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[] = {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", 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(usermode_helper, argv, envp, UMH_WAIT_PROC);
369 drbd_warn(device, "helper command: %s %s %s exit code %u (0x%x)\n",
370 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 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 static int 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[] = {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", usermode_helper, cmd, resource_name);
406 /* TODO: conn_bcast_event() ?? */
407 notify_helper(NOTIFY_CALL, NULL, connection, cmd, 0);
409 ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
411 drbd_warn(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
412 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 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);
445 if (fp == FP_NOT_AVAIL) {
446 /* IO Suspending works on the whole resource.
447 Do it only for one device. */
449 peer_device = idr_get_next(&connection->peer_devices, &vnr);
450 drbd_change_state(peer_device->device, CS_VERBOSE | CS_HARD, NS(susp_fen, 0));
456 bool conn_try_outdate_peer(struct drbd_connection *connection)
458 unsigned int connect_cnt;
459 union drbd_state mask = { };
460 union drbd_state val = { };
461 enum drbd_fencing_p fp;
465 spin_lock_irq(&connection->resource->req_lock);
466 if (connection->cstate >= C_WF_REPORT_PARAMS) {
467 drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n");
468 spin_unlock_irq(&connection->resource->req_lock);
472 connect_cnt = connection->connect_cnt;
473 spin_unlock_irq(&connection->resource->req_lock);
475 fp = highest_fencing_policy(connection);
478 drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n");
485 r = conn_khelper(connection, "fence-peer");
487 switch ((r>>8) & 0xff) {
488 case 3: /* peer is inconsistent */
489 ex_to_string = "peer is inconsistent or worse";
491 val.pdsk = D_INCONSISTENT;
493 case 4: /* peer got outdated, or was already outdated */
494 ex_to_string = "peer was fenced";
496 val.pdsk = D_OUTDATED;
498 case 5: /* peer was down */
499 if (conn_highest_disk(connection) == D_UP_TO_DATE) {
500 /* we will(have) create(d) a new UUID anyways... */
501 ex_to_string = "peer is unreachable, assumed to be dead";
503 val.pdsk = D_OUTDATED;
505 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
508 case 6: /* Peer is primary, voluntarily outdate myself.
509 * This is useful when an unconnected R_SECONDARY is asked to
510 * become R_PRIMARY, but finds the other peer being active. */
511 ex_to_string = "peer is active";
512 drbd_warn(connection, "Peer is primary, outdating myself.\n");
514 val.disk = D_OUTDATED;
517 if (fp != FP_STONITH)
518 drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n");
519 ex_to_string = "peer was stonithed";
521 val.pdsk = D_OUTDATED;
524 /* The script is broken ... */
525 drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
526 return false; /* Eventually leave IO frozen */
529 drbd_info(connection, "fence-peer helper returned %d (%s)\n",
530 (r>>8) & 0xff, ex_to_string);
535 conn_request_state(connection, mask, val, CS_VERBOSE);
536 here, because we might were able to re-establish the connection in the
538 spin_lock_irq(&connection->resource->req_lock);
539 if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) {
540 if (connection->connect_cnt != connect_cnt)
541 /* In case the connection was established and droped
542 while the fence-peer handler was running, ignore it */
543 drbd_info(connection, "Ignoring fence-peer exit code\n");
545 _conn_request_state(connection, mask, val, CS_VERBOSE);
547 spin_unlock_irq(&connection->resource->req_lock);
549 return conn_highest_pdsk(connection) <= D_OUTDATED;
552 static int _try_outdate_peer_async(void *data)
554 struct drbd_connection *connection = (struct drbd_connection *)data;
556 conn_try_outdate_peer(connection);
558 kref_put(&connection->kref, drbd_destroy_connection);
562 void conn_try_outdate_peer_async(struct drbd_connection *connection)
564 struct task_struct *opa;
566 kref_get(&connection->kref);
567 /* We may just have force_sig()'ed this thread
568 * to get it out of some blocking network function.
569 * Clear signals; otherwise kthread_run(), which internally uses
570 * wait_on_completion_killable(), will mistake our pending signal
571 * for a new fatal signal and fail. */
572 flush_signals(current);
573 opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h");
575 drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n");
576 kref_put(&connection->kref, drbd_destroy_connection);
581 drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force)
583 struct drbd_peer_device *const peer_device = first_peer_device(device);
584 struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL;
585 const int max_tries = 4;
586 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
590 union drbd_state mask, val;
592 if (new_role == R_PRIMARY) {
593 struct drbd_connection *connection;
595 /* Detect dead peers as soon as possible. */
598 for_each_connection(connection, device->resource)
599 request_ping(connection);
603 mutex_lock(device->state_mutex);
605 mask.i = 0; mask.role = R_MASK;
606 val.i = 0; val.role = new_role;
608 while (try++ < max_tries) {
609 rv = _drbd_request_state_holding_state_mutex(device, mask, val, CS_WAIT_COMPLETE);
611 /* in case we first succeeded to outdate,
612 * but now suddenly could establish a connection */
613 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
619 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
620 (device->state.disk < D_UP_TO_DATE &&
621 device->state.disk >= D_INCONSISTENT)) {
623 val.disk = D_UP_TO_DATE;
628 if (rv == SS_NO_UP_TO_DATE_DISK &&
629 device->state.disk == D_CONSISTENT && mask.pdsk == 0) {
630 D_ASSERT(device, device->state.pdsk == D_UNKNOWN);
632 if (conn_try_outdate_peer(connection)) {
633 val.disk = D_UP_TO_DATE;
639 if (rv == SS_NOTHING_TO_DO)
641 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
642 if (!conn_try_outdate_peer(connection) && force) {
643 drbd_warn(device, "Forced into split brain situation!\n");
645 val.pdsk = D_OUTDATED;
650 if (rv == SS_TWO_PRIMARIES) {
651 /* Maybe the peer is detected as dead very soon...
652 retry at most once more in this case. */
655 nc = rcu_dereference(connection->net_conf);
656 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
658 schedule_timeout_interruptible(timeo);
663 if (rv < SS_SUCCESS) {
664 rv = _drbd_request_state(device, mask, val,
665 CS_VERBOSE + CS_WAIT_COMPLETE);
676 drbd_warn(device, "Forced to consider local data as UpToDate!\n");
678 /* Wait until nothing is on the fly :) */
679 wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0);
681 /* FIXME also wait for all pending P_BARRIER_ACK? */
683 if (new_role == R_SECONDARY) {
684 if (get_ldev(device)) {
685 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
689 mutex_lock(&device->resource->conf_update);
690 nc = connection->net_conf;
692 nc->discard_my_data = 0; /* without copy; single bit op is atomic */
693 mutex_unlock(&device->resource->conf_update);
695 if (get_ldev(device)) {
696 if (((device->state.conn < C_CONNECTED ||
697 device->state.pdsk <= D_FAILED)
698 && device->ldev->md.uuid[UI_BITMAP] == 0) || forced)
699 drbd_uuid_new_current(device);
701 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
706 /* writeout of activity log covered areas of the bitmap
707 * to stable storage done in after state change already */
709 if (device->state.conn >= C_WF_REPORT_PARAMS) {
710 /* if this was forced, we should consider sync */
712 drbd_send_uuids(peer_device);
713 drbd_send_current_state(peer_device);
716 drbd_md_sync(device);
717 set_disk_ro(device->vdisk, new_role == R_SECONDARY);
718 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
720 mutex_unlock(device->state_mutex);
724 static const char *from_attrs_err_to_txt(int err)
726 return err == -ENOMSG ? "required attribute missing" :
727 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
728 err == -EEXIST ? "can not change invariant setting" :
729 "invalid attribute value";
732 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
734 struct drbd_config_context adm_ctx;
735 struct set_role_parms parms;
737 enum drbd_ret_code retcode;
739 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
740 if (!adm_ctx.reply_skb)
742 if (retcode != NO_ERROR)
745 memset(&parms, 0, sizeof(parms));
746 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
747 err = set_role_parms_from_attrs(&parms, info);
749 retcode = ERR_MANDATORY_TAG;
750 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
755 mutex_lock(&adm_ctx.resource->adm_mutex);
757 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
758 retcode = drbd_set_role(adm_ctx.device, R_PRIMARY, parms.assume_uptodate);
760 retcode = drbd_set_role(adm_ctx.device, R_SECONDARY, 0);
762 mutex_unlock(&adm_ctx.resource->adm_mutex);
765 drbd_adm_finish(&adm_ctx, info, retcode);
769 /* Initializes the md.*_offset members, so we are able to find
770 * the on disk meta data.
772 * We currently have two possible layouts:
774 * |----------- md_size_sect ------------------|
775 * [ 4k superblock ][ activity log ][ Bitmap ]
777 * | bm_offset = al_offset + X |
778 * ==> bitmap sectors = md_size_sect - bm_offset
781 * |----------- md_size_sect ------------------|
782 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
784 * | bm_offset = al_offset - Y |
785 * ==> bitmap sectors = Y = al_offset - bm_offset
787 * Activity log size used to be fixed 32kB,
788 * but is about to become configurable.
790 static void drbd_md_set_sector_offsets(struct drbd_device *device,
791 struct drbd_backing_dev *bdev)
793 sector_t md_size_sect = 0;
794 unsigned int al_size_sect = bdev->md.al_size_4k * 8;
796 bdev->md.md_offset = drbd_md_ss(bdev);
798 switch (bdev->md.meta_dev_idx) {
800 /* v07 style fixed size indexed meta data */
801 bdev->md.md_size_sect = MD_128MB_SECT;
802 bdev->md.al_offset = MD_4kB_SECT;
803 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
805 case DRBD_MD_INDEX_FLEX_EXT:
806 /* just occupy the full device; unit: sectors */
807 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
808 bdev->md.al_offset = MD_4kB_SECT;
809 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
811 case DRBD_MD_INDEX_INTERNAL:
812 case DRBD_MD_INDEX_FLEX_INT:
813 /* al size is still fixed */
814 bdev->md.al_offset = -al_size_sect;
815 /* we need (slightly less than) ~ this much bitmap sectors: */
816 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
817 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
818 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
819 md_size_sect = ALIGN(md_size_sect, 8);
821 /* plus the "drbd meta data super block",
822 * and the activity log; */
823 md_size_sect += MD_4kB_SECT + al_size_sect;
825 bdev->md.md_size_sect = md_size_sect;
826 /* bitmap offset is adjusted by 'super' block size */
827 bdev->md.bm_offset = -md_size_sect + MD_4kB_SECT;
832 /* input size is expected to be in KB */
833 char *ppsize(char *buf, unsigned long long size)
835 /* Needs 9 bytes at max including trailing NUL:
836 * -1ULL ==> "16384 EB" */
837 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
839 while (size >= 10000 && base < sizeof(units)-1) {
841 size = (size >> 10) + !!(size & (1<<9));
844 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
849 /* there is still a theoretical deadlock when called from receiver
850 * on an D_INCONSISTENT R_PRIMARY:
851 * remote READ does inc_ap_bio, receiver would need to receive answer
852 * packet from remote to dec_ap_bio again.
853 * receiver receive_sizes(), comes here,
854 * waits for ap_bio_cnt == 0. -> deadlock.
855 * but this cannot happen, actually, because:
856 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
857 * (not connected, or bad/no disk on peer):
858 * see drbd_fail_request_early, ap_bio_cnt is zero.
859 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
860 * peer may not initiate a resize.
862 /* Note these are not to be confused with
863 * drbd_adm_suspend_io/drbd_adm_resume_io,
864 * which are (sub) state changes triggered by admin (drbdsetup),
865 * and can be long lived.
866 * This changes an device->flag, is triggered by drbd internals,
867 * and should be short-lived. */
868 /* It needs to be a counter, since multiple threads might
869 independently suspend and resume IO. */
870 void drbd_suspend_io(struct drbd_device *device)
872 atomic_inc(&device->suspend_cnt);
873 if (drbd_suspended(device))
875 wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt));
878 void drbd_resume_io(struct drbd_device *device)
880 if (atomic_dec_and_test(&device->suspend_cnt))
881 wake_up(&device->misc_wait);
885 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
886 * @device: DRBD device.
888 * Returns 0 on success, negative return values indicate errors.
889 * You should call drbd_md_sync() after calling this function.
891 enum determine_dev_size
892 drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local)
894 struct md_offsets_and_sizes {
895 u64 last_agreed_sect;
902 u32 al_stripe_size_4k;
904 sector_t u_size, size;
905 struct drbd_md *md = &device->ldev->md;
909 int md_moved, la_size_changed;
910 enum determine_dev_size rv = DS_UNCHANGED;
912 /* We may change the on-disk offsets of our meta data below. Lock out
913 * anything that may cause meta data IO, to avoid acting on incomplete
914 * layout changes or scribbling over meta data that is in the process
917 * Move is not exactly correct, btw, currently we have all our meta
918 * data in core memory, to "move" it we just write it all out, there
920 drbd_suspend_io(device);
921 buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */
923 drbd_resume_io(device);
927 /* remember current offset and sizes */
928 prev.last_agreed_sect = md->la_size_sect;
929 prev.md_offset = md->md_offset;
930 prev.al_offset = md->al_offset;
931 prev.bm_offset = md->bm_offset;
932 prev.md_size_sect = md->md_size_sect;
933 prev.al_stripes = md->al_stripes;
934 prev.al_stripe_size_4k = md->al_stripe_size_4k;
937 /* rs is non NULL if we should change the AL layout only */
938 md->al_stripes = rs->al_stripes;
939 md->al_stripe_size_4k = rs->al_stripe_size / 4;
940 md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4;
943 drbd_md_set_sector_offsets(device, device->ldev);
946 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
948 size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED);
950 if (size < prev.last_agreed_sect) {
951 if (rs && u_size == 0) {
952 /* Remove "rs &&" later. This check should always be active, but
953 right now the receiver expects the permissive behavior */
954 drbd_warn(device, "Implicit shrink not allowed. "
955 "Use --size=%llus for explicit shrink.\n",
956 (unsigned long long)size);
957 rv = DS_ERROR_SHRINK;
960 rv = DS_ERROR_SPACE_MD;
961 if (rv != DS_UNCHANGED)
965 if (drbd_get_capacity(device->this_bdev) != size ||
966 drbd_bm_capacity(device) != size) {
968 err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC));
970 /* currently there is only one error: ENOMEM! */
971 size = drbd_bm_capacity(device);
973 drbd_err(device, "OUT OF MEMORY! "
974 "Could not allocate bitmap!\n");
976 drbd_err(device, "BM resizing failed. "
977 "Leaving size unchanged\n");
981 /* racy, see comments above. */
982 drbd_set_my_capacity(device, size);
983 md->la_size_sect = size;
984 drbd_info(device, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
985 (unsigned long long)size>>1);
990 la_size_changed = (prev.last_agreed_sect != md->la_size_sect);
992 md_moved = prev.md_offset != md->md_offset
993 || prev.md_size_sect != md->md_size_sect;
995 if (la_size_changed || md_moved || rs) {
998 /* We do some synchronous IO below, which may take some time.
999 * Clear the timer, to avoid scary "timer expired!" messages,
1000 * "Superblock" is written out at least twice below, anyways. */
1001 del_timer(&device->md_sync_timer);
1003 /* We won't change the "al-extents" setting, we just may need
1004 * to move the on-disk location of the activity log ringbuffer.
1005 * Lock for transaction is good enough, it may well be "dirty"
1006 * or even "starving". */
1007 wait_event(device->al_wait, lc_try_lock_for_transaction(device->act_log));
1009 /* mark current on-disk bitmap and activity log as unreliable */
1010 prev_flags = md->flags;
1011 md->flags |= MDF_FULL_SYNC | MDF_AL_DISABLED;
1012 drbd_md_write(device, buffer);
1014 drbd_al_initialize(device, buffer);
1016 drbd_info(device, "Writing the whole bitmap, %s\n",
1017 la_size_changed && md_moved ? "size changed and md moved" :
1018 la_size_changed ? "size changed" : "md moved");
1019 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
1020 drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
1021 "size changed", BM_LOCKED_MASK);
1023 /* on-disk bitmap and activity log is authoritative again
1024 * (unless there was an IO error meanwhile...) */
1025 md->flags = prev_flags;
1026 drbd_md_write(device, buffer);
1029 drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
1030 md->al_stripes, md->al_stripe_size_4k * 4);
1033 if (size > prev.last_agreed_sect)
1034 rv = prev.last_agreed_sect ? DS_GREW : DS_GREW_FROM_ZERO;
1035 if (size < prev.last_agreed_sect)
1040 /* restore previous offset and sizes */
1041 md->la_size_sect = prev.last_agreed_sect;
1042 md->md_offset = prev.md_offset;
1043 md->al_offset = prev.al_offset;
1044 md->bm_offset = prev.bm_offset;
1045 md->md_size_sect = prev.md_size_sect;
1046 md->al_stripes = prev.al_stripes;
1047 md->al_stripe_size_4k = prev.al_stripe_size_4k;
1048 md->al_size_4k = (u64)prev.al_stripes * prev.al_stripe_size_4k;
1050 lc_unlock(device->act_log);
1051 wake_up(&device->al_wait);
1052 drbd_md_put_buffer(device);
1053 drbd_resume_io(device);
1059 drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev,
1060 sector_t u_size, int assume_peer_has_space)
1062 sector_t p_size = device->p_size; /* partner's disk size. */
1063 sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
1064 sector_t m_size; /* my size */
1067 m_size = drbd_get_max_capacity(bdev);
1069 if (device->state.conn < C_CONNECTED && assume_peer_has_space) {
1070 drbd_warn(device, "Resize while not connected was forced by the user!\n");
1074 if (p_size && m_size) {
1075 size = min_t(sector_t, p_size, m_size);
1078 size = la_size_sect;
1079 if (m_size && m_size < size)
1081 if (p_size && p_size < size)
1092 drbd_err(device, "Both nodes diskless!\n");
1096 drbd_err(device, "Requested disk size is too big (%lu > %lu)\n",
1097 (unsigned long)u_size>>1, (unsigned long)size>>1);
1106 * drbd_check_al_size() - Ensures that the AL is of the right size
1107 * @device: DRBD device.
1109 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1110 * failed, and 0 on success. You should call drbd_md_sync() after you called
1113 static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1115 struct lru_cache *n, *t;
1116 struct lc_element *e;
1117 unsigned int in_use;
1120 if (device->act_log &&
1121 device->act_log->nr_elements == dc->al_extents)
1125 t = device->act_log;
1126 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
1127 dc->al_extents, sizeof(struct lc_element), 0);
1130 drbd_err(device, "Cannot allocate act_log lru!\n");
1133 spin_lock_irq(&device->al_lock);
1135 for (i = 0; i < t->nr_elements; i++) {
1136 e = lc_element_by_index(t, i);
1138 drbd_err(device, "refcnt(%d)==%d\n",
1139 e->lc_number, e->refcnt);
1140 in_use += e->refcnt;
1144 device->act_log = n;
1145 spin_unlock_irq(&device->al_lock);
1147 drbd_err(device, "Activity log still in use!\n");
1153 drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */
1157 static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev,
1158 unsigned int max_bio_size)
1160 struct request_queue * const q = device->rq_queue;
1161 unsigned int max_hw_sectors = max_bio_size >> 9;
1162 unsigned int max_segments = 0;
1163 struct request_queue *b = NULL;
1164 struct disk_conf *dc;
1165 bool discard_zeroes_if_aligned = true;
1168 b = bdev->backing_bdev->bd_disk->queue;
1170 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1172 dc = rcu_dereference(device->ldev->disk_conf);
1173 max_segments = dc->max_bio_bvecs;
1174 discard_zeroes_if_aligned = dc->discard_zeroes_if_aligned;
1177 blk_set_stacking_limits(&q->limits);
1178 blk_queue_max_write_same_sectors(q, 0);
1181 blk_queue_logical_block_size(q, 512);
1182 blk_queue_max_hw_sectors(q, max_hw_sectors);
1183 /* This is the workaround for "bio would need to, but cannot, be split" */
1184 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1185 blk_queue_segment_boundary(q, PAGE_SIZE-1);
1188 struct drbd_connection *connection = first_peer_device(device)->connection;
1190 blk_queue_max_discard_sectors(q, DRBD_MAX_DISCARD_SECTORS);
1192 if (blk_queue_discard(b) && (b->limits.discard_zeroes_data || discard_zeroes_if_aligned) &&
1193 (connection->cstate < C_CONNECTED || connection->agreed_features & FF_TRIM)) {
1194 /* We don't care, stacking below should fix it for the local device.
1195 * Whether or not it is a suitable granularity on the remote device
1196 * is not our problem, really. If you care, you need to
1197 * use devices with similar topology on all peers. */
1198 q->limits.discard_granularity = 512;
1199 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1201 blk_queue_max_discard_sectors(q, 0);
1202 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1203 q->limits.discard_granularity = 0;
1206 blk_queue_stack_limits(q, b);
1208 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1209 drbd_info(device, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1210 q->backing_dev_info.ra_pages,
1211 b->backing_dev_info.ra_pages);
1212 q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1215 /* To avoid confusion, if this queue does not support discard, clear
1216 * max_discard_sectors, which is what lsblk -D reports to the user. */
1217 if (!blk_queue_discard(q)) {
1218 blk_queue_max_discard_sectors(q, 0);
1219 q->limits.discard_granularity = 0;
1223 void drbd_reconsider_queue_parameters(struct drbd_device *device, struct drbd_backing_dev *bdev)
1225 unsigned int now, new, local, peer;
1227 now = queue_max_hw_sectors(device->rq_queue) << 9;
1228 local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */
1229 peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */
1232 local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9;
1233 device->local_max_bio_size = local;
1235 local = min(local, DRBD_MAX_BIO_SIZE);
1237 /* We may ignore peer limits if the peer is modern enough.
1238 Because new from 8.3.8 onwards the peer can use multiple
1239 BIOs for a single peer_request */
1240 if (device->state.conn >= C_WF_REPORT_PARAMS) {
1241 if (first_peer_device(device)->connection->agreed_pro_version < 94)
1242 peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1243 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1244 else if (first_peer_device(device)->connection->agreed_pro_version == 94)
1245 peer = DRBD_MAX_SIZE_H80_PACKET;
1246 else if (first_peer_device(device)->connection->agreed_pro_version < 100)
1247 peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */
1249 peer = DRBD_MAX_BIO_SIZE;
1251 /* We may later detach and re-attach on a disconnected Primary.
1252 * Avoid this setting to jump back in that case.
1253 * We want to store what we know the peer DRBD can handle,
1254 * not what the peer IO backend can handle. */
1255 if (peer > device->peer_max_bio_size)
1256 device->peer_max_bio_size = peer;
1258 new = min(local, peer);
1260 if (device->state.role == R_PRIMARY && new < now)
1261 drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1264 drbd_info(device, "max BIO size = %u\n", new);
1266 drbd_setup_queue_param(device, bdev, new);
1269 /* Starts the worker thread */
1270 static void conn_reconfig_start(struct drbd_connection *connection)
1272 drbd_thread_start(&connection->worker);
1273 drbd_flush_workqueue(&connection->sender_work);
1276 /* if still unconfigured, stops worker again. */
1277 static void conn_reconfig_done(struct drbd_connection *connection)
1280 spin_lock_irq(&connection->resource->req_lock);
1281 stop_threads = conn_all_vols_unconf(connection) &&
1282 connection->cstate == C_STANDALONE;
1283 spin_unlock_irq(&connection->resource->req_lock);
1285 /* ack_receiver thread and ack_sender workqueue are implicitly
1286 * stopped by receiver in conn_disconnect() */
1287 drbd_thread_stop(&connection->receiver);
1288 drbd_thread_stop(&connection->worker);
1292 /* Make sure IO is suspended before calling this function(). */
1293 static void drbd_suspend_al(struct drbd_device *device)
1297 if (!lc_try_lock(device->act_log)) {
1298 drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1302 drbd_al_shrink(device);
1303 spin_lock_irq(&device->resource->req_lock);
1304 if (device->state.conn < C_CONNECTED)
1305 s = !test_and_set_bit(AL_SUSPENDED, &device->flags);
1306 spin_unlock_irq(&device->resource->req_lock);
1307 lc_unlock(device->act_log);
1310 drbd_info(device, "Suspended AL updates\n");
1314 static bool should_set_defaults(struct genl_info *info)
1316 unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1317 return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1320 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1322 /* This is limited by 16 bit "slot" numbers,
1323 * and by available on-disk context storage.
1325 * Also (u16)~0 is special (denotes a "free" extent).
1327 * One transaction occupies one 4kB on-disk block,
1328 * we have n such blocks in the on disk ring buffer,
1329 * the "current" transaction may fail (n-1),
1330 * and there is 919 slot numbers context information per transaction.
1332 * 72 transaction blocks amounts to more than 2**16 context slots,
1333 * so cap there first.
1335 const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1336 const unsigned int sufficient_on_disk =
1337 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1338 /AL_CONTEXT_PER_TRANSACTION;
1340 unsigned int al_size_4k = bdev->md.al_size_4k;
1342 if (al_size_4k > sufficient_on_disk)
1345 return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1348 static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1350 return a->disk_barrier != b->disk_barrier ||
1351 a->disk_flushes != b->disk_flushes ||
1352 a->disk_drain != b->disk_drain;
1355 static void sanitize_disk_conf(struct drbd_device *device, struct disk_conf *disk_conf,
1356 struct drbd_backing_dev *nbc)
1358 struct request_queue * const q = nbc->backing_bdev->bd_disk->queue;
1360 if (disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1361 disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1362 if (disk_conf->al_extents > drbd_al_extents_max(nbc))
1363 disk_conf->al_extents = drbd_al_extents_max(nbc);
1365 if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data) {
1366 disk_conf->rs_discard_granularity = 0; /* disable feature */
1367 drbd_info(device, "rs_discard_granularity feature disabled\n");
1370 if (disk_conf->rs_discard_granularity) {
1371 int orig_value = disk_conf->rs_discard_granularity;
1374 if (q->limits.discard_granularity > disk_conf->rs_discard_granularity)
1375 disk_conf->rs_discard_granularity = q->limits.discard_granularity;
1377 remainder = disk_conf->rs_discard_granularity % q->limits.discard_granularity;
1378 disk_conf->rs_discard_granularity += remainder;
1380 if (disk_conf->rs_discard_granularity > q->limits.max_discard_sectors << 9)
1381 disk_conf->rs_discard_granularity = q->limits.max_discard_sectors << 9;
1383 if (disk_conf->rs_discard_granularity != orig_value)
1384 drbd_info(device, "rs_discard_granularity changed to %d\n",
1385 disk_conf->rs_discard_granularity);
1389 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1391 struct drbd_config_context adm_ctx;
1392 enum drbd_ret_code retcode;
1393 struct drbd_device *device;
1394 struct disk_conf *new_disk_conf, *old_disk_conf;
1395 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1398 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1399 if (!adm_ctx.reply_skb)
1401 if (retcode != NO_ERROR)
1404 device = adm_ctx.device;
1405 mutex_lock(&adm_ctx.resource->adm_mutex);
1407 /* we also need a disk
1408 * to change the options on */
1409 if (!get_ldev(device)) {
1410 retcode = ERR_NO_DISK;
1414 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1415 if (!new_disk_conf) {
1416 retcode = ERR_NOMEM;
1420 mutex_lock(&device->resource->conf_update);
1421 old_disk_conf = device->ldev->disk_conf;
1422 *new_disk_conf = *old_disk_conf;
1423 if (should_set_defaults(info))
1424 set_disk_conf_defaults(new_disk_conf);
1426 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1427 if (err && err != -ENOMSG) {
1428 retcode = ERR_MANDATORY_TAG;
1429 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1433 if (!expect(new_disk_conf->resync_rate >= 1))
1434 new_disk_conf->resync_rate = 1;
1436 sanitize_disk_conf(device, new_disk_conf, device->ldev);
1438 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1439 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1441 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1442 if (fifo_size != device->rs_plan_s->size) {
1443 new_plan = fifo_alloc(fifo_size);
1445 drbd_err(device, "kmalloc of fifo_buffer failed");
1446 retcode = ERR_NOMEM;
1451 drbd_suspend_io(device);
1452 wait_event(device->al_wait, lc_try_lock(device->act_log));
1453 drbd_al_shrink(device);
1454 err = drbd_check_al_size(device, new_disk_conf);
1455 lc_unlock(device->act_log);
1456 wake_up(&device->al_wait);
1457 drbd_resume_io(device);
1460 retcode = ERR_NOMEM;
1464 lock_all_resources();
1465 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1466 if (retcode == NO_ERROR) {
1467 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1468 drbd_resync_after_changed(device);
1470 unlock_all_resources();
1472 if (retcode != NO_ERROR)
1476 old_plan = device->rs_plan_s;
1477 rcu_assign_pointer(device->rs_plan_s, new_plan);
1480 mutex_unlock(&device->resource->conf_update);
1482 if (new_disk_conf->al_updates)
1483 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1485 device->ldev->md.flags |= MDF_AL_DISABLED;
1487 if (new_disk_conf->md_flushes)
1488 clear_bit(MD_NO_FUA, &device->flags);
1490 set_bit(MD_NO_FUA, &device->flags);
1492 if (write_ordering_changed(old_disk_conf, new_disk_conf))
1493 drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH);
1495 if (old_disk_conf->discard_zeroes_if_aligned != new_disk_conf->discard_zeroes_if_aligned)
1496 drbd_reconsider_queue_parameters(device, device->ldev);
1498 drbd_md_sync(device);
1500 if (device->state.conn >= C_CONNECTED) {
1501 struct drbd_peer_device *peer_device;
1503 for_each_peer_device(peer_device, device)
1504 drbd_send_sync_param(peer_device);
1508 kfree(old_disk_conf);
1510 mod_timer(&device->request_timer, jiffies + HZ);
1514 mutex_unlock(&device->resource->conf_update);
1516 kfree(new_disk_conf);
1521 mutex_unlock(&adm_ctx.resource->adm_mutex);
1523 drbd_adm_finish(&adm_ctx, info, retcode);
1527 static struct block_device *open_backing_dev(struct drbd_device *device,
1528 const char *bdev_path, void *claim_ptr, bool do_bd_link)
1530 struct block_device *bdev;
1533 bdev = blkdev_get_by_path(bdev_path,
1534 FMODE_READ | FMODE_WRITE | FMODE_EXCL, claim_ptr);
1536 drbd_err(device, "open(\"%s\") failed with %ld\n",
1537 bdev_path, PTR_ERR(bdev));
1544 err = bd_link_disk_holder(bdev, device->vdisk);
1546 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1547 drbd_err(device, "bd_link_disk_holder(\"%s\", ...) failed with %d\n",
1549 bdev = ERR_PTR(err);
1554 static int open_backing_devices(struct drbd_device *device,
1555 struct disk_conf *new_disk_conf,
1556 struct drbd_backing_dev *nbc)
1558 struct block_device *bdev;
1560 bdev = open_backing_dev(device, new_disk_conf->backing_dev, device, true);
1562 return ERR_OPEN_DISK;
1563 nbc->backing_bdev = bdev;
1566 * meta_dev_idx >= 0: external fixed size, possibly multiple
1567 * drbd sharing one meta device. TODO in that case, paranoia
1568 * check that [md_bdev, meta_dev_idx] is not yet used by some
1569 * other drbd minor! (if you use drbd.conf + drbdadm, that
1570 * should check it for you already; but if you don't, or
1571 * someone fooled it, we need to double check here)
1573 bdev = open_backing_dev(device, new_disk_conf->meta_dev,
1574 /* claim ptr: device, if claimed exclusively; shared drbd_m_holder,
1575 * if potentially shared with other drbd minors */
1576 (new_disk_conf->meta_dev_idx < 0) ? (void*)device : (void*)drbd_m_holder,
1577 /* avoid double bd_claim_by_disk() for the same (source,target) tuple,
1578 * as would happen with internal metadata. */
1579 (new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_FLEX_INT &&
1580 new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_INTERNAL));
1582 return ERR_OPEN_MD_DISK;
1583 nbc->md_bdev = bdev;
1587 static void close_backing_dev(struct drbd_device *device, struct block_device *bdev,
1593 bd_unlink_disk_holder(bdev, device->vdisk);
1594 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1597 void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev)
1602 close_backing_dev(device, ldev->md_bdev, ldev->md_bdev != ldev->backing_bdev);
1603 close_backing_dev(device, ldev->backing_bdev, true);
1605 kfree(ldev->disk_conf);
1609 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1611 struct drbd_config_context adm_ctx;
1612 struct drbd_device *device;
1613 struct drbd_peer_device *peer_device;
1614 struct drbd_connection *connection;
1616 enum drbd_ret_code retcode;
1617 enum determine_dev_size dd;
1618 sector_t max_possible_sectors;
1619 sector_t min_md_device_sectors;
1620 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1621 struct disk_conf *new_disk_conf = NULL;
1622 struct lru_cache *resync_lru = NULL;
1623 struct fifo_buffer *new_plan = NULL;
1624 union drbd_state ns, os;
1625 enum drbd_state_rv rv;
1626 struct net_conf *nc;
1628 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1629 if (!adm_ctx.reply_skb)
1631 if (retcode != NO_ERROR)
1634 device = adm_ctx.device;
1635 mutex_lock(&adm_ctx.resource->adm_mutex);
1636 peer_device = first_peer_device(device);
1637 connection = peer_device->connection;
1638 conn_reconfig_start(connection);
1640 /* if you want to reconfigure, please tear down first */
1641 if (device->state.disk > D_DISKLESS) {
1642 retcode = ERR_DISK_CONFIGURED;
1645 /* It may just now have detached because of IO error. Make sure
1646 * drbd_ldev_destroy is done already, we may end up here very fast,
1647 * e.g. if someone calls attach from the on-io-error handler,
1648 * to realize a "hot spare" feature (not that I'd recommend that) */
1649 wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1651 /* make sure there is no leftover from previous force-detach attempts */
1652 clear_bit(FORCE_DETACH, &device->flags);
1653 clear_bit(WAS_IO_ERROR, &device->flags);
1654 clear_bit(WAS_READ_ERROR, &device->flags);
1656 /* and no leftover from previously aborted resync or verify, either */
1657 device->rs_total = 0;
1658 device->rs_failed = 0;
1659 atomic_set(&device->rs_pending_cnt, 0);
1661 /* allocation not in the IO path, drbdsetup context */
1662 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1664 retcode = ERR_NOMEM;
1667 spin_lock_init(&nbc->md.uuid_lock);
1669 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1670 if (!new_disk_conf) {
1671 retcode = ERR_NOMEM;
1674 nbc->disk_conf = new_disk_conf;
1676 set_disk_conf_defaults(new_disk_conf);
1677 err = disk_conf_from_attrs(new_disk_conf, info);
1679 retcode = ERR_MANDATORY_TAG;
1680 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1684 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1685 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1687 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1689 retcode = ERR_NOMEM;
1693 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1694 retcode = ERR_MD_IDX_INVALID;
1699 nc = rcu_dereference(connection->net_conf);
1701 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1703 retcode = ERR_STONITH_AND_PROT_A;
1709 retcode = open_backing_devices(device, new_disk_conf, nbc);
1710 if (retcode != NO_ERROR)
1713 if ((nbc->backing_bdev == nbc->md_bdev) !=
1714 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1715 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1716 retcode = ERR_MD_IDX_INVALID;
1720 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1721 1, 61, sizeof(struct bm_extent),
1722 offsetof(struct bm_extent, lce));
1724 retcode = ERR_NOMEM;
1728 /* Read our meta data super block early.
1729 * This also sets other on-disk offsets. */
1730 retcode = drbd_md_read(device, nbc);
1731 if (retcode != NO_ERROR)
1734 sanitize_disk_conf(device, new_disk_conf, nbc);
1736 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1737 drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1738 (unsigned long long) drbd_get_max_capacity(nbc),
1739 (unsigned long long) new_disk_conf->disk_size);
1740 retcode = ERR_DISK_TOO_SMALL;
1744 if (new_disk_conf->meta_dev_idx < 0) {
1745 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1746 /* at least one MB, otherwise it does not make sense */
1747 min_md_device_sectors = (2<<10);
1749 max_possible_sectors = DRBD_MAX_SECTORS;
1750 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1753 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1754 retcode = ERR_MD_DISK_TOO_SMALL;
1755 drbd_warn(device, "refusing attach: md-device too small, "
1756 "at least %llu sectors needed for this meta-disk type\n",
1757 (unsigned long long) min_md_device_sectors);
1761 /* Make sure the new disk is big enough
1762 * (we may currently be R_PRIMARY with no local disk...) */
1763 if (drbd_get_max_capacity(nbc) <
1764 drbd_get_capacity(device->this_bdev)) {
1765 retcode = ERR_DISK_TOO_SMALL;
1769 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1771 if (nbc->known_size > max_possible_sectors) {
1772 drbd_warn(device, "==> truncating very big lower level device "
1773 "to currently maximum possible %llu sectors <==\n",
1774 (unsigned long long) max_possible_sectors);
1775 if (new_disk_conf->meta_dev_idx >= 0)
1776 drbd_warn(device, "==>> using internal or flexible "
1777 "meta data may help <<==\n");
1780 drbd_suspend_io(device);
1781 /* also wait for the last barrier ack. */
1782 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1783 * We need a way to either ignore barrier acks for barriers sent before a device
1784 * was attached, or a way to wait for all pending barrier acks to come in.
1785 * As barriers are counted per resource,
1786 * we'd need to suspend io on all devices of a resource.
1788 wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1789 /* and for any other previously queued work */
1790 drbd_flush_workqueue(&connection->sender_work);
1792 rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1793 retcode = rv; /* FIXME: Type mismatch. */
1794 drbd_resume_io(device);
1795 if (rv < SS_SUCCESS)
1798 if (!get_ldev_if_state(device, D_ATTACHING))
1799 goto force_diskless;
1801 if (!device->bitmap) {
1802 if (drbd_bm_init(device)) {
1803 retcode = ERR_NOMEM;
1804 goto force_diskless_dec;
1808 if (device->state.conn < C_CONNECTED &&
1809 device->state.role == R_PRIMARY && device->ed_uuid &&
1810 (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1811 drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1812 (unsigned long long)device->ed_uuid);
1813 retcode = ERR_DATA_NOT_CURRENT;
1814 goto force_diskless_dec;
1817 /* Since we are diskless, fix the activity log first... */
1818 if (drbd_check_al_size(device, new_disk_conf)) {
1819 retcode = ERR_NOMEM;
1820 goto force_diskless_dec;
1823 /* Prevent shrinking of consistent devices ! */
1824 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1825 drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1826 drbd_warn(device, "refusing to truncate a consistent device\n");
1827 retcode = ERR_DISK_TOO_SMALL;
1828 goto force_diskless_dec;
1831 lock_all_resources();
1832 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1833 if (retcode != NO_ERROR) {
1834 unlock_all_resources();
1835 goto force_diskless_dec;
1838 /* Reset the "barriers don't work" bits here, then force meta data to
1839 * be written, to ensure we determine if barriers are supported. */
1840 if (new_disk_conf->md_flushes)
1841 clear_bit(MD_NO_FUA, &device->flags);
1843 set_bit(MD_NO_FUA, &device->flags);
1845 /* Point of no return reached.
1846 * Devices and memory are no longer released by error cleanup below.
1847 * now device takes over responsibility, and the state engine should
1848 * clean it up somewhere. */
1849 D_ASSERT(device, device->ldev == NULL);
1851 device->resync = resync_lru;
1852 device->rs_plan_s = new_plan;
1855 new_disk_conf = NULL;
1858 drbd_resync_after_changed(device);
1859 drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH);
1860 unlock_all_resources();
1862 if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
1863 set_bit(CRASHED_PRIMARY, &device->flags);
1865 clear_bit(CRASHED_PRIMARY, &device->flags);
1867 if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1868 !(device->state.role == R_PRIMARY && device->resource->susp_nod))
1869 set_bit(CRASHED_PRIMARY, &device->flags);
1871 device->send_cnt = 0;
1872 device->recv_cnt = 0;
1873 device->read_cnt = 0;
1874 device->writ_cnt = 0;
1876 drbd_reconsider_queue_parameters(device, device->ldev);
1878 /* If I am currently not R_PRIMARY,
1879 * but meta data primary indicator is set,
1880 * I just now recover from a hard crash,
1881 * and have been R_PRIMARY before that crash.
1883 * Now, if I had no connection before that crash
1884 * (have been degraded R_PRIMARY), chances are that
1885 * I won't find my peer now either.
1887 * In that case, and _only_ in that case,
1888 * we use the degr-wfc-timeout instead of the default,
1889 * so we can automatically recover from a crash of a
1890 * degraded but active "cluster" after a certain timeout.
1892 clear_bit(USE_DEGR_WFC_T, &device->flags);
1893 if (device->state.role != R_PRIMARY &&
1894 drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1895 !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
1896 set_bit(USE_DEGR_WFC_T, &device->flags);
1898 dd = drbd_determine_dev_size(device, 0, NULL);
1899 if (dd <= DS_ERROR) {
1900 retcode = ERR_NOMEM_BITMAP;
1901 goto force_diskless_dec;
1902 } else if (dd == DS_GREW)
1903 set_bit(RESYNC_AFTER_NEG, &device->flags);
1905 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
1906 (test_bit(CRASHED_PRIMARY, &device->flags) &&
1907 drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
1908 drbd_info(device, "Assuming that all blocks are out of sync "
1909 "(aka FullSync)\n");
1910 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
1911 "set_n_write from attaching", BM_LOCKED_MASK)) {
1912 retcode = ERR_IO_MD_DISK;
1913 goto force_diskless_dec;
1916 if (drbd_bitmap_io(device, &drbd_bm_read,
1917 "read from attaching", BM_LOCKED_MASK)) {
1918 retcode = ERR_IO_MD_DISK;
1919 goto force_diskless_dec;
1923 if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
1924 drbd_suspend_al(device); /* IO is still suspended here... */
1926 spin_lock_irq(&device->resource->req_lock);
1927 os = drbd_read_state(device);
1929 /* If MDF_CONSISTENT is not set go into inconsistent state,
1930 otherwise investigate MDF_WasUpToDate...
1931 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1932 otherwise into D_CONSISTENT state.
1934 if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
1935 if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
1936 ns.disk = D_CONSISTENT;
1938 ns.disk = D_OUTDATED;
1940 ns.disk = D_INCONSISTENT;
1943 if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
1944 ns.pdsk = D_OUTDATED;
1947 if (ns.disk == D_CONSISTENT &&
1948 (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
1949 ns.disk = D_UP_TO_DATE;
1951 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1952 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1953 this point, because drbd_request_state() modifies these
1956 if (rcu_dereference(device->ldev->disk_conf)->al_updates)
1957 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1959 device->ldev->md.flags |= MDF_AL_DISABLED;
1963 /* In case we are C_CONNECTED postpone any decision on the new disk
1964 state after the negotiation phase. */
1965 if (device->state.conn == C_CONNECTED) {
1966 device->new_state_tmp.i = ns.i;
1968 ns.disk = D_NEGOTIATING;
1970 /* We expect to receive up-to-date UUIDs soon.
1971 To avoid a race in receive_state, free p_uuid while
1972 holding req_lock. I.e. atomic with the state change */
1973 kfree(device->p_uuid);
1974 device->p_uuid = NULL;
1977 rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
1978 spin_unlock_irq(&device->resource->req_lock);
1980 if (rv < SS_SUCCESS)
1981 goto force_diskless_dec;
1983 mod_timer(&device->request_timer, jiffies + HZ);
1985 if (device->state.role == R_PRIMARY)
1986 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
1988 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1990 drbd_md_mark_dirty(device);
1991 drbd_md_sync(device);
1993 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
1995 conn_reconfig_done(connection);
1996 mutex_unlock(&adm_ctx.resource->adm_mutex);
1997 drbd_adm_finish(&adm_ctx, info, retcode);
2003 drbd_force_state(device, NS(disk, D_DISKLESS));
2004 drbd_md_sync(device);
2006 conn_reconfig_done(connection);
2008 close_backing_dev(device, nbc->md_bdev, nbc->md_bdev != nbc->backing_bdev);
2009 close_backing_dev(device, nbc->backing_bdev, true);
2012 kfree(new_disk_conf);
2013 lc_destroy(resync_lru);
2015 mutex_unlock(&adm_ctx.resource->adm_mutex);
2017 drbd_adm_finish(&adm_ctx, info, retcode);
2021 static int adm_detach(struct drbd_device *device, int force)
2023 enum drbd_state_rv retcode;
2028 set_bit(FORCE_DETACH, &device->flags);
2029 drbd_force_state(device, NS(disk, D_FAILED));
2030 retcode = SS_SUCCESS;
2034 drbd_suspend_io(device); /* so no-one is stuck in drbd_al_begin_io */
2035 buffer = drbd_md_get_buffer(device, __func__); /* make sure there is no in-flight meta-data IO */
2037 retcode = drbd_request_state(device, NS(disk, D_FAILED));
2038 drbd_md_put_buffer(device);
2039 } else /* already <= D_FAILED */
2040 retcode = SS_NOTHING_TO_DO;
2041 /* D_FAILED will transition to DISKLESS. */
2042 drbd_resume_io(device);
2043 ret = wait_event_interruptible(device->misc_wait,
2044 device->state.disk != D_FAILED);
2045 if ((int)retcode == (int)SS_IS_DISKLESS)
2046 retcode = SS_NOTHING_TO_DO;
2053 /* Detaching the disk is a process in multiple stages. First we need to lock
2054 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
2055 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
2056 * internal references as well.
2057 * Only then we have finally detached. */
2058 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
2060 struct drbd_config_context adm_ctx;
2061 enum drbd_ret_code retcode;
2062 struct detach_parms parms = { };
2065 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2066 if (!adm_ctx.reply_skb)
2068 if (retcode != NO_ERROR)
2071 if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
2072 err = detach_parms_from_attrs(&parms, info);
2074 retcode = ERR_MANDATORY_TAG;
2075 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2080 mutex_lock(&adm_ctx.resource->adm_mutex);
2081 retcode = adm_detach(adm_ctx.device, parms.force_detach);
2082 mutex_unlock(&adm_ctx.resource->adm_mutex);
2084 drbd_adm_finish(&adm_ctx, info, retcode);
2088 static bool conn_resync_running(struct drbd_connection *connection)
2090 struct drbd_peer_device *peer_device;
2095 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2096 struct drbd_device *device = peer_device->device;
2097 if (device->state.conn == C_SYNC_SOURCE ||
2098 device->state.conn == C_SYNC_TARGET ||
2099 device->state.conn == C_PAUSED_SYNC_S ||
2100 device->state.conn == C_PAUSED_SYNC_T) {
2110 static bool conn_ov_running(struct drbd_connection *connection)
2112 struct drbd_peer_device *peer_device;
2117 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2118 struct drbd_device *device = peer_device->device;
2119 if (device->state.conn == C_VERIFY_S ||
2120 device->state.conn == C_VERIFY_T) {
2130 static enum drbd_ret_code
2131 _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2133 struct drbd_peer_device *peer_device;
2136 if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2137 if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2138 return ERR_NEED_APV_100;
2140 if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2141 return ERR_NEED_APV_100;
2143 if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2144 return ERR_NEED_APV_100;
2147 if (!new_net_conf->two_primaries &&
2148 conn_highest_role(connection) == R_PRIMARY &&
2149 conn_highest_peer(connection) == R_PRIMARY)
2150 return ERR_NEED_ALLOW_TWO_PRI;
2152 if (new_net_conf->two_primaries &&
2153 (new_net_conf->wire_protocol != DRBD_PROT_C))
2154 return ERR_NOT_PROTO_C;
2156 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2157 struct drbd_device *device = peer_device->device;
2158 if (get_ldev(device)) {
2159 enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2161 if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2162 return ERR_STONITH_AND_PROT_A;
2164 if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2165 return ERR_DISCARD_IMPOSSIBLE;
2168 if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2169 return ERR_CONG_NOT_PROTO_A;
2174 static enum drbd_ret_code
2175 check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2177 static enum drbd_ret_code rv;
2178 struct drbd_peer_device *peer_device;
2182 rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2185 /* connection->peer_devices protected by genl_lock() here */
2186 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2187 struct drbd_device *device = peer_device->device;
2188 if (!device->bitmap) {
2189 if (drbd_bm_init(device))
2198 struct crypto_ahash *verify_tfm;
2199 struct crypto_ahash *csums_tfm;
2200 struct crypto_shash *cram_hmac_tfm;
2201 struct crypto_ahash *integrity_tfm;
2205 alloc_shash(struct crypto_shash **tfm, char *tfm_name, int err_alg)
2210 *tfm = crypto_alloc_shash(tfm_name, 0, 0);
2220 alloc_ahash(struct crypto_ahash **tfm, char *tfm_name, int err_alg)
2225 *tfm = crypto_alloc_ahash(tfm_name, 0, CRYPTO_ALG_ASYNC);
2234 static enum drbd_ret_code
2235 alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2237 char hmac_name[CRYPTO_MAX_ALG_NAME];
2238 enum drbd_ret_code rv;
2240 rv = alloc_ahash(&crypto->csums_tfm, new_net_conf->csums_alg,
2244 rv = alloc_ahash(&crypto->verify_tfm, new_net_conf->verify_alg,
2248 rv = alloc_ahash(&crypto->integrity_tfm, new_net_conf->integrity_alg,
2252 if (new_net_conf->cram_hmac_alg[0] != 0) {
2253 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
2254 new_net_conf->cram_hmac_alg);
2256 rv = alloc_shash(&crypto->cram_hmac_tfm, hmac_name,
2263 static void free_crypto(struct crypto *crypto)
2265 crypto_free_shash(crypto->cram_hmac_tfm);
2266 crypto_free_ahash(crypto->integrity_tfm);
2267 crypto_free_ahash(crypto->csums_tfm);
2268 crypto_free_ahash(crypto->verify_tfm);
2271 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2273 struct drbd_config_context adm_ctx;
2274 enum drbd_ret_code retcode;
2275 struct drbd_connection *connection;
2276 struct net_conf *old_net_conf, *new_net_conf = NULL;
2278 int ovr; /* online verify running */
2279 int rsr; /* re-sync running */
2280 struct crypto crypto = { };
2282 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2283 if (!adm_ctx.reply_skb)
2285 if (retcode != NO_ERROR)
2288 connection = adm_ctx.connection;
2289 mutex_lock(&adm_ctx.resource->adm_mutex);
2291 new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
2292 if (!new_net_conf) {
2293 retcode = ERR_NOMEM;
2297 conn_reconfig_start(connection);
2299 mutex_lock(&connection->data.mutex);
2300 mutex_lock(&connection->resource->conf_update);
2301 old_net_conf = connection->net_conf;
2303 if (!old_net_conf) {
2304 drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect");
2305 retcode = ERR_INVALID_REQUEST;
2309 *new_net_conf = *old_net_conf;
2310 if (should_set_defaults(info))
2311 set_net_conf_defaults(new_net_conf);
2313 err = net_conf_from_attrs_for_change(new_net_conf, info);
2314 if (err && err != -ENOMSG) {
2315 retcode = ERR_MANDATORY_TAG;
2316 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2320 retcode = check_net_options(connection, new_net_conf);
2321 if (retcode != NO_ERROR)
2324 /* re-sync running */
2325 rsr = conn_resync_running(connection);
2326 if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2327 retcode = ERR_CSUMS_RESYNC_RUNNING;
2331 /* online verify running */
2332 ovr = conn_ov_running(connection);
2333 if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2334 retcode = ERR_VERIFY_RUNNING;
2338 retcode = alloc_crypto(&crypto, new_net_conf);
2339 if (retcode != NO_ERROR)
2342 rcu_assign_pointer(connection->net_conf, new_net_conf);
2345 crypto_free_ahash(connection->csums_tfm);
2346 connection->csums_tfm = crypto.csums_tfm;
2347 crypto.csums_tfm = NULL;
2350 crypto_free_ahash(connection->verify_tfm);
2351 connection->verify_tfm = crypto.verify_tfm;
2352 crypto.verify_tfm = NULL;
2355 crypto_free_ahash(connection->integrity_tfm);
2356 connection->integrity_tfm = crypto.integrity_tfm;
2357 if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2358 /* Do this without trying to take connection->data.mutex again. */
2359 __drbd_send_protocol(connection, P_PROTOCOL_UPDATE);
2361 crypto_free_shash(connection->cram_hmac_tfm);
2362 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2364 mutex_unlock(&connection->resource->conf_update);
2365 mutex_unlock(&connection->data.mutex);
2367 kfree(old_net_conf);
2369 if (connection->cstate >= C_WF_REPORT_PARAMS) {
2370 struct drbd_peer_device *peer_device;
2373 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2374 drbd_send_sync_param(peer_device);
2380 mutex_unlock(&connection->resource->conf_update);
2381 mutex_unlock(&connection->data.mutex);
2382 free_crypto(&crypto);
2383 kfree(new_net_conf);
2385 conn_reconfig_done(connection);
2387 mutex_unlock(&adm_ctx.resource->adm_mutex);
2389 drbd_adm_finish(&adm_ctx, info, retcode);
2393 static void connection_to_info(struct connection_info *info,
2394 struct drbd_connection *connection)
2396 info->conn_connection_state = connection->cstate;
2397 info->conn_role = conn_highest_peer(connection);
2400 static void peer_device_to_info(struct peer_device_info *info,
2401 struct drbd_peer_device *peer_device)
2403 struct drbd_device *device = peer_device->device;
2405 info->peer_repl_state =
2406 max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn);
2407 info->peer_disk_state = device->state.pdsk;
2408 info->peer_resync_susp_user = device->state.user_isp;
2409 info->peer_resync_susp_peer = device->state.peer_isp;
2410 info->peer_resync_susp_dependency = device->state.aftr_isp;
2413 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2415 struct connection_info connection_info;
2416 enum drbd_notification_type flags;
2417 unsigned int peer_devices = 0;
2418 struct drbd_config_context adm_ctx;
2419 struct drbd_peer_device *peer_device;
2420 struct net_conf *old_net_conf, *new_net_conf = NULL;
2421 struct crypto crypto = { };
2422 struct drbd_resource *resource;
2423 struct drbd_connection *connection;
2424 enum drbd_ret_code retcode;
2428 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2430 if (!adm_ctx.reply_skb)
2432 if (retcode != NO_ERROR)
2434 if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2435 drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing");
2436 retcode = ERR_INVALID_REQUEST;
2440 /* No need for _rcu here. All reconfiguration is
2441 * strictly serialized on genl_lock(). We are protected against
2442 * concurrent reconfiguration/addition/deletion */
2443 for_each_resource(resource, &drbd_resources) {
2444 for_each_connection(connection, resource) {
2445 if (nla_len(adm_ctx.my_addr) == connection->my_addr_len &&
2446 !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr,
2447 connection->my_addr_len)) {
2448 retcode = ERR_LOCAL_ADDR;
2452 if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len &&
2453 !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr,
2454 connection->peer_addr_len)) {
2455 retcode = ERR_PEER_ADDR;
2461 mutex_lock(&adm_ctx.resource->adm_mutex);
2462 connection = first_connection(adm_ctx.resource);
2463 conn_reconfig_start(connection);
2465 if (connection->cstate > C_STANDALONE) {
2466 retcode = ERR_NET_CONFIGURED;
2470 /* allocation not in the IO path, drbdsetup / netlink process context */
2471 new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL);
2472 if (!new_net_conf) {
2473 retcode = ERR_NOMEM;
2477 set_net_conf_defaults(new_net_conf);
2479 err = net_conf_from_attrs(new_net_conf, info);
2480 if (err && err != -ENOMSG) {
2481 retcode = ERR_MANDATORY_TAG;
2482 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2486 retcode = check_net_options(connection, new_net_conf);
2487 if (retcode != NO_ERROR)
2490 retcode = alloc_crypto(&crypto, new_net_conf);
2491 if (retcode != NO_ERROR)
2494 ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2496 drbd_flush_workqueue(&connection->sender_work);
2498 mutex_lock(&adm_ctx.resource->conf_update);
2499 old_net_conf = connection->net_conf;
2501 retcode = ERR_NET_CONFIGURED;
2502 mutex_unlock(&adm_ctx.resource->conf_update);
2505 rcu_assign_pointer(connection->net_conf, new_net_conf);
2507 conn_free_crypto(connection);
2508 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2509 connection->integrity_tfm = crypto.integrity_tfm;
2510 connection->csums_tfm = crypto.csums_tfm;
2511 connection->verify_tfm = crypto.verify_tfm;
2513 connection->my_addr_len = nla_len(adm_ctx.my_addr);
2514 memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2515 connection->peer_addr_len = nla_len(adm_ctx.peer_addr);
2516 memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2518 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2522 connection_to_info(&connection_info, connection);
2523 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2524 mutex_lock(¬ification_mutex);
2525 notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags);
2526 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2527 struct peer_device_info peer_device_info;
2529 peer_device_to_info(&peer_device_info, peer_device);
2530 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2531 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags);
2533 mutex_unlock(¬ification_mutex);
2534 mutex_unlock(&adm_ctx.resource->conf_update);
2537 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2538 struct drbd_device *device = peer_device->device;
2539 device->send_cnt = 0;
2540 device->recv_cnt = 0;
2544 retcode = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2546 conn_reconfig_done(connection);
2547 mutex_unlock(&adm_ctx.resource->adm_mutex);
2548 drbd_adm_finish(&adm_ctx, info, retcode);
2552 free_crypto(&crypto);
2553 kfree(new_net_conf);
2555 conn_reconfig_done(connection);
2556 mutex_unlock(&adm_ctx.resource->adm_mutex);
2558 drbd_adm_finish(&adm_ctx, info, retcode);
2562 static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2564 enum drbd_state_rv rv;
2566 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2567 force ? CS_HARD : 0);
2570 case SS_NOTHING_TO_DO:
2572 case SS_ALREADY_STANDALONE:
2574 case SS_PRIMARY_NOP:
2575 /* Our state checking code wants to see the peer outdated. */
2576 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2578 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2579 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2582 case SS_CW_FAILED_BY_PEER:
2583 /* The peer probably wants to see us outdated. */
2584 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2585 disk, D_OUTDATED), 0);
2586 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2587 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2592 /* no special handling necessary */
2595 if (rv >= SS_SUCCESS) {
2596 enum drbd_state_rv rv2;
2597 /* No one else can reconfigure the network while I am here.
2598 * The state handling only uses drbd_thread_stop_nowait(),
2599 * we want to really wait here until the receiver is no more.
2601 drbd_thread_stop(&connection->receiver);
2603 /* Race breaker. This additional state change request may be
2604 * necessary, if this was a forced disconnect during a receiver
2605 * restart. We may have "killed" the receiver thread just
2606 * after drbd_receiver() returned. Typically, we should be
2607 * C_STANDALONE already, now, and this becomes a no-op.
2609 rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2610 CS_VERBOSE | CS_HARD);
2611 if (rv2 < SS_SUCCESS)
2612 drbd_err(connection,
2613 "unexpected rv2=%d in conn_try_disconnect()\n",
2615 /* Unlike in DRBD 9, the state engine has generated
2616 * NOTIFY_DESTROY events before clearing connection->net_conf. */
2621 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2623 struct drbd_config_context adm_ctx;
2624 struct disconnect_parms parms;
2625 struct drbd_connection *connection;
2626 enum drbd_state_rv rv;
2627 enum drbd_ret_code retcode;
2630 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2631 if (!adm_ctx.reply_skb)
2633 if (retcode != NO_ERROR)
2636 connection = adm_ctx.connection;
2637 memset(&parms, 0, sizeof(parms));
2638 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2639 err = disconnect_parms_from_attrs(&parms, info);
2641 retcode = ERR_MANDATORY_TAG;
2642 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2647 mutex_lock(&adm_ctx.resource->adm_mutex);
2648 rv = conn_try_disconnect(connection, parms.force_disconnect);
2649 if (rv < SS_SUCCESS)
2650 retcode = rv; /* FIXME: Type mismatch. */
2653 mutex_unlock(&adm_ctx.resource->adm_mutex);
2655 drbd_adm_finish(&adm_ctx, info, retcode);
2659 void resync_after_online_grow(struct drbd_device *device)
2661 int iass; /* I am sync source */
2663 drbd_info(device, "Resync of new storage after online grow\n");
2664 if (device->state.role != device->state.peer)
2665 iass = (device->state.role == R_PRIMARY);
2667 iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2670 drbd_start_resync(device, C_SYNC_SOURCE);
2672 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2675 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2677 struct drbd_config_context adm_ctx;
2678 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2679 struct resize_parms rs;
2680 struct drbd_device *device;
2681 enum drbd_ret_code retcode;
2682 enum determine_dev_size dd;
2683 bool change_al_layout = false;
2684 enum dds_flags ddsf;
2688 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2689 if (!adm_ctx.reply_skb)
2691 if (retcode != NO_ERROR)
2694 mutex_lock(&adm_ctx.resource->adm_mutex);
2695 device = adm_ctx.device;
2696 if (!get_ldev(device)) {
2697 retcode = ERR_NO_DISK;
2701 memset(&rs, 0, sizeof(struct resize_parms));
2702 rs.al_stripes = device->ldev->md.al_stripes;
2703 rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2704 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2705 err = resize_parms_from_attrs(&rs, info);
2707 retcode = ERR_MANDATORY_TAG;
2708 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2713 if (device->state.conn > C_CONNECTED) {
2714 retcode = ERR_RESIZE_RESYNC;
2718 if (device->state.role == R_SECONDARY &&
2719 device->state.peer == R_SECONDARY) {
2720 retcode = ERR_NO_PRIMARY;
2724 if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2725 retcode = ERR_NEED_APV_93;
2730 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2732 if (u_size != (sector_t)rs.resize_size) {
2733 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2734 if (!new_disk_conf) {
2735 retcode = ERR_NOMEM;
2740 if (device->ldev->md.al_stripes != rs.al_stripes ||
2741 device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2742 u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2744 if (al_size_k > (16 * 1024 * 1024)) {
2745 retcode = ERR_MD_LAYOUT_TOO_BIG;
2749 if (al_size_k < MD_32kB_SECT/2) {
2750 retcode = ERR_MD_LAYOUT_TOO_SMALL;
2754 if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2755 retcode = ERR_MD_LAYOUT_CONNECTED;
2759 change_al_layout = true;
2762 if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev))
2763 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
2765 if (new_disk_conf) {
2766 mutex_lock(&device->resource->conf_update);
2767 old_disk_conf = device->ldev->disk_conf;
2768 *new_disk_conf = *old_disk_conf;
2769 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2770 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2771 mutex_unlock(&device->resource->conf_update);
2773 kfree(old_disk_conf);
2774 new_disk_conf = NULL;
2777 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2778 dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL);
2779 drbd_md_sync(device);
2781 if (dd == DS_ERROR) {
2782 retcode = ERR_NOMEM_BITMAP;
2784 } else if (dd == DS_ERROR_SPACE_MD) {
2785 retcode = ERR_MD_LAYOUT_NO_FIT;
2787 } else if (dd == DS_ERROR_SHRINK) {
2788 retcode = ERR_IMPLICIT_SHRINK;
2792 if (device->state.conn == C_CONNECTED) {
2794 set_bit(RESIZE_PENDING, &device->flags);
2796 drbd_send_uuids(first_peer_device(device));
2797 drbd_send_sizes(first_peer_device(device), 1, ddsf);
2801 mutex_unlock(&adm_ctx.resource->adm_mutex);
2803 drbd_adm_finish(&adm_ctx, info, retcode);
2808 kfree(new_disk_conf);
2812 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2814 struct drbd_config_context adm_ctx;
2815 enum drbd_ret_code retcode;
2816 struct res_opts res_opts;
2819 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2820 if (!adm_ctx.reply_skb)
2822 if (retcode != NO_ERROR)
2825 res_opts = adm_ctx.resource->res_opts;
2826 if (should_set_defaults(info))
2827 set_res_opts_defaults(&res_opts);
2829 err = res_opts_from_attrs(&res_opts, info);
2830 if (err && err != -ENOMSG) {
2831 retcode = ERR_MANDATORY_TAG;
2832 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2836 mutex_lock(&adm_ctx.resource->adm_mutex);
2837 err = set_resource_options(adm_ctx.resource, &res_opts);
2839 retcode = ERR_INVALID_REQUEST;
2841 retcode = ERR_NOMEM;
2843 mutex_unlock(&adm_ctx.resource->adm_mutex);
2846 drbd_adm_finish(&adm_ctx, info, retcode);
2850 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2852 struct drbd_config_context adm_ctx;
2853 struct drbd_device *device;
2854 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2856 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2857 if (!adm_ctx.reply_skb)
2859 if (retcode != NO_ERROR)
2862 device = adm_ctx.device;
2863 if (!get_ldev(device)) {
2864 retcode = ERR_NO_DISK;
2868 mutex_lock(&adm_ctx.resource->adm_mutex);
2870 /* If there is still bitmap IO pending, probably because of a previous
2871 * resync just being finished, wait for it before requesting a new resync.
2872 * Also wait for it's after_state_ch(). */
2873 drbd_suspend_io(device);
2874 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2875 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2877 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2878 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2879 * try to start a resync handshake as sync target for full sync.
2881 if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
2882 retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
2883 if (retcode >= SS_SUCCESS) {
2884 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2885 "set_n_write from invalidate", BM_LOCKED_MASK))
2886 retcode = ERR_IO_MD_DISK;
2889 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
2890 drbd_resume_io(device);
2891 mutex_unlock(&adm_ctx.resource->adm_mutex);
2894 drbd_adm_finish(&adm_ctx, info, retcode);
2898 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2899 union drbd_state mask, union drbd_state val)
2901 struct drbd_config_context adm_ctx;
2902 enum drbd_ret_code retcode;
2904 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2905 if (!adm_ctx.reply_skb)
2907 if (retcode != NO_ERROR)
2910 mutex_lock(&adm_ctx.resource->adm_mutex);
2911 retcode = drbd_request_state(adm_ctx.device, mask, val);
2912 mutex_unlock(&adm_ctx.resource->adm_mutex);
2914 drbd_adm_finish(&adm_ctx, info, retcode);
2918 static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local)
2922 rv = drbd_bmio_set_n_write(device);
2923 drbd_suspend_al(device);
2927 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2929 struct drbd_config_context adm_ctx;
2930 int retcode; /* drbd_ret_code, drbd_state_rv */
2931 struct drbd_device *device;
2933 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2934 if (!adm_ctx.reply_skb)
2936 if (retcode != NO_ERROR)
2939 device = adm_ctx.device;
2940 if (!get_ldev(device)) {
2941 retcode = ERR_NO_DISK;
2945 mutex_lock(&adm_ctx.resource->adm_mutex);
2947 /* If there is still bitmap IO pending, probably because of a previous
2948 * resync just being finished, wait for it before requesting a new resync.
2949 * Also wait for it's after_state_ch(). */
2950 drbd_suspend_io(device);
2951 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2952 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2954 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
2955 * in the bitmap. Otherwise, try to start a resync handshake
2956 * as sync source for full sync.
2958 if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
2959 /* The peer will get a resync upon connect anyways. Just make that
2960 into a full resync. */
2961 retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
2962 if (retcode >= SS_SUCCESS) {
2963 if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al,
2964 "set_n_write from invalidate_peer",
2965 BM_LOCKED_SET_ALLOWED))
2966 retcode = ERR_IO_MD_DISK;
2969 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
2970 drbd_resume_io(device);
2971 mutex_unlock(&adm_ctx.resource->adm_mutex);
2974 drbd_adm_finish(&adm_ctx, info, retcode);
2978 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2980 struct drbd_config_context adm_ctx;
2981 enum drbd_ret_code retcode;
2983 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2984 if (!adm_ctx.reply_skb)
2986 if (retcode != NO_ERROR)
2989 mutex_lock(&adm_ctx.resource->adm_mutex);
2990 if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2991 retcode = ERR_PAUSE_IS_SET;
2992 mutex_unlock(&adm_ctx.resource->adm_mutex);
2994 drbd_adm_finish(&adm_ctx, info, retcode);
2998 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
3000 struct drbd_config_context adm_ctx;
3001 union drbd_dev_state s;
3002 enum drbd_ret_code retcode;
3004 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3005 if (!adm_ctx.reply_skb)
3007 if (retcode != NO_ERROR)
3010 mutex_lock(&adm_ctx.resource->adm_mutex);
3011 if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
3012 s = adm_ctx.device->state;
3013 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
3014 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
3015 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
3017 retcode = ERR_PAUSE_IS_CLEAR;
3020 mutex_unlock(&adm_ctx.resource->adm_mutex);
3022 drbd_adm_finish(&adm_ctx, info, retcode);
3026 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
3028 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
3031 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
3033 struct drbd_config_context adm_ctx;
3034 struct drbd_device *device;
3035 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3037 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3038 if (!adm_ctx.reply_skb)
3040 if (retcode != NO_ERROR)
3043 mutex_lock(&adm_ctx.resource->adm_mutex);
3044 device = adm_ctx.device;
3045 if (test_bit(NEW_CUR_UUID, &device->flags)) {
3046 if (get_ldev_if_state(device, D_ATTACHING)) {
3047 drbd_uuid_new_current(device);
3050 /* This is effectively a multi-stage "forced down".
3051 * The NEW_CUR_UUID bit is supposedly only set, if we
3052 * lost the replication connection, and are configured
3053 * to freeze IO and wait for some fence-peer handler.
3054 * So we still don't have a replication connection.
3055 * And now we don't have a local disk either. After
3056 * resume, we will fail all pending and new IO, because
3057 * we don't have any data anymore. Which means we will
3058 * eventually be able to terminate all users of this
3059 * device, and then take it down. By bumping the
3060 * "effective" data uuid, we make sure that you really
3061 * need to tear down before you reconfigure, we will
3062 * the refuse to re-connect or re-attach (because no
3063 * matching real data uuid exists).
3066 get_random_bytes(&val, sizeof(u64));
3067 drbd_set_ed_uuid(device, val);
3068 drbd_warn(device, "Resumed without access to data; please tear down before attempting to re-configure.\n");
3070 clear_bit(NEW_CUR_UUID, &device->flags);
3072 drbd_suspend_io(device);
3073 retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
3074 if (retcode == SS_SUCCESS) {
3075 if (device->state.conn < C_CONNECTED)
3076 tl_clear(first_peer_device(device)->connection);
3077 if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
3078 tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO);
3080 drbd_resume_io(device);
3081 mutex_unlock(&adm_ctx.resource->adm_mutex);
3083 drbd_adm_finish(&adm_ctx, info, retcode);
3087 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
3089 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
3092 static int nla_put_drbd_cfg_context(struct sk_buff *skb,
3093 struct drbd_resource *resource,
3094 struct drbd_connection *connection,
3095 struct drbd_device *device)
3098 nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
3100 goto nla_put_failure;
3102 nla_put_u32(skb, T_ctx_volume, device->vnr))
3103 goto nla_put_failure;
3104 if (nla_put_string(skb, T_ctx_resource_name, resource->name))
3105 goto nla_put_failure;
3107 if (connection->my_addr_len &&
3108 nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr))
3109 goto nla_put_failure;
3110 if (connection->peer_addr_len &&
3111 nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr))
3112 goto nla_put_failure;
3114 nla_nest_end(skb, nla);
3119 nla_nest_cancel(skb, nla);
3124 * The generic netlink dump callbacks are called outside the genl_lock(), so
3125 * they cannot use the simple attribute parsing code which uses global
3128 static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr)
3130 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3131 const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3134 nla = nla_find(nlmsg_attrdata(nlh, hdrlen), nlmsg_attrlen(nlh, hdrlen),
3135 DRBD_NLA_CFG_CONTEXT);
3138 return drbd_nla_find_nested(maxtype, nla, __nla_type(attr));
3141 static void resource_to_info(struct resource_info *, struct drbd_resource *);
3143 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb)
3145 struct drbd_genlmsghdr *dh;
3146 struct drbd_resource *resource;
3147 struct resource_info resource_info;
3148 struct resource_statistics resource_statistics;
3153 for_each_resource_rcu(resource, &drbd_resources)
3154 if (resource == (struct drbd_resource *)cb->args[0])
3155 goto found_resource;
3156 err = 0; /* resource was probably deleted */
3159 resource = list_entry(&drbd_resources,
3160 struct drbd_resource, resources);
3163 list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) {
3170 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3171 cb->nlh->nlmsg_seq, &drbd_genl_family,
3172 NLM_F_MULTI, DRBD_ADM_GET_RESOURCES);
3177 dh->ret_code = NO_ERROR;
3178 err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL);
3181 err = res_opts_to_skb(skb, &resource->res_opts, !capable(CAP_SYS_ADMIN));
3184 resource_to_info(&resource_info, resource);
3185 err = resource_info_to_skb(skb, &resource_info, !capable(CAP_SYS_ADMIN));
3188 resource_statistics.res_stat_write_ordering = resource->write_ordering;
3189 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
3192 cb->args[0] = (long)resource;
3193 genlmsg_end(skb, dh);
3203 static void device_to_statistics(struct device_statistics *s,
3204 struct drbd_device *device)
3206 memset(s, 0, sizeof(*s));
3207 s->dev_upper_blocked = !may_inc_ap_bio(device);
3208 if (get_ldev(device)) {
3209 struct drbd_md *md = &device->ldev->md;
3210 u64 *history_uuids = (u64 *)s->history_uuids;
3211 struct request_queue *q;
3214 spin_lock_irq(&md->uuid_lock);
3215 s->dev_current_uuid = md->uuid[UI_CURRENT];
3216 BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1);
3217 for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++)
3218 history_uuids[n] = md->uuid[UI_HISTORY_START + n];
3219 for (; n < HISTORY_UUIDS; n++)
3220 history_uuids[n] = 0;
3221 s->history_uuids_len = HISTORY_UUIDS;
3222 spin_unlock_irq(&md->uuid_lock);
3224 s->dev_disk_flags = md->flags;
3225 q = bdev_get_queue(device->ldev->backing_bdev);
3226 s->dev_lower_blocked =
3227 bdi_congested(&q->backing_dev_info,
3228 (1 << WB_async_congested) |
3229 (1 << WB_sync_congested));
3232 s->dev_size = drbd_get_capacity(device->this_bdev);
3233 s->dev_read = device->read_cnt;
3234 s->dev_write = device->writ_cnt;
3235 s->dev_al_writes = device->al_writ_cnt;
3236 s->dev_bm_writes = device->bm_writ_cnt;
3237 s->dev_upper_pending = atomic_read(&device->ap_bio_cnt);
3238 s->dev_lower_pending = atomic_read(&device->local_cnt);
3239 s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags);
3240 s->dev_exposed_data_uuid = device->ed_uuid;
3243 static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr)
3246 struct drbd_resource *resource =
3247 (struct drbd_resource *)cb->args[0];
3248 kref_put(&resource->kref, drbd_destroy_resource);
3254 int drbd_adm_dump_devices_done(struct netlink_callback *cb) {
3255 return put_resource_in_arg0(cb, 7);
3258 static void device_to_info(struct device_info *, struct drbd_device *);
3260 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb)
3262 struct nlattr *resource_filter;
3263 struct drbd_resource *resource;
3264 struct drbd_device *uninitialized_var(device);
3265 int minor, err, retcode;
3266 struct drbd_genlmsghdr *dh;
3267 struct device_info device_info;
3268 struct device_statistics device_statistics;
3269 struct idr *idr_to_search;
3271 resource = (struct drbd_resource *)cb->args[0];
3272 if (!cb->args[0] && !cb->args[1]) {
3273 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3274 if (resource_filter) {
3275 retcode = ERR_RES_NOT_KNOWN;
3276 resource = drbd_find_resource(nla_data(resource_filter));
3279 cb->args[0] = (long)resource;
3284 minor = cb->args[1];
3285 idr_to_search = resource ? &resource->devices : &drbd_devices;
3286 device = idr_get_next(idr_to_search, &minor);
3291 idr_for_each_entry_continue(idr_to_search, device, minor) {
3293 goto put_result; /* only one iteration */
3296 goto out; /* no more devices */
3299 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3300 cb->nlh->nlmsg_seq, &drbd_genl_family,
3301 NLM_F_MULTI, DRBD_ADM_GET_DEVICES);
3305 dh->ret_code = retcode;
3307 if (retcode == NO_ERROR) {
3308 dh->minor = device->minor;
3309 err = nla_put_drbd_cfg_context(skb, device->resource, NULL, device);
3312 if (get_ldev(device)) {
3313 struct disk_conf *disk_conf =
3314 rcu_dereference(device->ldev->disk_conf);
3316 err = disk_conf_to_skb(skb, disk_conf, !capable(CAP_SYS_ADMIN));
3321 device_to_info(&device_info, device);
3322 err = device_info_to_skb(skb, &device_info, !capable(CAP_SYS_ADMIN));
3326 device_to_statistics(&device_statistics, device);
3327 err = device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
3330 cb->args[1] = minor + 1;
3332 genlmsg_end(skb, dh);
3342 int drbd_adm_dump_connections_done(struct netlink_callback *cb)
3344 return put_resource_in_arg0(cb, 6);
3347 enum { SINGLE_RESOURCE, ITERATE_RESOURCES };
3349 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb)
3351 struct nlattr *resource_filter;
3352 struct drbd_resource *resource = NULL, *next_resource;
3353 struct drbd_connection *uninitialized_var(connection);
3354 int err = 0, retcode;
3355 struct drbd_genlmsghdr *dh;
3356 struct connection_info connection_info;
3357 struct connection_statistics connection_statistics;
3360 resource = (struct drbd_resource *)cb->args[0];
3362 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3363 if (resource_filter) {
3364 retcode = ERR_RES_NOT_KNOWN;
3365 resource = drbd_find_resource(nla_data(resource_filter));
3368 cb->args[0] = (long)resource;
3369 cb->args[1] = SINGLE_RESOURCE;
3373 if (list_empty(&drbd_resources))
3375 resource = list_first_entry(&drbd_resources, struct drbd_resource, resources);
3376 kref_get(&resource->kref);
3377 cb->args[0] = (long)resource;
3378 cb->args[1] = ITERATE_RESOURCES;
3383 mutex_lock(&resource->conf_update);
3386 for_each_connection_rcu(connection, resource)
3387 if (connection == (struct drbd_connection *)cb->args[2])
3388 goto found_connection;
3389 /* connection was probably deleted */
3390 goto no_more_connections;
3392 connection = list_entry(&resource->connections, struct drbd_connection, connections);
3395 list_for_each_entry_continue_rcu(connection, &resource->connections, connections) {
3396 if (!has_net_conf(connection))
3399 goto put_result; /* only one iteration */
3402 no_more_connections:
3403 if (cb->args[1] == ITERATE_RESOURCES) {
3404 for_each_resource_rcu(next_resource, &drbd_resources) {
3405 if (next_resource == resource)
3406 goto found_resource;
3408 /* resource was probably deleted */
3413 list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) {
3414 mutex_unlock(&resource->conf_update);
3415 kref_put(&resource->kref, drbd_destroy_resource);
3416 resource = next_resource;
3417 kref_get(&resource->kref);
3418 cb->args[0] = (long)resource;
3422 goto out; /* no more resources */
3425 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3426 cb->nlh->nlmsg_seq, &drbd_genl_family,
3427 NLM_F_MULTI, DRBD_ADM_GET_CONNECTIONS);
3431 dh->ret_code = retcode;
3433 if (retcode == NO_ERROR) {
3434 struct net_conf *net_conf;
3436 err = nla_put_drbd_cfg_context(skb, resource, connection, NULL);
3439 net_conf = rcu_dereference(connection->net_conf);
3441 err = net_conf_to_skb(skb, net_conf, !capable(CAP_SYS_ADMIN));
3445 connection_to_info(&connection_info, connection);
3446 err = connection_info_to_skb(skb, &connection_info, !capable(CAP_SYS_ADMIN));
3449 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
3450 err = connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
3453 cb->args[2] = (long)connection;
3455 genlmsg_end(skb, dh);
3461 mutex_unlock(&resource->conf_update);
3467 enum mdf_peer_flag {
3468 MDF_PEER_CONNECTED = 1 << 0,
3469 MDF_PEER_OUTDATED = 1 << 1,
3470 MDF_PEER_FENCING = 1 << 2,
3471 MDF_PEER_FULL_SYNC = 1 << 3,
3474 static void peer_device_to_statistics(struct peer_device_statistics *s,
3475 struct drbd_peer_device *peer_device)
3477 struct drbd_device *device = peer_device->device;
3479 memset(s, 0, sizeof(*s));
3480 s->peer_dev_received = device->recv_cnt;
3481 s->peer_dev_sent = device->send_cnt;
3482 s->peer_dev_pending = atomic_read(&device->ap_pending_cnt) +
3483 atomic_read(&device->rs_pending_cnt);
3484 s->peer_dev_unacked = atomic_read(&device->unacked_cnt);
3485 s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9);
3486 s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9);
3487 if (get_ldev(device)) {
3488 struct drbd_md *md = &device->ldev->md;
3490 spin_lock_irq(&md->uuid_lock);
3491 s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP];
3492 spin_unlock_irq(&md->uuid_lock);
3494 (drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ?
3495 MDF_PEER_CONNECTED : 0) +
3496 (drbd_md_test_flag(device->ldev, MDF_CONSISTENT) &&
3497 !drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ?
3498 MDF_PEER_OUTDATED : 0) +
3499 /* FIXME: MDF_PEER_FENCING? */
3500 (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ?
3501 MDF_PEER_FULL_SYNC : 0);
3506 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb)
3508 return put_resource_in_arg0(cb, 9);
3511 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb)
3513 struct nlattr *resource_filter;
3514 struct drbd_resource *resource;
3515 struct drbd_device *uninitialized_var(device);
3516 struct drbd_peer_device *peer_device = NULL;
3517 int minor, err, retcode;
3518 struct drbd_genlmsghdr *dh;
3519 struct idr *idr_to_search;
3521 resource = (struct drbd_resource *)cb->args[0];
3522 if (!cb->args[0] && !cb->args[1]) {
3523 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3524 if (resource_filter) {
3525 retcode = ERR_RES_NOT_KNOWN;
3526 resource = drbd_find_resource(nla_data(resource_filter));
3530 cb->args[0] = (long)resource;
3534 minor = cb->args[1];
3535 idr_to_search = resource ? &resource->devices : &drbd_devices;
3536 device = idr_find(idr_to_search, minor);
3541 device = idr_get_next(idr_to_search, &minor);
3548 for_each_peer_device(peer_device, device)
3549 if (peer_device == (struct drbd_peer_device *)cb->args[2])
3550 goto found_peer_device;
3551 /* peer device was probably deleted */
3554 /* Make peer_device point to the list head (not the first entry). */
3555 peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices);
3558 list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) {
3559 if (!has_net_conf(peer_device->connection))
3562 goto put_result; /* only one iteration */
3567 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3568 cb->nlh->nlmsg_seq, &drbd_genl_family,
3569 NLM_F_MULTI, DRBD_ADM_GET_PEER_DEVICES);
3573 dh->ret_code = retcode;
3575 if (retcode == NO_ERROR) {
3576 struct peer_device_info peer_device_info;
3577 struct peer_device_statistics peer_device_statistics;
3580 err = nla_put_drbd_cfg_context(skb, device->resource, peer_device->connection, device);
3583 peer_device_to_info(&peer_device_info, peer_device);
3584 err = peer_device_info_to_skb(skb, &peer_device_info, !capable(CAP_SYS_ADMIN));
3587 peer_device_to_statistics(&peer_device_statistics, peer_device);
3588 err = peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
3591 cb->args[1] = minor;
3592 cb->args[2] = (long)peer_device;
3594 genlmsg_end(skb, dh);
3604 * Return the connection of @resource if @resource has exactly one connection.
3606 static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
3608 struct list_head *connections = &resource->connections;
3610 if (list_empty(connections) || connections->next->next != connections)
3612 return list_first_entry(&resource->connections, struct drbd_connection, connections);
3615 static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
3616 const struct sib_info *sib)
3618 struct drbd_resource *resource = device->resource;
3619 struct state_info *si = NULL; /* for sizeof(si->member); */
3623 int exclude_sensitive;
3625 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
3626 * to. So we better exclude_sensitive information.
3628 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
3629 * in the context of the requesting user process. Exclude sensitive
3630 * information, unless current has superuser.
3632 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
3633 * relies on the current implementation of netlink_dump(), which
3634 * executes the dump callback successively from netlink_recvmsg(),
3635 * always in the context of the receiving process */
3636 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
3638 got_ldev = get_ldev(device);
3640 /* We need to add connection name and volume number information still.
3641 * Minor number is in drbd_genlmsghdr. */
3642 if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device))
3643 goto nla_put_failure;
3645 if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive))
3646 goto nla_put_failure;
3650 struct disk_conf *disk_conf;
3652 disk_conf = rcu_dereference(device->ldev->disk_conf);
3653 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
3656 struct net_conf *nc;
3658 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3660 err = net_conf_to_skb(skb, nc, exclude_sensitive);
3664 goto nla_put_failure;
3666 nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
3668 goto nla_put_failure;
3669 if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
3670 nla_put_u32(skb, T_current_state, device->state.i) ||
3671 nla_put_u64_0pad(skb, T_ed_uuid, device->ed_uuid) ||
3672 nla_put_u64_0pad(skb, T_capacity,
3673 drbd_get_capacity(device->this_bdev)) ||
3674 nla_put_u64_0pad(skb, T_send_cnt, device->send_cnt) ||
3675 nla_put_u64_0pad(skb, T_recv_cnt, device->recv_cnt) ||
3676 nla_put_u64_0pad(skb, T_read_cnt, device->read_cnt) ||
3677 nla_put_u64_0pad(skb, T_writ_cnt, device->writ_cnt) ||
3678 nla_put_u64_0pad(skb, T_al_writ_cnt, device->al_writ_cnt) ||
3679 nla_put_u64_0pad(skb, T_bm_writ_cnt, device->bm_writ_cnt) ||
3680 nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) ||
3681 nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) ||
3682 nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt)))
3683 goto nla_put_failure;
3688 spin_lock_irq(&device->ldev->md.uuid_lock);
3689 err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid);
3690 spin_unlock_irq(&device->ldev->md.uuid_lock);
3693 goto nla_put_failure;
3695 if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) ||
3696 nla_put_u64_0pad(skb, T_bits_total, drbd_bm_bits(device)) ||
3697 nla_put_u64_0pad(skb, T_bits_oos,
3698 drbd_bm_total_weight(device)))
3699 goto nla_put_failure;
3700 if (C_SYNC_SOURCE <= device->state.conn &&
3701 C_PAUSED_SYNC_T >= device->state.conn) {
3702 if (nla_put_u64_0pad(skb, T_bits_rs_total,
3703 device->rs_total) ||
3704 nla_put_u64_0pad(skb, T_bits_rs_failed,
3706 goto nla_put_failure;
3711 switch(sib->sib_reason) {
3712 case SIB_SYNC_PROGRESS:
3713 case SIB_GET_STATUS_REPLY:
3715 case SIB_STATE_CHANGE:
3716 if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
3717 nla_put_u32(skb, T_new_state, sib->ns.i))
3718 goto nla_put_failure;
3720 case SIB_HELPER_POST:
3721 if (nla_put_u32(skb, T_helper_exit_code,
3722 sib->helper_exit_code))
3723 goto nla_put_failure;
3725 case SIB_HELPER_PRE:
3726 if (nla_put_string(skb, T_helper, sib->helper_name))
3727 goto nla_put_failure;
3731 nla_nest_end(skb, nla);
3741 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3743 struct drbd_config_context adm_ctx;
3744 enum drbd_ret_code retcode;
3747 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3748 if (!adm_ctx.reply_skb)
3750 if (retcode != NO_ERROR)
3753 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL);
3755 nlmsg_free(adm_ctx.reply_skb);
3759 drbd_adm_finish(&adm_ctx, info, retcode);
3763 static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3765 struct drbd_device *device;
3766 struct drbd_genlmsghdr *dh;
3767 struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3768 struct drbd_resource *resource = NULL;
3769 struct drbd_resource *tmp;
3770 unsigned volume = cb->args[1];
3772 /* Open coded, deferred, iteration:
3773 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3774 * connection = "first connection of resource or undefined";
3775 * idr_for_each_entry(&resource->devices, device, i) {
3779 * where resource is cb->args[0];
3780 * and i is cb->args[1];
3782 * cb->args[2] indicates if we shall loop over all resources,
3783 * or just dump all volumes of a single resource.
3785 * This may miss entries inserted after this dump started,
3786 * or entries deleted before they are reached.
3788 * We need to make sure the device won't disappear while
3789 * we are looking at it, and revalidate our iterators
3790 * on each iteration.
3793 /* synchronize with conn_create()/drbd_destroy_connection() */
3795 /* revalidate iterator position */
3796 for_each_resource_rcu(tmp, &drbd_resources) {
3798 /* first iteration */
3810 device = idr_get_next(&resource->devices, &volume);
3812 /* No more volumes to dump on this resource.
3813 * Advance resource iterator. */
3814 pos = list_entry_rcu(resource->resources.next,
3815 struct drbd_resource, resources);
3816 /* Did we dump any volume of this resource yet? */
3818 /* If we reached the end of the list,
3819 * or only a single resource dump was requested,
3821 if (&pos->resources == &drbd_resources || cb->args[2])
3829 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3830 cb->nlh->nlmsg_seq, &drbd_genl_family,
3831 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
3836 /* This is a connection without a single volume.
3837 * Suprisingly enough, it may have a network
3839 struct drbd_connection *connection;
3842 dh->ret_code = NO_ERROR;
3843 connection = the_only_connection(resource);
3844 if (nla_put_drbd_cfg_context(skb, resource, connection, NULL))
3847 struct net_conf *nc;
3849 nc = rcu_dereference(connection->net_conf);
3850 if (nc && net_conf_to_skb(skb, nc, 1) != 0)
3856 D_ASSERT(device, device->vnr == volume);
3857 D_ASSERT(device, device->resource == resource);
3859 dh->minor = device_to_minor(device);
3860 dh->ret_code = NO_ERROR;
3862 if (nla_put_status_info(skb, device, NULL)) {
3864 genlmsg_cancel(skb, dh);
3868 genlmsg_end(skb, dh);
3873 /* where to start the next iteration */
3874 cb->args[0] = (long)pos;
3875 cb->args[1] = (pos == resource) ? volume + 1 : 0;
3877 /* No more resources/volumes/minors found results in an empty skb.
3878 * Which will terminate the dump. */
3883 * Request status of all resources, or of all volumes within a single resource.
3885 * This is a dump, as the answer may not fit in a single reply skb otherwise.
3886 * Which means we cannot use the family->attrbuf or other such members, because
3887 * dump is NOT protected by the genl_lock(). During dump, we only have access
3888 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3890 * Once things are setup properly, we call into get_one_status().
3892 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
3894 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3896 const char *resource_name;
3897 struct drbd_resource *resource;
3900 /* Is this a followup call? */
3902 /* ... of a single resource dump,
3903 * and the resource iterator has been advanced already? */
3904 if (cb->args[2] && cb->args[2] != cb->args[0])
3905 return 0; /* DONE. */
3909 /* First call (from netlink_dump_start). We need to figure out
3910 * which resource(s) the user wants us to dump. */
3911 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
3912 nlmsg_attrlen(cb->nlh, hdrlen),
3913 DRBD_NLA_CFG_CONTEXT);
3915 /* No explicit context given. Dump all. */
3918 maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3919 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
3921 return PTR_ERR(nla);
3922 /* context given, but no name present? */
3925 resource_name = nla_data(nla);
3926 if (!*resource_name)
3928 resource = drbd_find_resource(resource_name);
3932 kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */
3934 /* prime iterators, and set "filter" mode mark:
3935 * only dump this connection. */
3936 cb->args[0] = (long)resource;
3937 /* cb->args[1] = 0; passed in this way. */
3938 cb->args[2] = (long)resource;
3941 return get_one_status(skb, cb);
3944 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
3946 struct drbd_config_context adm_ctx;
3947 enum drbd_ret_code retcode;
3948 struct timeout_parms tp;
3951 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3952 if (!adm_ctx.reply_skb)
3954 if (retcode != NO_ERROR)
3958 adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
3959 test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED :
3962 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
3964 nlmsg_free(adm_ctx.reply_skb);
3968 drbd_adm_finish(&adm_ctx, info, retcode);
3972 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
3974 struct drbd_config_context adm_ctx;
3975 struct drbd_device *device;
3976 enum drbd_ret_code retcode;
3977 struct start_ov_parms parms;
3979 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3980 if (!adm_ctx.reply_skb)
3982 if (retcode != NO_ERROR)
3985 device = adm_ctx.device;
3987 /* resume from last known position, if possible */
3988 parms.ov_start_sector = device->ov_start_sector;
3989 parms.ov_stop_sector = ULLONG_MAX;
3990 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
3991 int err = start_ov_parms_from_attrs(&parms, info);
3993 retcode = ERR_MANDATORY_TAG;
3994 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
3998 mutex_lock(&adm_ctx.resource->adm_mutex);
4000 /* w_make_ov_request expects position to be aligned */
4001 device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
4002 device->ov_stop_sector = parms.ov_stop_sector;
4004 /* If there is still bitmap IO pending, e.g. previous resync or verify
4005 * just being finished, wait for it before requesting a new resync. */
4006 drbd_suspend_io(device);
4007 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
4008 retcode = drbd_request_state(device, NS(conn, C_VERIFY_S));
4009 drbd_resume_io(device);
4011 mutex_unlock(&adm_ctx.resource->adm_mutex);
4013 drbd_adm_finish(&adm_ctx, info, retcode);
4018 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
4020 struct drbd_config_context adm_ctx;
4021 struct drbd_device *device;
4022 enum drbd_ret_code retcode;
4023 int skip_initial_sync = 0;
4025 struct new_c_uuid_parms args;
4027 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4028 if (!adm_ctx.reply_skb)
4030 if (retcode != NO_ERROR)
4033 device = adm_ctx.device;
4034 memset(&args, 0, sizeof(args));
4035 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
4036 err = new_c_uuid_parms_from_attrs(&args, info);
4038 retcode = ERR_MANDATORY_TAG;
4039 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4044 mutex_lock(&adm_ctx.resource->adm_mutex);
4045 mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */
4047 if (!get_ldev(device)) {
4048 retcode = ERR_NO_DISK;
4052 /* this is "skip initial sync", assume to be clean */
4053 if (device->state.conn == C_CONNECTED &&
4054 first_peer_device(device)->connection->agreed_pro_version >= 90 &&
4055 device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
4056 drbd_info(device, "Preparing to skip initial sync\n");
4057 skip_initial_sync = 1;
4058 } else if (device->state.conn != C_STANDALONE) {
4059 retcode = ERR_CONNECTED;
4063 drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
4064 drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
4066 if (args.clear_bm) {
4067 err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
4068 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
4070 drbd_err(device, "Writing bitmap failed with %d\n", err);
4071 retcode = ERR_IO_MD_DISK;
4073 if (skip_initial_sync) {
4074 drbd_send_uuids_skip_initial_sync(first_peer_device(device));
4075 _drbd_uuid_set(device, UI_BITMAP, 0);
4076 drbd_print_uuids(device, "cleared bitmap UUID");
4077 spin_lock_irq(&device->resource->req_lock);
4078 _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
4080 spin_unlock_irq(&device->resource->req_lock);
4084 drbd_md_sync(device);
4088 mutex_unlock(device->state_mutex);
4089 mutex_unlock(&adm_ctx.resource->adm_mutex);
4091 drbd_adm_finish(&adm_ctx, info, retcode);
4095 static enum drbd_ret_code
4096 drbd_check_resource_name(struct drbd_config_context *adm_ctx)
4098 const char *name = adm_ctx->resource_name;
4099 if (!name || !name[0]) {
4100 drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing");
4101 return ERR_MANDATORY_TAG;
4103 /* if we want to use these in sysfs/configfs/debugfs some day,
4104 * we must not allow slashes */
4105 if (strchr(name, '/')) {
4106 drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name");
4107 return ERR_INVALID_REQUEST;
4112 static void resource_to_info(struct resource_info *info,
4113 struct drbd_resource *resource)
4115 info->res_role = conn_highest_role(first_connection(resource));
4116 info->res_susp = resource->susp;
4117 info->res_susp_nod = resource->susp_nod;
4118 info->res_susp_fen = resource->susp_fen;
4121 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
4123 struct drbd_connection *connection;
4124 struct drbd_config_context adm_ctx;
4125 enum drbd_ret_code retcode;
4126 struct res_opts res_opts;
4129 retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0);
4130 if (!adm_ctx.reply_skb)
4132 if (retcode != NO_ERROR)
4135 set_res_opts_defaults(&res_opts);
4136 err = res_opts_from_attrs(&res_opts, info);
4137 if (err && err != -ENOMSG) {
4138 retcode = ERR_MANDATORY_TAG;
4139 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4143 retcode = drbd_check_resource_name(&adm_ctx);
4144 if (retcode != NO_ERROR)
4147 if (adm_ctx.resource) {
4148 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
4149 retcode = ERR_INVALID_REQUEST;
4150 drbd_msg_put_info(adm_ctx.reply_skb, "resource exists");
4152 /* else: still NO_ERROR */
4156 /* not yet safe for genl_family.parallel_ops */
4157 mutex_lock(&resources_mutex);
4158 connection = conn_create(adm_ctx.resource_name, &res_opts);
4159 mutex_unlock(&resources_mutex);
4162 struct resource_info resource_info;
4164 mutex_lock(¬ification_mutex);
4165 resource_to_info(&resource_info, connection->resource);
4166 notify_resource_state(NULL, 0, connection->resource,
4167 &resource_info, NOTIFY_CREATE);
4168 mutex_unlock(¬ification_mutex);
4170 retcode = ERR_NOMEM;
4173 drbd_adm_finish(&adm_ctx, info, retcode);
4177 static void device_to_info(struct device_info *info,
4178 struct drbd_device *device)
4180 info->dev_disk_state = device->state.disk;
4184 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info)
4186 struct drbd_config_context adm_ctx;
4187 struct drbd_genlmsghdr *dh = info->userhdr;
4188 enum drbd_ret_code retcode;
4190 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4191 if (!adm_ctx.reply_skb)
4193 if (retcode != NO_ERROR)
4196 if (dh->minor > MINORMASK) {
4197 drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range");
4198 retcode = ERR_INVALID_REQUEST;
4201 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
4202 drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range");
4203 retcode = ERR_INVALID_REQUEST;
4207 /* drbd_adm_prepare made sure already
4208 * that first_peer_device(device)->connection and device->vnr match the request. */
4209 if (adm_ctx.device) {
4210 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
4211 retcode = ERR_MINOR_OR_VOLUME_EXISTS;
4212 /* else: still NO_ERROR */
4216 mutex_lock(&adm_ctx.resource->adm_mutex);
4217 retcode = drbd_create_device(&adm_ctx, dh->minor);
4218 if (retcode == NO_ERROR) {
4219 struct drbd_device *device;
4220 struct drbd_peer_device *peer_device;
4221 struct device_info info;
4222 unsigned int peer_devices = 0;
4223 enum drbd_notification_type flags;
4225 device = minor_to_device(dh->minor);
4226 for_each_peer_device(peer_device, device) {
4227 if (!has_net_conf(peer_device->connection))
4232 device_to_info(&info, device);
4233 mutex_lock(¬ification_mutex);
4234 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4235 notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags);
4236 for_each_peer_device(peer_device, device) {
4237 struct peer_device_info peer_device_info;
4239 if (!has_net_conf(peer_device->connection))
4241 peer_device_to_info(&peer_device_info, peer_device);
4242 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4243 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info,
4244 NOTIFY_CREATE | flags);
4246 mutex_unlock(¬ification_mutex);
4248 mutex_unlock(&adm_ctx.resource->adm_mutex);
4250 drbd_adm_finish(&adm_ctx, info, retcode);
4254 static enum drbd_ret_code adm_del_minor(struct drbd_device *device)
4256 struct drbd_peer_device *peer_device;
4258 if (device->state.disk == D_DISKLESS &&
4259 /* no need to be device->state.conn == C_STANDALONE &&
4260 * we may want to delete a minor from a live replication group.
4262 device->state.role == R_SECONDARY) {
4263 struct drbd_connection *connection =
4264 first_connection(device->resource);
4266 _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS),
4267 CS_VERBOSE + CS_WAIT_COMPLETE);
4269 /* If the state engine hasn't stopped the sender thread yet, we
4270 * need to flush the sender work queue before generating the
4271 * DESTROY events here. */
4272 if (get_t_state(&connection->worker) == RUNNING)
4273 drbd_flush_workqueue(&connection->sender_work);
4275 mutex_lock(¬ification_mutex);
4276 for_each_peer_device(peer_device, device) {
4277 if (!has_net_conf(peer_device->connection))
4279 notify_peer_device_state(NULL, 0, peer_device, NULL,
4280 NOTIFY_DESTROY | NOTIFY_CONTINUES);
4282 notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY);
4283 mutex_unlock(¬ification_mutex);
4285 drbd_delete_device(device);
4288 return ERR_MINOR_CONFIGURED;
4291 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info)
4293 struct drbd_config_context adm_ctx;
4294 enum drbd_ret_code retcode;
4296 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4297 if (!adm_ctx.reply_skb)
4299 if (retcode != NO_ERROR)
4302 mutex_lock(&adm_ctx.resource->adm_mutex);
4303 retcode = adm_del_minor(adm_ctx.device);
4304 mutex_unlock(&adm_ctx.resource->adm_mutex);
4306 drbd_adm_finish(&adm_ctx, info, retcode);
4310 static int adm_del_resource(struct drbd_resource *resource)
4312 struct drbd_connection *connection;
4314 for_each_connection(connection, resource) {
4315 if (connection->cstate > C_STANDALONE)
4316 return ERR_NET_CONFIGURED;
4318 if (!idr_is_empty(&resource->devices))
4319 return ERR_RES_IN_USE;
4321 /* The state engine has stopped the sender thread, so we don't
4322 * need to flush the sender work queue before generating the
4323 * DESTROY event here. */
4324 mutex_lock(¬ification_mutex);
4325 notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY);
4326 mutex_unlock(¬ification_mutex);
4328 mutex_lock(&resources_mutex);
4329 list_del_rcu(&resource->resources);
4330 mutex_unlock(&resources_mutex);
4331 /* Make sure all threads have actually stopped: state handling only
4332 * does drbd_thread_stop_nowait(). */
4333 list_for_each_entry(connection, &resource->connections, connections)
4334 drbd_thread_stop(&connection->worker);
4336 drbd_free_resource(resource);
4340 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
4342 struct drbd_config_context adm_ctx;
4343 struct drbd_resource *resource;
4344 struct drbd_connection *connection;
4345 struct drbd_device *device;
4346 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
4349 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4350 if (!adm_ctx.reply_skb)
4352 if (retcode != NO_ERROR)
4355 resource = adm_ctx.resource;
4356 mutex_lock(&resource->adm_mutex);
4358 for_each_connection(connection, resource) {
4359 struct drbd_peer_device *peer_device;
4361 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
4362 retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0);
4363 if (retcode < SS_SUCCESS) {
4364 drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote");
4369 retcode = conn_try_disconnect(connection, 0);
4370 if (retcode < SS_SUCCESS) {
4371 drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect");
4377 idr_for_each_entry(&resource->devices, device, i) {
4378 retcode = adm_detach(device, 0);
4379 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
4380 drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach");
4385 /* delete volumes */
4386 idr_for_each_entry(&resource->devices, device, i) {
4387 retcode = adm_del_minor(device);
4388 if (retcode != NO_ERROR) {
4389 /* "can not happen" */
4390 drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume");
4395 retcode = adm_del_resource(resource);
4397 mutex_unlock(&resource->adm_mutex);
4399 drbd_adm_finish(&adm_ctx, info, retcode);
4403 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
4405 struct drbd_config_context adm_ctx;
4406 struct drbd_resource *resource;
4407 enum drbd_ret_code retcode;
4409 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4410 if (!adm_ctx.reply_skb)
4412 if (retcode != NO_ERROR)
4414 resource = adm_ctx.resource;
4416 mutex_lock(&resource->adm_mutex);
4417 retcode = adm_del_resource(resource);
4418 mutex_unlock(&resource->adm_mutex);
4420 drbd_adm_finish(&adm_ctx, info, retcode);
4424 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib)
4426 struct sk_buff *msg;
4427 struct drbd_genlmsghdr *d_out;
4431 seq = atomic_inc_return(&drbd_genl_seq);
4432 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4437 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
4438 if (!d_out) /* cannot happen, but anyways. */
4439 goto nla_put_failure;
4440 d_out->minor = device_to_minor(device);
4441 d_out->ret_code = NO_ERROR;
4443 if (nla_put_status_info(msg, device, sib))
4444 goto nla_put_failure;
4445 genlmsg_end(msg, d_out);
4446 err = drbd_genl_multicast_events(msg, GFP_NOWAIT);
4447 /* msg has been consumed or freed in netlink_broadcast() */
4448 if (err && err != -ESRCH)
4456 drbd_err(device, "Error %d while broadcasting event. "
4457 "Event seq:%u sib_reason:%u\n",
4458 err, seq, sib->sib_reason);
4461 static int nla_put_notification_header(struct sk_buff *msg,
4462 enum drbd_notification_type type)
4464 struct drbd_notification_header nh = {
4468 return drbd_notification_header_to_skb(msg, &nh, true);
4471 void notify_resource_state(struct sk_buff *skb,
4473 struct drbd_resource *resource,
4474 struct resource_info *resource_info,
4475 enum drbd_notification_type type)
4477 struct resource_statistics resource_statistics;
4478 struct drbd_genlmsghdr *dh;
4479 bool multicast = false;
4483 seq = atomic_inc_return(¬ify_genl_seq);
4484 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4492 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_RESOURCE_STATE);
4494 goto nla_put_failure;
4496 dh->ret_code = NO_ERROR;
4497 if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) ||
4498 nla_put_notification_header(skb, type) ||
4499 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4500 resource_info_to_skb(skb, resource_info, true)))
4501 goto nla_put_failure;
4502 resource_statistics.res_stat_write_ordering = resource->write_ordering;
4503 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
4505 goto nla_put_failure;
4506 genlmsg_end(skb, dh);
4508 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4509 /* skb has been consumed or freed in netlink_broadcast() */
4510 if (err && err != -ESRCH)
4518 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4522 void notify_device_state(struct sk_buff *skb,
4524 struct drbd_device *device,
4525 struct device_info *device_info,
4526 enum drbd_notification_type type)
4528 struct device_statistics device_statistics;
4529 struct drbd_genlmsghdr *dh;
4530 bool multicast = false;
4534 seq = atomic_inc_return(¬ify_genl_seq);
4535 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4543 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_DEVICE_STATE);
4545 goto nla_put_failure;
4546 dh->minor = device->minor;
4547 dh->ret_code = NO_ERROR;
4548 if (nla_put_drbd_cfg_context(skb, device->resource, NULL, device) ||
4549 nla_put_notification_header(skb, type) ||
4550 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4551 device_info_to_skb(skb, device_info, true)))
4552 goto nla_put_failure;
4553 device_to_statistics(&device_statistics, device);
4554 device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
4555 genlmsg_end(skb, dh);
4557 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4558 /* skb has been consumed or freed in netlink_broadcast() */
4559 if (err && err != -ESRCH)
4567 drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n",
4571 void notify_connection_state(struct sk_buff *skb,
4573 struct drbd_connection *connection,
4574 struct connection_info *connection_info,
4575 enum drbd_notification_type type)
4577 struct connection_statistics connection_statistics;
4578 struct drbd_genlmsghdr *dh;
4579 bool multicast = false;
4583 seq = atomic_inc_return(¬ify_genl_seq);
4584 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4592 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_CONNECTION_STATE);
4594 goto nla_put_failure;
4596 dh->ret_code = NO_ERROR;
4597 if (nla_put_drbd_cfg_context(skb, connection->resource, connection, NULL) ||
4598 nla_put_notification_header(skb, type) ||
4599 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4600 connection_info_to_skb(skb, connection_info, true)))
4601 goto nla_put_failure;
4602 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
4603 connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
4604 genlmsg_end(skb, dh);
4606 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4607 /* skb has been consumed or freed in netlink_broadcast() */
4608 if (err && err != -ESRCH)
4616 drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n",
4620 void notify_peer_device_state(struct sk_buff *skb,
4622 struct drbd_peer_device *peer_device,
4623 struct peer_device_info *peer_device_info,
4624 enum drbd_notification_type type)
4626 struct peer_device_statistics peer_device_statistics;
4627 struct drbd_resource *resource = peer_device->device->resource;
4628 struct drbd_genlmsghdr *dh;
4629 bool multicast = false;
4633 seq = atomic_inc_return(¬ify_genl_seq);
4634 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4642 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_PEER_DEVICE_STATE);
4644 goto nla_put_failure;
4646 dh->ret_code = NO_ERROR;
4647 if (nla_put_drbd_cfg_context(skb, resource, peer_device->connection, peer_device->device) ||
4648 nla_put_notification_header(skb, type) ||
4649 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4650 peer_device_info_to_skb(skb, peer_device_info, true)))
4651 goto nla_put_failure;
4652 peer_device_to_statistics(&peer_device_statistics, peer_device);
4653 peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
4654 genlmsg_end(skb, dh);
4656 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4657 /* skb has been consumed or freed in netlink_broadcast() */
4658 if (err && err != -ESRCH)
4666 drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n",
4670 void notify_helper(enum drbd_notification_type type,
4671 struct drbd_device *device, struct drbd_connection *connection,
4672 const char *name, int status)
4674 struct drbd_resource *resource = device ? device->resource : connection->resource;
4675 struct drbd_helper_info helper_info;
4676 unsigned int seq = atomic_inc_return(¬ify_genl_seq);
4677 struct sk_buff *skb = NULL;
4678 struct drbd_genlmsghdr *dh;
4681 strlcpy(helper_info.helper_name, name, sizeof(helper_info.helper_name));
4682 helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name));
4683 helper_info.helper_status = status;
4685 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4691 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_HELPER);
4694 dh->minor = device ? device->minor : -1;
4695 dh->ret_code = NO_ERROR;
4696 mutex_lock(¬ification_mutex);
4697 if (nla_put_drbd_cfg_context(skb, resource, connection, device) ||
4698 nla_put_notification_header(skb, type) ||
4699 drbd_helper_info_to_skb(skb, &helper_info, true))
4701 genlmsg_end(skb, dh);
4702 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4704 /* skb has been consumed or freed in netlink_broadcast() */
4705 if (err && err != -ESRCH)
4707 mutex_unlock(¬ification_mutex);
4711 mutex_unlock(¬ification_mutex);
4714 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4718 static void notify_initial_state_done(struct sk_buff *skb, unsigned int seq)
4720 struct drbd_genlmsghdr *dh;
4724 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_INITIAL_STATE_DONE);
4726 goto nla_put_failure;
4728 dh->ret_code = NO_ERROR;
4729 if (nla_put_notification_header(skb, NOTIFY_EXISTS))
4730 goto nla_put_failure;
4731 genlmsg_end(skb, dh);
4736 pr_err("Error %d sending event. Event seq:%u\n", err, seq);
4739 static void free_state_changes(struct list_head *list)
4741 while (!list_empty(list)) {
4742 struct drbd_state_change *state_change =
4743 list_first_entry(list, struct drbd_state_change, list);
4744 list_del(&state_change->list);
4745 forget_state_change(state_change);
4749 static unsigned int notifications_for_state_change(struct drbd_state_change *state_change)
4752 state_change->n_connections +
4753 state_change->n_devices +
4754 state_change->n_devices * state_change->n_connections;
4757 static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4759 struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0];
4760 unsigned int seq = cb->args[2];
4762 enum drbd_notification_type flags = 0;
4764 /* There is no need for taking notification_mutex here: it doesn't
4765 matter if the initial state events mix with later state chage
4766 events; we can always tell the events apart by the NOTIFY_EXISTS
4770 if (cb->args[5] == 1) {
4771 notify_initial_state_done(skb, seq);
4775 if (cb->args[4] < cb->args[3])
4776 flags |= NOTIFY_CONTINUES;
4778 notify_resource_state_change(skb, seq, state_change->resource,
4779 NOTIFY_EXISTS | flags);
4783 if (n < state_change->n_connections) {
4784 notify_connection_state_change(skb, seq, &state_change->connections[n],
4785 NOTIFY_EXISTS | flags);
4788 n -= state_change->n_connections;
4789 if (n < state_change->n_devices) {
4790 notify_device_state_change(skb, seq, &state_change->devices[n],
4791 NOTIFY_EXISTS | flags);
4794 n -= state_change->n_devices;
4795 if (n < state_change->n_devices * state_change->n_connections) {
4796 notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n],
4797 NOTIFY_EXISTS | flags);
4802 if (cb->args[4] == cb->args[3]) {
4803 struct drbd_state_change *next_state_change =
4804 list_entry(state_change->list.next,
4805 struct drbd_state_change, list);
4806 cb->args[0] = (long)next_state_change;
4807 cb->args[3] = notifications_for_state_change(next_state_change);
4814 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4816 struct drbd_resource *resource;
4819 if (cb->args[5] >= 1) {
4820 if (cb->args[5] > 1)
4821 return get_initial_state(skb, cb);
4823 struct drbd_state_change *state_change =
4824 (struct drbd_state_change *)cb->args[0];
4826 /* connect list to head */
4827 list_add(&head, &state_change->list);
4828 free_state_changes(&head);
4833 cb->args[5] = 2; /* number of iterations */
4834 mutex_lock(&resources_mutex);
4835 for_each_resource(resource, &drbd_resources) {
4836 struct drbd_state_change *state_change;
4838 state_change = remember_old_state(resource, GFP_KERNEL);
4839 if (!state_change) {
4840 if (!list_empty(&head))
4841 free_state_changes(&head);
4842 mutex_unlock(&resources_mutex);
4845 copy_old_to_new_state_change(state_change);
4846 list_add_tail(&state_change->list, &head);
4847 cb->args[5] += notifications_for_state_change(state_change);
4849 mutex_unlock(&resources_mutex);
4851 if (!list_empty(&head)) {
4852 struct drbd_state_change *state_change =
4853 list_entry(head.next, struct drbd_state_change, list);
4854 cb->args[0] = (long)state_change;
4855 cb->args[3] = notifications_for_state_change(state_change);
4856 list_del(&head); /* detach list from head */
4859 cb->args[2] = cb->nlh->nlmsg_seq;
4860 return get_initial_state(skb, cb);