prepare made sure already
4301 * that first_peer_device(device)->connection and device->vnr match the request. */
4302 if (adm_ctx.device) {
4303 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
4304 retcode = ERR_MINOR_OR_VOLUME_EXISTS;
4305 /* else: still NO_ERROR */
4309 mutex_lock(&adm_ctx.resource->adm_mutex);
4310 retcode = drbd_create_device(&adm_ctx, dh->minor);
4311 if (retcode == NO_ERROR) {
4312 struct drbd_device *device;
4313 struct drbd_peer_device *peer_device;
4314 struct device_info info;
4315 unsigned int peer_devices = 0;
4316 enum drbd_notification_type flags;
4318 device = minor_to_device(dh->minor);
4319 for_each_peer_device(peer_device, device) {
4320 if (!has_net_conf(peer_device->connection))
4325 device_to_info(&info, device);
4326 mutex_lock(¬ification_mutex);
4327 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4328 notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags);
4329 for_each_peer_device(peer_device, device) {
4330 struct peer_device_info peer_device_info;
4332 if (!has_net_conf(peer_device->connection))
4334 peer_device_to_info(&peer_device_info, peer_device);
4335 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4336 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info,
4337 NOTIFY_CREATE | flags);
4339 mutex_unlock(¬ification_mutex);
4341 mutex_unlock(&adm_ctx.resource->adm_mutex);
4343 drbd_adm_finish(&adm_ctx, info, retcode);
4347 static enum drbd_ret_code adm_del_minor(struct drbd_device *device)
4349 struct drbd_peer_device *peer_device;
4351 if (device->state.disk == D_DISKLESS &&
4352 /* no need to be device->state.conn == C_STANDALONE &&
4353 * we may want to delete a minor from a live replication group.
4355 device->state.role == R_SECONDARY) {
4356 struct drbd_connection *connection =
4357 first_connection(device->resource);
4359 _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS),
4360 CS_VERBOSE + CS_WAIT_COMPLETE);
4362 /* If the state engine hasn't stopped the sender thread yet, we
4363 * need to flush the sender work queue before generating the
4364 * DESTROY events here. */
4365 if (get_t_state(&connection->worker) == RUNNING)
4366 drbd_flush_workqueue(&connection->sender_work);
4368 mutex_lock(¬ification_mutex);
4369 for_each_peer_device(peer_device, device) {
4370 if (!has_net_conf(peer_device->connection))
4372 notify_peer_device_state(NULL, 0, peer_device, NULL,
4373 NOTIFY_DESTROY | NOTIFY_CONTINUES);
4375 notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY);
4376 mutex_unlock(¬ification_mutex);
4378 drbd_delete_device(device);
4381 return ERR_MINOR_CONFIGURED;
4384 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info)
4386 struct drbd_config_context adm_ctx;
4387 enum drbd_ret_code retcode;
4389 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4390 if (!adm_ctx.reply_skb)
4392 if (retcode != NO_ERROR)
4395 mutex_lock(&adm_ctx.resource->adm_mutex);
4396 retcode = adm_del_minor(adm_ctx.device);
4397 mutex_unlock(&adm_ctx.resource->adm_mutex);
4399 drbd_adm_finish(&adm_ctx, info, retcode);
4403 static int adm_del_resource(struct drbd_resource *resource)
4405 struct drbd_connection *connection;
4407 for_each_connection(connection, resource) {
4408 if (connection->cstate > C_STANDALONE)
4409 return ERR_NET_CONFIGURED;
4411 if (!idr_is_empty(&resource->devices))
4412 return ERR_RES_IN_USE;
4414 /* The state engine has stopped the sender thread, so we don't
4415 * need to flush the sender work queue before generating the
4416 * DESTROY event here. */
4417 mutex_lock(¬ification_mutex);
4418 notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY);
4419 mutex_unlock(¬ification_mutex);
4421 mutex_lock(&resources_mutex);
4422 list_del_rcu(&resource->resources);
4423 mutex_unlock(&resources_mutex);
4424 /* Make sure all threads have actually stopped: state handling only
4425 * does drbd_thread_stop_nowait(). */
4426 list_for_each_entry(connection, &resource->connections, connections)
4427 drbd_thread_stop(&connection->worker);
4429 drbd_free_resource(resource);
4433 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
4435 struct drbd_config_context adm_ctx;
4436 struct drbd_resource *resource;
4437 struct drbd_connection *connection;
4438 struct drbd_device *device;
4439 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
4442 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4443 if (!adm_ctx.reply_skb)
4445 if (retcode != NO_ERROR)
4448 resource = adm_ctx.resource;
4449 mutex_lock(&resource->adm_mutex);
4451 for_each_connection(connection, resource) {
4452 struct drbd_peer_device *peer_device;
4454 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
4455 retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0);
4456 if (retcode < SS_SUCCESS) {
4457 drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote");
4462 retcode = conn_try_disconnect(connection, 0);
4463 if (retcode < SS_SUCCESS) {
4464 drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect");
4470 idr_for_each_entry(&resource->devices, device, i) {
4471 retcode = adm_detach(device, 0);
4472 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
4473 drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach");
4478 /* delete volumes */
4479 idr_for_each_entry(&resource->devices, device, i) {
4480 retcode = adm_del_minor(device);
4481 if (retcode != NO_ERROR) {
4482 /* "can not happen" */
4483 drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume");
4488 retcode = adm_del_resource(resource);
4490 mutex_unlock(&resource->adm_mutex);
4492 drbd_adm_finish(&adm_ctx, info, retcode);
4496 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
4498 struct drbd_config_context adm_ctx;
4499 struct drbd_resource *resource;
4500 enum drbd_ret_code retcode;
4502 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4503 if (!adm_ctx.reply_skb)
4505 if (retcode != NO_ERROR)
4507 resource = adm_ctx.resource;
4509 mutex_lock(&resource->adm_mutex);
4510 retcode = adm_del_resource(resource);
4511 mutex_unlock(&resource->adm_mutex);
4513 drbd_adm_finish(&adm_ctx, info, retcode);
4517 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib)
4519 struct sk_buff *msg;
4520 struct drbd_genlmsghdr *d_out;
4524 seq = atomic_inc_return(&drbd_genl_seq);
4525 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4530 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
4531 if (!d_out) /* cannot happen, but anyways. */
4532 goto nla_put_failure;
4533 d_out->minor = device_to_minor(device);
4534 d_out->ret_code = NO_ERROR;
4536 if (nla_put_status_info(msg, device, sib))
4537 goto nla_put_failure;
4538 genlmsg_end(msg, d_out);
4539 err = drbd_genl_multicast_events(msg, GFP_NOWAIT);
4540 /* msg has been consumed or freed in netlink_broadcast() */
4541 if (err && err != -ESRCH)
4549 drbd_err(device, "Error %d while broadcasting event. "
4550 "Event seq:%u sib_reason:%u\n",
4551 err, seq, sib->sib_reason);
4554 static int nla_put_notification_header(struct sk_buff *msg,
4555 enum drbd_notification_type type)
4557 struct drbd_notification_header nh = {
4561 return drbd_notification_header_to_skb(msg, &nh, true);
4564 void notify_resource_state(struct sk_buff *skb,
4566 struct drbd_resource *resource,
4567 struct resource_info *resource_info,
4568 enum drbd_notification_type type)
4570 struct resource_statistics resource_statistics;
4571 struct drbd_genlmsghdr *dh;
4572 bool multicast = false;
4576 seq = atomic_inc_return(¬ify_genl_seq);
4577 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4585 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_RESOURCE_STATE);
4587 goto nla_put_failure;
4589 dh->ret_code = NO_ERROR;
4590 if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) ||
4591 nla_put_notification_header(skb, type) ||
4592 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4593 resource_info_to_skb(skb, resource_info, true)))
4594 goto nla_put_failure;
4595 resource_statistics.res_stat_write_ordering = resource->write_ordering;
4596 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
4598 goto nla_put_failure;
4599 genlmsg_end(skb, dh);
4601 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4602 /* skb has been consumed or freed in netlink_broadcast() */
4603 if (err && err != -ESRCH)
4611 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4615 void notify_device_state(struct sk_buff *skb,
4617 struct drbd_device *device,
4618 struct device_info *device_info,
4619 enum drbd_notification_type type)
4621 struct device_statistics device_statistics;
4622 struct drbd_genlmsghdr *dh;
4623 bool multicast = false;
4627 seq = atomic_inc_return(¬ify_genl_seq);
4628 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4636 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_DEVICE_STATE);
4638 goto nla_put_failure;
4639 dh->minor = device->minor;
4640 dh->ret_code = NO_ERROR;
4641 if (nla_put_drbd_cfg_context(skb, device->resource, NULL, device) ||
4642 nla_put_notification_header(skb, type) ||
4643 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4644 device_info_to_skb(skb, device_info, true)))
4645 goto nla_put_failure;
4646 device_to_statistics(&device_statistics, device);
4647 device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
4648 genlmsg_end(skb, dh);
4650 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4651 /* skb has been consumed or freed in netlink_broadcast() */
4652 if (err && err != -ESRCH)
4660 drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n",
4664 void notify_connection_state(struct sk_buff *skb,
4666 struct drbd_connection *connection,
4667 struct connection_info *connection_info,
4668 enum drbd_notification_type type)
4670 struct connection_statistics connection_statistics;
4671 struct drbd_genlmsghdr *dh;
4672 bool multicast = false;
4676 seq = atomic_inc_return(¬ify_genl_seq);
4677 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4685 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_CONNECTION_STATE);
4687 goto nla_put_failure;
4689 dh->ret_code = NO_ERROR;
4690 if (nla_put_drbd_cfg_context(skb, connection->resource, connection, NULL) ||
4691 nla_put_notification_header(skb, type) ||
4692 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4693 connection_info_to_skb(skb, connection_info, true)))
4694 goto nla_put_failure;
4695 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
4696 connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
4697 genlmsg_end(skb, dh);
4699 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4700 /* skb has been consumed or freed in netlink_broadcast() */
4701 if (err && err != -ESRCH)
4709 drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n",
4713 void notify_peer_device_state(struct sk_buff *skb,
4715 struct drbd_peer_device *peer_device,
4716 struct peer_device_info *peer_device_info,
4717 enum drbd_notification_type type)
4719 struct peer_device_statistics peer_device_statistics;
4720 struct drbd_resource *resource = peer_device->device->resource;
4721 struct drbd_genlmsghdr *dh;
4722 bool multicast = false;
4726 seq = atomic_inc_return(¬ify_genl_seq);
4727 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4735 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_PEER_DEVICE_STATE);
4737 goto nla_put_failure;
4739 dh->ret_code = NO_ERROR;
4740 if (nla_put_drbd_cfg_context(skb, resource, peer_device->connection, peer_device->device) ||
4741 nla_put_notification_header(skb, type) ||
4742 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4743 peer_device_info_to_skb(skb, peer_device_info, true)))
4744 goto nla_put_failure;
4745 peer_device_to_statistics(&peer_device_statistics, peer_device);
4746 peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
4747 genlmsg_end(skb, dh);
4749 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4750 /* skb has been consumed or freed in netlink_broadcast() */
4751 if (err && err != -ESRCH)
4759 drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n",
4763 void notify_helper(enum drbd_notification_type type,
4764 struct drbd_device *device, struct drbd_connection *connection,
4765 const char *name, int status)
4767 struct drbd_resource *resource = device ? device->resource : connection->resource;
4768 struct drbd_helper_info helper_info;
4769 unsigned int seq = atomic_inc_return(¬ify_genl_seq);
4770 struct sk_buff *skb = NULL;
4771 struct drbd_genlmsghdr *dh;
4774 strlcpy(helper_info.helper_name, name, sizeof(helper_info.helper_name));
4775 helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name));
4776 helper_info.helper_status = status;
4778 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4784 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_HELPER);
4787 dh->minor = device ? device->minor : -1;
4788 dh->ret_code = NO_ERROR;
4789 mutex_lock(¬ification_mutex);
4790 if (nla_put_drbd_cfg_context(skb, resource, connection, device) ||
4791 nla_put_notification_header(skb, type) ||
4792 drbd_helper_info_to_skb(skb, &helper_info, true))
4794 genlmsg_end(skb, dh);
4795 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4797 /* skb has been consumed or freed in netlink_broadcast() */
4798 if (err && err != -ESRCH)
4800 mutex_unlock(¬ification_mutex);
4804 mutex_unlock(¬ification_mutex);
4807 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4811 static void notify_initial_state_done(struct sk_buff *skb, unsigned int seq)
4813 struct drbd_genlmsghdr *dh;
4817 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_INITIAL_STATE_DONE);
4819 goto nla_put_failure;
4821 dh->ret_code = NO_ERROR;
4822 if (nla_put_notification_header(skb, NOTIFY_EXISTS))
4823 goto nla_put_failure;
4824 genlmsg_end(skb, dh);
4829 pr_err("Error %d sending event. Event seq:%u\n", err, seq);
4832 static void free_state_changes(struct list_head *list)
4834 while (!list_empty(list)) {
4835 struct drbd_state_change *state_change =
4836 list_first_entry(list, struct drbd_state_change, list);
4837 list_del(&state_change->list);
4838 forget_state_change(state_change);
4842 static unsigned int notifications_for_state_change(struct drbd_state_change *state_change)
4845 state_change->n_connections +
4846 state_change->n_devices +
4847 state_change->n_devices * state_change->n_connections;
4850 static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4852 struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0];
4853 unsigned int seq = cb->args[2];
4855 enum drbd_notification_type flags = 0;
4857 /* There is no need for taking notification_mutex here: it doesn't
4858 matter if the initial state events mix with later state chage
4859 events; we can always tell the events apart by the NOTIFY_EXISTS
4863 if (cb->args[5] == 1) {
4864 notify_initial_state_done(skb, seq);
4868 if (cb->args[4] < cb->args[3])
4869 flags |= NOTIFY_CONTINUES;
4871 notify_resource_state_change(skb, seq, state_change->resource,
4872 NOTIFY_EXISTS | flags);
4876 if (n < state_change->n_connections) {
4877 notify_connection_state_change(skb, seq, &state_change->connections[n],
4878 NOTIFY_EXISTS | flags);
4881 n -= state_change->n_connections;
4882 if (n < state_change->n_devices) {
4883 notify_device_state_change(skb, seq, &state_change->devices[n],
4884 NOTIFY_EXISTS | flags);
4887 n -= state_change->n_devices;
4888 if (n < state_change->n_devices * state_change->n_connections) {
4889 notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n],
4890 NOTIFY_EXISTS | flags);
4895 if (cb->args[4] == cb->args[3]) {
4896 struct drbd_state_change *next_state_change =
4897 list_entry(state_change->list.next,
4898 struct drbd_state_change, list);
4899 cb->args[0] = (long)next_state_change;
4900 cb->args[3] = notifications_for_state_change(next_state_change);
4907 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4909 struct drbd_resource *resource;
4912 if (cb->args[5] >= 1) {
4913 if (cb->args[5] > 1)
4914 return get_initial_state(skb, cb);
4916 struct drbd_state_change *state_change =
4917 (struct drbd_state_change *)cb->args[0];
4919 /* connect list to head */
4920 list_add(&head, &state_change->list);
4921 free_state_changes(&head);
4926 cb->args[5] = 2; /* number of iterations */
4927 mutex_lock(&resources_mutex);
4928 for_each_resource(resource, &drbd_resources) {
4929 struct drbd_state_change *state_change;
4931 state_change = remember_old_state(resource, GFP_KERNEL);
4932 if (!state_change) {
4933 if (!list_empty(&head))
4934 free_state_changes(&head);
4935 mutex_unlock(&resources_mutex);
4938 copy_old_to_new_state_change(state_change);
4939 list_add_tail(&state_change->list, &head);
4940 cb->args[5] += notifications_for_state_change(state_change);
4942 mutex_unlock(&resources_mutex);
4944 if (!list_empty(&head)) {
4945 struct drbd_state_change *state_change =
4946 list_entry(head.next, struct drbd_state_change, list);
4947 cb->args[0] = (long)state_change;
4948 cb->args[3] = notifications_for_state_change(state_change);
4949 list_del(&head); /* detach list from head */
4952 cb->args[2] = cb->nlh->nlmsg_seq;
4953 return get_initial_state(skb, cb);