1 // SPDX-License-Identifier: GPL-2.0-or-later
5 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
7 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
8 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
9 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16 #include <linux/module.h>
17 #include <linux/drbd.h>
20 #include <linux/file.h>
21 #include <linux/slab.h>
22 #include <linux/blkpg.h>
23 #include <linux/cpumask.h>
25 #include "drbd_protocol.h"
27 #include "drbd_state_change.h"
28 #include <asm/unaligned.h>
29 #include <linux/drbd_limits.h>
30 #include <linux/kthread.h>
32 #include <net/genetlink.h>
35 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
36 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
38 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info);
39 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info);
41 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info);
42 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info);
43 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
45 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
46 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
47 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
48 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
49 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
50 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
51 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
52 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
53 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
54 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
55 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
56 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
57 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
58 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
59 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
60 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
61 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
62 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
63 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
64 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
66 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
67 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb);
68 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb);
69 int drbd_adm_dump_devices_done(struct netlink_callback *cb);
70 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb);
71 int drbd_adm_dump_connections_done(struct netlink_callback *cb);
72 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb);
73 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb);
74 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb);
76 #include <linux/drbd_genl_api.h>
78 #include <linux/genl_magic_func.h>
80 static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
81 static atomic_t notify_genl_seq = ATOMIC_INIT(2); /* two. */
83 DEFINE_MUTEX(notification_mutex);
85 /* used blkdev_get_by_path, to claim our meta data device(s) */
86 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
88 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
90 genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
91 if (genlmsg_reply(skb, info))
92 pr_err("error sending genl reply\n");
95 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
96 * reason it could fail was no space in skb, and there are 4k available. */
97 static int drbd_msg_put_info(struct sk_buff *skb, const char *info)
102 if (!info || !info[0])
105 nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_REPLY);
109 err = nla_put_string(skb, T_info_text, info);
111 nla_nest_cancel(skb, nla);
114 nla_nest_end(skb, nla);
119 static int drbd_msg_sprintf_info(struct sk_buff *skb, const char *fmt, ...)
122 struct nlattr *nla, *txt;
126 nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_REPLY);
130 txt = nla_reserve(skb, T_info_text, 256);
132 nla_nest_cancel(skb, nla);
136 len = vscnprintf(nla_data(txt), 256, fmt, args);
139 /* maybe: retry with larger reserve, if truncated */
140 txt->nla_len = nla_attr_size(len+1);
141 nlmsg_trim(skb, (char*)txt + NLA_ALIGN(txt->nla_len));
142 nla_nest_end(skb, nla);
147 /* This would be a good candidate for a "pre_doit" hook,
148 * and per-family private info->pointers.
149 * But we need to stay compatible with older kernels.
150 * If it returns successfully, adm_ctx members are valid.
152 * At this point, we still rely on the global genl_lock().
153 * If we want to avoid that, and allow "genl_family.parallel_ops", we may need
154 * to add additional synchronization against object destruction/modification.
156 #define DRBD_ADM_NEED_MINOR 1
157 #define DRBD_ADM_NEED_RESOURCE 2
158 #define DRBD_ADM_NEED_CONNECTION 4
159 static int drbd_adm_prepare(struct drbd_config_context *adm_ctx,
160 struct sk_buff *skb, struct genl_info *info, unsigned flags)
162 struct drbd_genlmsghdr *d_in = info->userhdr;
163 const u8 cmd = info->genlhdr->cmd;
166 memset(adm_ctx, 0, sizeof(*adm_ctx));
168 /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
169 if (cmd != DRBD_ADM_GET_STATUS && !capable(CAP_NET_ADMIN))
172 adm_ctx->reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
173 if (!adm_ctx->reply_skb) {
178 adm_ctx->reply_dh = genlmsg_put_reply(adm_ctx->reply_skb,
179 info, &drbd_genl_family, 0, cmd);
180 /* put of a few bytes into a fresh skb of >= 4k will always succeed.
182 if (!adm_ctx->reply_dh) {
187 adm_ctx->reply_dh->minor = d_in->minor;
188 adm_ctx->reply_dh->ret_code = NO_ERROR;
190 adm_ctx->volume = VOLUME_UNSPECIFIED;
191 if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
193 /* parse and validate only */
194 err = drbd_cfg_context_from_attrs(NULL, info);
198 /* It was present, and valid,
199 * copy it over to the reply skb. */
200 err = nla_put_nohdr(adm_ctx->reply_skb,
201 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
202 info->attrs[DRBD_NLA_CFG_CONTEXT]);
206 /* and assign stuff to the adm_ctx */
207 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
209 adm_ctx->volume = nla_get_u32(nla);
210 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
212 adm_ctx->resource_name = nla_data(nla);
213 adm_ctx->my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)];
214 adm_ctx->peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)];
215 if ((adm_ctx->my_addr &&
216 nla_len(adm_ctx->my_addr) > sizeof(adm_ctx->connection->my_addr)) ||
217 (adm_ctx->peer_addr &&
218 nla_len(adm_ctx->peer_addr) > sizeof(adm_ctx->connection->peer_addr))) {
224 adm_ctx->minor = d_in->minor;
225 adm_ctx->device = minor_to_device(d_in->minor);
227 /* We are protected by the global genl_lock().
228 * But we may explicitly drop it/retake it in drbd_adm_set_role(),
229 * so make sure this object stays around. */
231 kref_get(&adm_ctx->device->kref);
233 if (adm_ctx->resource_name) {
234 adm_ctx->resource = drbd_find_resource(adm_ctx->resource_name);
237 if (!adm_ctx->device && (flags & DRBD_ADM_NEED_MINOR)) {
238 drbd_msg_put_info(adm_ctx->reply_skb, "unknown minor");
239 return ERR_MINOR_INVALID;
241 if (!adm_ctx->resource && (flags & DRBD_ADM_NEED_RESOURCE)) {
242 drbd_msg_put_info(adm_ctx->reply_skb, "unknown resource");
243 if (adm_ctx->resource_name)
244 return ERR_RES_NOT_KNOWN;
245 return ERR_INVALID_REQUEST;
248 if (flags & DRBD_ADM_NEED_CONNECTION) {
249 if (adm_ctx->resource) {
250 drbd_msg_put_info(adm_ctx->reply_skb, "no resource name expected");
251 return ERR_INVALID_REQUEST;
253 if (adm_ctx->device) {
254 drbd_msg_put_info(adm_ctx->reply_skb, "no minor number expected");
255 return ERR_INVALID_REQUEST;
257 if (adm_ctx->my_addr && adm_ctx->peer_addr)
258 adm_ctx->connection = conn_get_by_addrs(nla_data(adm_ctx->my_addr),
259 nla_len(adm_ctx->my_addr),
260 nla_data(adm_ctx->peer_addr),
261 nla_len(adm_ctx->peer_addr));
262 if (!adm_ctx->connection) {
263 drbd_msg_put_info(adm_ctx->reply_skb, "unknown connection");
264 return ERR_INVALID_REQUEST;
268 /* some more paranoia, if the request was over-determined */
269 if (adm_ctx->device && adm_ctx->resource &&
270 adm_ctx->device->resource != adm_ctx->resource) {
271 pr_warning("request: minor=%u, resource=%s; but that minor belongs to resource %s\n",
272 adm_ctx->minor, adm_ctx->resource->name,
273 adm_ctx->device->resource->name);
274 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists in different resource");
275 return ERR_INVALID_REQUEST;
277 if (adm_ctx->device &&
278 adm_ctx->volume != VOLUME_UNSPECIFIED &&
279 adm_ctx->volume != adm_ctx->device->vnr) {
280 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
281 adm_ctx->minor, adm_ctx->volume,
282 adm_ctx->device->vnr,
283 adm_ctx->device->resource->name);
284 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists as different volume");
285 return ERR_INVALID_REQUEST;
288 /* still, provide adm_ctx->resource always, if possible. */
289 if (!adm_ctx->resource) {
290 adm_ctx->resource = adm_ctx->device ? adm_ctx->device->resource
291 : adm_ctx->connection ? adm_ctx->connection->resource : NULL;
292 if (adm_ctx->resource)
293 kref_get(&adm_ctx->resource->kref);
299 nlmsg_free(adm_ctx->reply_skb);
300 adm_ctx->reply_skb = NULL;
304 static int drbd_adm_finish(struct drbd_config_context *adm_ctx,
305 struct genl_info *info, int retcode)
307 if (adm_ctx->device) {
308 kref_put(&adm_ctx->device->kref, drbd_destroy_device);
309 adm_ctx->device = NULL;
311 if (adm_ctx->connection) {
312 kref_put(&adm_ctx->connection->kref, &drbd_destroy_connection);
313 adm_ctx->connection = NULL;
315 if (adm_ctx->resource) {
316 kref_put(&adm_ctx->resource->kref, drbd_destroy_resource);
317 adm_ctx->resource = NULL;
320 if (!adm_ctx->reply_skb)
323 adm_ctx->reply_dh->ret_code = retcode;
324 drbd_adm_send_reply(adm_ctx->reply_skb, info);
328 static void setup_khelper_env(struct drbd_connection *connection, char **envp)
332 /* FIXME: A future version will not allow this case. */
333 if (connection->my_addr_len == 0 || connection->peer_addr_len == 0)
336 switch (((struct sockaddr *)&connection->peer_addr)->sa_family) {
339 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
340 &((struct sockaddr_in6 *)&connection->peer_addr)->sin6_addr);
344 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
345 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
349 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
350 &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
352 snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
355 int drbd_khelper(struct drbd_device *device, char *cmd)
357 char *envp[] = { "HOME=/",
359 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
360 (char[20]) { }, /* address family */
361 (char[60]) { }, /* address */
364 char *argv[] = {drbd_usermode_helper, cmd, mb, NULL };
365 struct drbd_connection *connection = first_peer_device(device)->connection;
369 if (current == connection->worker.task)
370 set_bit(CALLBACK_PENDING, &connection->flags);
372 snprintf(mb, 14, "minor-%d", device_to_minor(device));
373 setup_khelper_env(connection, envp);
375 /* The helper may take some time.
376 * write out any unsynced meta data changes now */
377 drbd_md_sync(device);
379 drbd_info(device, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, mb);
380 sib.sib_reason = SIB_HELPER_PRE;
381 sib.helper_name = cmd;
382 drbd_bcast_event(device, &sib);
383 notify_helper(NOTIFY_CALL, device, connection, cmd, 0);
384 ret = call_usermodehelper(drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
386 drbd_warn(device, "helper command: %s %s %s exit code %u (0x%x)\n",
387 drbd_usermode_helper, cmd, mb,
388 (ret >> 8) & 0xff, ret);
390 drbd_info(device, "helper command: %s %s %s exit code %u (0x%x)\n",
391 drbd_usermode_helper, cmd, mb,
392 (ret >> 8) & 0xff, ret);
393 sib.sib_reason = SIB_HELPER_POST;
394 sib.helper_exit_code = ret;
395 drbd_bcast_event(device, &sib);
396 notify_helper(NOTIFY_RESPONSE, device, connection, cmd, ret);
398 if (current == connection->worker.task)
399 clear_bit(CALLBACK_PENDING, &connection->flags);
401 if (ret < 0) /* Ignore any ERRNOs we got. */
407 enum drbd_peer_state conn_khelper(struct drbd_connection *connection, char *cmd)
409 char *envp[] = { "HOME=/",
411 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
412 (char[20]) { }, /* address family */
413 (char[60]) { }, /* address */
415 char *resource_name = connection->resource->name;
416 char *argv[] = {drbd_usermode_helper, cmd, resource_name, NULL };
419 setup_khelper_env(connection, envp);
420 conn_md_sync(connection);
422 drbd_info(connection, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, resource_name);
423 /* TODO: conn_bcast_event() ?? */
424 notify_helper(NOTIFY_CALL, NULL, connection, cmd, 0);
426 ret = call_usermodehelper(drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
428 drbd_warn(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
429 drbd_usermode_helper, cmd, resource_name,
430 (ret >> 8) & 0xff, ret);
432 drbd_info(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
433 drbd_usermode_helper, cmd, resource_name,
434 (ret >> 8) & 0xff, ret);
435 /* TODO: conn_bcast_event() ?? */
436 notify_helper(NOTIFY_RESPONSE, NULL, connection, cmd, ret);
438 if (ret < 0) /* Ignore any ERRNOs we got. */
444 static enum drbd_fencing_p highest_fencing_policy(struct drbd_connection *connection)
446 enum drbd_fencing_p fp = FP_NOT_AVAIL;
447 struct drbd_peer_device *peer_device;
451 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
452 struct drbd_device *device = peer_device->device;
453 if (get_ldev_if_state(device, D_CONSISTENT)) {
454 struct disk_conf *disk_conf =
455 rcu_dereference(peer_device->device->ldev->disk_conf);
456 fp = max_t(enum drbd_fencing_p, fp, disk_conf->fencing);
465 static bool resource_is_supended(struct drbd_resource *resource)
467 return resource->susp || resource->susp_fen || resource->susp_nod;
470 bool conn_try_outdate_peer(struct drbd_connection *connection)
472 struct drbd_resource * const resource = connection->resource;
473 unsigned int connect_cnt;
474 union drbd_state mask = { };
475 union drbd_state val = { };
476 enum drbd_fencing_p fp;
480 spin_lock_irq(&resource->req_lock);
481 if (connection->cstate >= C_WF_REPORT_PARAMS) {
482 drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n");
483 spin_unlock_irq(&resource->req_lock);
487 connect_cnt = connection->connect_cnt;
488 spin_unlock_irq(&resource->req_lock);
490 fp = highest_fencing_policy(connection);
493 drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n");
494 spin_lock_irq(&resource->req_lock);
495 if (connection->cstate < C_WF_REPORT_PARAMS) {
496 _conn_request_state(connection,
497 (union drbd_state) { { .susp_fen = 1 } },
498 (union drbd_state) { { .susp_fen = 0 } },
499 CS_VERBOSE | CS_HARD | CS_DC_SUSP);
500 /* We are no longer suspended due to the fencing policy.
501 * We may still be suspended due to the on-no-data-accessible policy.
502 * If that was OND_IO_ERROR, fail pending requests. */
503 if (!resource_is_supended(resource))
504 _tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
506 /* Else: in case we raced with a connection handshake,
507 * let the handshake figure out if we maybe can RESEND,
508 * and do not resume/fail pending requests here.
509 * Worst case is we stay suspended for now, which may be
510 * resolved by either re-establishing the replication link, or
511 * the next link failure, or eventually the administrator. */
512 spin_unlock_irq(&resource->req_lock);
520 r = conn_khelper(connection, "fence-peer");
522 switch ((r>>8) & 0xff) {
523 case P_INCONSISTENT: /* peer is inconsistent */
524 ex_to_string = "peer is inconsistent or worse";
526 val.pdsk = D_INCONSISTENT;
528 case P_OUTDATED: /* peer got outdated, or was already outdated */
529 ex_to_string = "peer was fenced";
531 val.pdsk = D_OUTDATED;
533 case P_DOWN: /* peer was down */
534 if (conn_highest_disk(connection) == D_UP_TO_DATE) {
535 /* we will(have) create(d) a new UUID anyways... */
536 ex_to_string = "peer is unreachable, assumed to be dead";
538 val.pdsk = D_OUTDATED;
540 ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
543 case P_PRIMARY: /* Peer is primary, voluntarily outdate myself.
544 * This is useful when an unconnected R_SECONDARY is asked to
545 * become R_PRIMARY, but finds the other peer being active. */
546 ex_to_string = "peer is active";
547 drbd_warn(connection, "Peer is primary, outdating myself.\n");
549 val.disk = D_OUTDATED;
552 /* THINK: do we need to handle this
553 * like case 4, or more like case 5? */
554 if (fp != FP_STONITH)
555 drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n");
556 ex_to_string = "peer was stonithed";
558 val.pdsk = D_OUTDATED;
561 /* The script is broken ... */
562 drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
563 return false; /* Eventually leave IO frozen */
566 drbd_info(connection, "fence-peer helper returned %d (%s)\n",
567 (r>>8) & 0xff, ex_to_string);
570 conn_request_state(connection, mask, val, CS_VERBOSE);
571 here, because we might were able to re-establish the connection in the
573 spin_lock_irq(&resource->req_lock);
574 if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) {
575 if (connection->connect_cnt != connect_cnt)
576 /* In case the connection was established and droped
577 while the fence-peer handler was running, ignore it */
578 drbd_info(connection, "Ignoring fence-peer exit code\n");
580 _conn_request_state(connection, mask, val, CS_VERBOSE);
582 spin_unlock_irq(&resource->req_lock);
584 return conn_highest_pdsk(connection) <= D_OUTDATED;
587 static int _try_outdate_peer_async(void *data)
589 struct drbd_connection *connection = (struct drbd_connection *)data;
591 conn_try_outdate_peer(connection);
593 kref_put(&connection->kref, drbd_destroy_connection);
597 void conn_try_outdate_peer_async(struct drbd_connection *connection)
599 struct task_struct *opa;
601 kref_get(&connection->kref);
602 /* We may just have force_sig()'ed this thread
603 * to get it out of some blocking network function.
604 * Clear signals; otherwise kthread_run(), which internally uses
605 * wait_on_completion_killable(), will mistake our pending signal
606 * for a new fatal signal and fail. */
607 flush_signals(current);
608 opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h");
610 drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n");
611 kref_put(&connection->kref, drbd_destroy_connection);
616 drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force)
618 struct drbd_peer_device *const peer_device = first_peer_device(device);
619 struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL;
620 const int max_tries = 4;
621 enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
625 union drbd_state mask, val;
627 if (new_role == R_PRIMARY) {
628 struct drbd_connection *connection;
630 /* Detect dead peers as soon as possible. */
633 for_each_connection(connection, device->resource)
634 request_ping(connection);
638 mutex_lock(device->state_mutex);
640 mask.i = 0; mask.role = R_MASK;
641 val.i = 0; val.role = new_role;
643 while (try++ < max_tries) {
644 rv = _drbd_request_state_holding_state_mutex(device, mask, val, CS_WAIT_COMPLETE);
646 /* in case we first succeeded to outdate,
647 * but now suddenly could establish a connection */
648 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
654 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
655 (device->state.disk < D_UP_TO_DATE &&
656 device->state.disk >= D_INCONSISTENT)) {
658 val.disk = D_UP_TO_DATE;
663 if (rv == SS_NO_UP_TO_DATE_DISK &&
664 device->state.disk == D_CONSISTENT && mask.pdsk == 0) {
665 D_ASSERT(device, device->state.pdsk == D_UNKNOWN);
667 if (conn_try_outdate_peer(connection)) {
668 val.disk = D_UP_TO_DATE;
674 if (rv == SS_NOTHING_TO_DO)
676 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
677 if (!conn_try_outdate_peer(connection) && force) {
678 drbd_warn(device, "Forced into split brain situation!\n");
680 val.pdsk = D_OUTDATED;
685 if (rv == SS_TWO_PRIMARIES) {
686 /* Maybe the peer is detected as dead very soon...
687 retry at most once more in this case. */
688 if (try < max_tries) {
692 nc = rcu_dereference(connection->net_conf);
693 timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
695 schedule_timeout_interruptible(timeo);
699 if (rv < SS_SUCCESS) {
700 rv = _drbd_request_state(device, mask, val,
701 CS_VERBOSE + CS_WAIT_COMPLETE);
712 drbd_warn(device, "Forced to consider local data as UpToDate!\n");
714 /* Wait until nothing is on the fly :) */
715 wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0);
717 /* FIXME also wait for all pending P_BARRIER_ACK? */
719 if (new_role == R_SECONDARY) {
720 if (get_ldev(device)) {
721 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
725 mutex_lock(&device->resource->conf_update);
726 nc = connection->net_conf;
728 nc->discard_my_data = 0; /* without copy; single bit op is atomic */
729 mutex_unlock(&device->resource->conf_update);
731 if (get_ldev(device)) {
732 if (((device->state.conn < C_CONNECTED ||
733 device->state.pdsk <= D_FAILED)
734 && device->ldev->md.uuid[UI_BITMAP] == 0) || forced)
735 drbd_uuid_new_current(device);
737 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
742 /* writeout of activity log covered areas of the bitmap
743 * to stable storage done in after state change already */
745 if (device->state.conn >= C_WF_REPORT_PARAMS) {
746 /* if this was forced, we should consider sync */
748 drbd_send_uuids(peer_device);
749 drbd_send_current_state(peer_device);
752 drbd_md_sync(device);
753 set_disk_ro(device->vdisk, new_role == R_SECONDARY);
754 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
756 mutex_unlock(device->state_mutex);
760 static const char *from_attrs_err_to_txt(int err)
762 return err == -ENOMSG ? "required attribute missing" :
763 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
764 err == -EEXIST ? "can not change invariant setting" :
765 "invalid attribute value";
768 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
770 struct drbd_config_context adm_ctx;
771 struct set_role_parms parms;
773 enum drbd_ret_code retcode;
775 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
776 if (!adm_ctx.reply_skb)
778 if (retcode != NO_ERROR)
781 memset(&parms, 0, sizeof(parms));
782 if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
783 err = set_role_parms_from_attrs(&parms, info);
785 retcode = ERR_MANDATORY_TAG;
786 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
791 mutex_lock(&adm_ctx.resource->adm_mutex);
793 if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
794 retcode = drbd_set_role(adm_ctx.device, R_PRIMARY, parms.assume_uptodate);
796 retcode = drbd_set_role(adm_ctx.device, R_SECONDARY, 0);
798 mutex_unlock(&adm_ctx.resource->adm_mutex);
801 drbd_adm_finish(&adm_ctx, info, retcode);
805 /* Initializes the md.*_offset members, so we are able to find
806 * the on disk meta data.
808 * We currently have two possible layouts:
810 * |----------- md_size_sect ------------------|
811 * [ 4k superblock ][ activity log ][ Bitmap ]
813 * | bm_offset = al_offset + X |
814 * ==> bitmap sectors = md_size_sect - bm_offset
817 * |----------- md_size_sect ------------------|
818 * [data.....][ Bitmap ][ activity log ][ 4k superblock ]
820 * | bm_offset = al_offset - Y |
821 * ==> bitmap sectors = Y = al_offset - bm_offset
823 * Activity log size used to be fixed 32kB,
824 * but is about to become configurable.
826 static void drbd_md_set_sector_offsets(struct drbd_device *device,
827 struct drbd_backing_dev *bdev)
829 sector_t md_size_sect = 0;
830 unsigned int al_size_sect = bdev->md.al_size_4k * 8;
832 bdev->md.md_offset = drbd_md_ss(bdev);
834 switch (bdev->md.meta_dev_idx) {
836 /* v07 style fixed size indexed meta data */
837 bdev->md.md_size_sect = MD_128MB_SECT;
838 bdev->md.al_offset = MD_4kB_SECT;
839 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
841 case DRBD_MD_INDEX_FLEX_EXT:
842 /* just occupy the full device; unit: sectors */
843 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
844 bdev->md.al_offset = MD_4kB_SECT;
845 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
847 case DRBD_MD_INDEX_INTERNAL:
848 case DRBD_MD_INDEX_FLEX_INT:
849 /* al size is still fixed */
850 bdev->md.al_offset = -al_size_sect;
851 /* we need (slightly less than) ~ this much bitmap sectors: */
852 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
853 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
854 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
855 md_size_sect = ALIGN(md_size_sect, 8);
857 /* plus the "drbd meta data super block",
858 * and the activity log; */
859 md_size_sect += MD_4kB_SECT + al_size_sect;
861 bdev->md.md_size_sect = md_size_sect;
862 /* bitmap offset is adjusted by 'super' block size */
863 bdev->md.bm_offset = -md_size_sect + MD_4kB_SECT;
868 /* input size is expected to be in KB */
869 char *ppsize(char *buf, unsigned long long size)
871 /* Needs 9 bytes at max including trailing NUL:
872 * -1ULL ==> "16384 EB" */
873 static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
875 while (size >= 10000 && base < sizeof(units)-1) {
877 size = (size >> 10) + !!(size & (1<<9));
880 sprintf(buf, "%u %cB", (unsigned)size, units[base]);
885 /* there is still a theoretical deadlock when called from receiver
886 * on an D_INCONSISTENT R_PRIMARY:
887 * remote READ does inc_ap_bio, receiver would need to receive answer
888 * packet from remote to dec_ap_bio again.
889 * receiver receive_sizes(), comes here,
890 * waits for ap_bio_cnt == 0. -> deadlock.
891 * but this cannot happen, actually, because:
892 * R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
893 * (not connected, or bad/no disk on peer):
894 * see drbd_fail_request_early, ap_bio_cnt is zero.
895 * R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
896 * peer may not initiate a resize.
898 /* Note these are not to be confused with
899 * drbd_adm_suspend_io/drbd_adm_resume_io,
900 * which are (sub) state changes triggered by admin (drbdsetup),
901 * and can be long lived.
902 * This changes an device->flag, is triggered by drbd internals,
903 * and should be short-lived. */
904 /* It needs to be a counter, since multiple threads might
905 independently suspend and resume IO. */
906 void drbd_suspend_io(struct drbd_device *device)
908 atomic_inc(&device->suspend_cnt);
909 if (drbd_suspended(device))
911 wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt));
914 void drbd_resume_io(struct drbd_device *device)
916 if (atomic_dec_and_test(&device->suspend_cnt))
917 wake_up(&device->misc_wait);
921 * drbd_determine_dev_size() - Sets the right device size obeying all constraints
922 * @device: DRBD device.
924 * Returns 0 on success, negative return values indicate errors.
925 * You should call drbd_md_sync() after calling this function.
927 enum determine_dev_size
928 drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local)
930 struct md_offsets_and_sizes {
931 u64 last_agreed_sect;
938 u32 al_stripe_size_4k;
940 sector_t u_size, size;
941 struct drbd_md *md = &device->ldev->md;
944 int md_moved, la_size_changed;
945 enum determine_dev_size rv = DS_UNCHANGED;
947 /* We may change the on-disk offsets of our meta data below. Lock out
948 * anything that may cause meta data IO, to avoid acting on incomplete
949 * layout changes or scribbling over meta data that is in the process
952 * Move is not exactly correct, btw, currently we have all our meta
953 * data in core memory, to "move" it we just write it all out, there
955 drbd_suspend_io(device);
956 buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */
958 drbd_resume_io(device);
962 /* remember current offset and sizes */
963 prev.last_agreed_sect = md->la_size_sect;
964 prev.md_offset = md->md_offset;
965 prev.al_offset = md->al_offset;
966 prev.bm_offset = md->bm_offset;
967 prev.md_size_sect = md->md_size_sect;
968 prev.al_stripes = md->al_stripes;
969 prev.al_stripe_size_4k = md->al_stripe_size_4k;
972 /* rs is non NULL if we should change the AL layout only */
973 md->al_stripes = rs->al_stripes;
974 md->al_stripe_size_4k = rs->al_stripe_size / 4;
975 md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4;
978 drbd_md_set_sector_offsets(device, device->ldev);
981 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
983 size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED);
985 if (size < prev.last_agreed_sect) {
986 if (rs && u_size == 0) {
987 /* Remove "rs &&" later. This check should always be active, but
988 right now the receiver expects the permissive behavior */
989 drbd_warn(device, "Implicit shrink not allowed. "
990 "Use --size=%llus for explicit shrink.\n",
991 (unsigned long long)size);
992 rv = DS_ERROR_SHRINK;
995 rv = DS_ERROR_SPACE_MD;
996 if (rv != DS_UNCHANGED)
1000 if (drbd_get_capacity(device->this_bdev) != size ||
1001 drbd_bm_capacity(device) != size) {
1003 err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC));
1004 if (unlikely(err)) {
1005 /* currently there is only one error: ENOMEM! */
1006 size = drbd_bm_capacity(device);
1008 drbd_err(device, "OUT OF MEMORY! "
1009 "Could not allocate bitmap!\n");
1011 drbd_err(device, "BM resizing failed. "
1012 "Leaving size unchanged\n");
1016 /* racy, see comments above. */
1017 drbd_set_my_capacity(device, size);
1018 md->la_size_sect = size;
1023 la_size_changed = (prev.last_agreed_sect != md->la_size_sect);
1025 md_moved = prev.md_offset != md->md_offset
1026 || prev.md_size_sect != md->md_size_sect;
1028 if (la_size_changed || md_moved || rs) {
1031 /* We do some synchronous IO below, which may take some time.
1032 * Clear the timer, to avoid scary "timer expired!" messages,
1033 * "Superblock" is written out at least twice below, anyways. */
1034 del_timer(&device->md_sync_timer);
1036 /* We won't change the "al-extents" setting, we just may need
1037 * to move the on-disk location of the activity log ringbuffer.
1038 * Lock for transaction is good enough, it may well be "dirty"
1039 * or even "starving". */
1040 wait_event(device->al_wait, lc_try_lock_for_transaction(device->act_log));
1042 /* mark current on-disk bitmap and activity log as unreliable */
1043 prev_flags = md->flags;
1044 md->flags |= MDF_FULL_SYNC | MDF_AL_DISABLED;
1045 drbd_md_write(device, buffer);
1047 drbd_al_initialize(device, buffer);
1049 drbd_info(device, "Writing the whole bitmap, %s\n",
1050 la_size_changed && md_moved ? "size changed and md moved" :
1051 la_size_changed ? "size changed" : "md moved");
1052 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
1053 drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
1054 "size changed", BM_LOCKED_MASK);
1056 /* on-disk bitmap and activity log is authoritative again
1057 * (unless there was an IO error meanwhile...) */
1058 md->flags = prev_flags;
1059 drbd_md_write(device, buffer);
1062 drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
1063 md->al_stripes, md->al_stripe_size_4k * 4);
1066 if (size > prev.last_agreed_sect)
1067 rv = prev.last_agreed_sect ? DS_GREW : DS_GREW_FROM_ZERO;
1068 if (size < prev.last_agreed_sect)
1073 /* restore previous offset and sizes */
1074 md->la_size_sect = prev.last_agreed_sect;
1075 md->md_offset = prev.md_offset;
1076 md->al_offset = prev.al_offset;
1077 md->bm_offset = prev.bm_offset;
1078 md->md_size_sect = prev.md_size_sect;
1079 md->al_stripes = prev.al_stripes;
1080 md->al_stripe_size_4k = prev.al_stripe_size_4k;
1081 md->al_size_4k = (u64)prev.al_stripes * prev.al_stripe_size_4k;
1083 lc_unlock(device->act_log);
1084 wake_up(&device->al_wait);
1085 drbd_md_put_buffer(device);
1086 drbd_resume_io(device);
1092 drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev,
1093 sector_t u_size, int assume_peer_has_space)
1095 sector_t p_size = device->p_size; /* partner's disk size. */
1096 sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
1097 sector_t m_size; /* my size */
1100 m_size = drbd_get_max_capacity(bdev);
1102 if (device->state.conn < C_CONNECTED && assume_peer_has_space) {
1103 drbd_warn(device, "Resize while not connected was forced by the user!\n");
1107 if (p_size && m_size) {
1108 size = min_t(sector_t, p_size, m_size);
1111 size = la_size_sect;
1112 if (m_size && m_size < size)
1114 if (p_size && p_size < size)
1125 drbd_err(device, "Both nodes diskless!\n");
1129 drbd_err(device, "Requested disk size is too big (%lu > %lu)\n",
1130 (unsigned long)u_size>>1, (unsigned long)size>>1);
1139 * drbd_check_al_size() - Ensures that the AL is of the right size
1140 * @device: DRBD device.
1142 * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1143 * failed, and 0 on success. You should call drbd_md_sync() after you called
1146 static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1148 struct lru_cache *n, *t;
1149 struct lc_element *e;
1150 unsigned int in_use;
1153 if (device->act_log &&
1154 device->act_log->nr_elements == dc->al_extents)
1158 t = device->act_log;
1159 n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
1160 dc->al_extents, sizeof(struct lc_element), 0);
1163 drbd_err(device, "Cannot allocate act_log lru!\n");
1166 spin_lock_irq(&device->al_lock);
1168 for (i = 0; i < t->nr_elements; i++) {
1169 e = lc_element_by_index(t, i);
1171 drbd_err(device, "refcnt(%d)==%d\n",
1172 e->lc_number, e->refcnt);
1173 in_use += e->refcnt;
1177 device->act_log = n;
1178 spin_unlock_irq(&device->al_lock);
1180 drbd_err(device, "Activity log still in use!\n");
1186 drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */
1190 static void blk_queue_discard_granularity(struct request_queue *q, unsigned int granularity)
1192 q->limits.discard_granularity = granularity;
1195 static unsigned int drbd_max_discard_sectors(struct drbd_connection *connection)
1197 /* when we introduced REQ_WRITE_SAME support, we also bumped
1198 * our maximum supported batch bio size used for discards. */
1199 if (connection->agreed_features & DRBD_FF_WSAME)
1200 return DRBD_MAX_BBIO_SECTORS;
1201 /* before, with DRBD <= 8.4.6, we only allowed up to one AL_EXTENT_SIZE. */
1202 return AL_EXTENT_SIZE >> 9;
1205 static void decide_on_discard_support(struct drbd_device *device,
1206 struct request_queue *q,
1207 struct request_queue *b,
1208 bool discard_zeroes_if_aligned)
1210 /* q = drbd device queue (device->rq_queue)
1211 * b = backing device queue (device->ldev->backing_bdev->bd_disk->queue),
1212 * or NULL if diskless
1214 struct drbd_connection *connection = first_peer_device(device)->connection;
1215 bool can_do = b ? blk_queue_discard(b) : true;
1217 if (can_do && connection->cstate >= C_CONNECTED && !(connection->agreed_features & DRBD_FF_TRIM)) {
1219 drbd_info(connection, "peer DRBD too old, does not support TRIM: disabling discards\n");
1222 /* We don't care for the granularity, really.
1223 * Stacking limits below should fix it for the local
1224 * device. Whether or not it is a suitable granularity
1225 * on the remote device is not our problem, really. If
1226 * you care, you need to use devices with similar
1227 * topology on all peers. */
1228 blk_queue_discard_granularity(q, 512);
1229 q->limits.max_discard_sectors = drbd_max_discard_sectors(connection);
1230 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1231 q->limits.max_write_zeroes_sectors = drbd_max_discard_sectors(connection);
1233 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1234 blk_queue_discard_granularity(q, 0);
1235 q->limits.max_discard_sectors = 0;
1236 q->limits.max_write_zeroes_sectors = 0;
1240 static void fixup_discard_if_not_supported(struct request_queue *q)
1242 /* To avoid confusion, if this queue does not support discard, clear
1243 * max_discard_sectors, which is what lsblk -D reports to the user.
1244 * Older kernels got this wrong in "stack limits".
1246 if (!blk_queue_discard(q)) {
1247 blk_queue_max_discard_sectors(q, 0);
1248 blk_queue_discard_granularity(q, 0);
1252 static void fixup_write_zeroes(struct drbd_device *device, struct request_queue *q)
1254 /* Fixup max_write_zeroes_sectors after blk_queue_stack_limits():
1255 * if we can handle "zeroes" efficiently on the protocol,
1256 * we want to do that, even if our backend does not announce
1257 * max_write_zeroes_sectors itself. */
1258 struct drbd_connection *connection = first_peer_device(device)->connection;
1259 /* If the peer announces WZEROES support, use it. Otherwise, rather
1260 * send explicit zeroes than rely on some discard-zeroes-data magic. */
1261 if (connection->agreed_features & DRBD_FF_WZEROES)
1262 q->limits.max_write_zeroes_sectors = DRBD_MAX_BBIO_SECTORS;
1264 q->limits.max_write_zeroes_sectors = 0;
1267 static void decide_on_write_same_support(struct drbd_device *device,
1268 struct request_queue *q,
1269 struct request_queue *b, struct o_qlim *o,
1270 bool disable_write_same)
1272 struct drbd_peer_device *peer_device = first_peer_device(device);
1273 struct drbd_connection *connection = peer_device->connection;
1274 bool can_do = b ? b->limits.max_write_same_sectors : true;
1276 if (can_do && disable_write_same) {
1278 drbd_info(peer_device, "WRITE_SAME disabled by config\n");
1281 if (can_do && connection->cstate >= C_CONNECTED && !(connection->agreed_features & DRBD_FF_WSAME)) {
1283 drbd_info(peer_device, "peer does not support WRITE_SAME\n");
1287 /* logical block size; queue_logical_block_size(NULL) is 512 */
1288 unsigned int peer_lbs = be32_to_cpu(o->logical_block_size);
1289 unsigned int me_lbs_b = queue_logical_block_size(b);
1290 unsigned int me_lbs = queue_logical_block_size(q);
1292 if (me_lbs_b != me_lbs) {
1294 "logical block size of local backend does not match (drbd:%u, backend:%u); was this a late attach?\n",
1296 /* rather disable write same than trigger some BUG_ON later in the scsi layer. */
1299 if (me_lbs_b != peer_lbs) {
1300 drbd_warn(peer_device, "logical block sizes do not match (me:%u, peer:%u); this may cause problems.\n",
1303 drbd_dbg(peer_device, "logical block size mismatch: WRITE_SAME disabled.\n");
1306 me_lbs = max(me_lbs, me_lbs_b);
1307 /* We cannot change the logical block size of an in-use queue.
1308 * We can only hope that access happens to be properly aligned.
1309 * If not, the peer will likely produce an IO error, and detach. */
1310 if (peer_lbs > me_lbs) {
1311 if (device->state.role != R_PRIMARY) {
1312 blk_queue_logical_block_size(q, peer_lbs);
1313 drbd_warn(peer_device, "logical block size set to %u\n", peer_lbs);
1315 drbd_warn(peer_device,
1316 "current Primary must NOT adjust logical block size (%u -> %u); hope for the best.\n",
1321 if (can_do && !o->write_same_capable) {
1322 /* If we introduce an open-coded write-same loop on the receiving side,
1323 * the peer would present itself as "capable". */
1324 drbd_dbg(peer_device, "WRITE_SAME disabled (peer device not capable)\n");
1329 blk_queue_max_write_same_sectors(q, can_do ? DRBD_MAX_BBIO_SECTORS : 0);
1332 static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev,
1333 unsigned int max_bio_size, struct o_qlim *o)
1335 struct request_queue * const q = device->rq_queue;
1336 unsigned int max_hw_sectors = max_bio_size >> 9;
1337 unsigned int max_segments = 0;
1338 struct request_queue *b = NULL;
1339 struct disk_conf *dc;
1340 bool discard_zeroes_if_aligned = true;
1341 bool disable_write_same = false;
1344 b = bdev->backing_bdev->bd_disk->queue;
1346 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1348 dc = rcu_dereference(device->ldev->disk_conf);
1349 max_segments = dc->max_bio_bvecs;
1350 discard_zeroes_if_aligned = dc->discard_zeroes_if_aligned;
1351 disable_write_same = dc->disable_write_same;
1354 blk_set_stacking_limits(&q->limits);
1357 blk_queue_max_hw_sectors(q, max_hw_sectors);
1358 /* This is the workaround for "bio would need to, but cannot, be split" */
1359 blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1360 blk_queue_segment_boundary(q, PAGE_SIZE-1);
1361 decide_on_discard_support(device, q, b, discard_zeroes_if_aligned);
1362 decide_on_write_same_support(device, q, b, o, disable_write_same);
1365 blk_queue_stack_limits(q, b);
1367 if (q->backing_dev_info->ra_pages !=
1368 b->backing_dev_info->ra_pages) {
1369 drbd_info(device, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1370 q->backing_dev_info->ra_pages,
1371 b->backing_dev_info->ra_pages);
1372 q->backing_dev_info->ra_pages =
1373 b->backing_dev_info->ra_pages;
1376 fixup_discard_if_not_supported(q);
1377 fixup_write_zeroes(device, q);
1380 void drbd_reconsider_queue_parameters(struct drbd_device *device, struct drbd_backing_dev *bdev, struct o_qlim *o)
1382 unsigned int now, new, local, peer;
1384 now = queue_max_hw_sectors(device->rq_queue) << 9;
1385 local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */
1386 peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */
1389 local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9;
1390 device->local_max_bio_size = local;
1392 local = min(local, DRBD_MAX_BIO_SIZE);
1394 /* We may ignore peer limits if the peer is modern enough.
1395 Because new from 8.3.8 onwards the peer can use multiple
1396 BIOs for a single peer_request */
1397 if (device->state.conn >= C_WF_REPORT_PARAMS) {
1398 if (first_peer_device(device)->connection->agreed_pro_version < 94)
1399 peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1400 /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1401 else if (first_peer_device(device)->connection->agreed_pro_version == 94)
1402 peer = DRBD_MAX_SIZE_H80_PACKET;
1403 else if (first_peer_device(device)->connection->agreed_pro_version < 100)
1404 peer = DRBD_MAX_BIO_SIZE_P95; /* drbd 8.3.8 onwards, before 8.4.0 */
1406 peer = DRBD_MAX_BIO_SIZE;
1408 /* We may later detach and re-attach on a disconnected Primary.
1409 * Avoid this setting to jump back in that case.
1410 * We want to store what we know the peer DRBD can handle,
1411 * not what the peer IO backend can handle. */
1412 if (peer > device->peer_max_bio_size)
1413 device->peer_max_bio_size = peer;
1415 new = min(local, peer);
1417 if (device->state.role == R_PRIMARY && new < now)
1418 drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1421 drbd_info(device, "max BIO size = %u\n", new);
1423 drbd_setup_queue_param(device, bdev, new, o);
1426 /* Starts the worker thread */
1427 static void conn_reconfig_start(struct drbd_connection *connection)
1429 drbd_thread_start(&connection->worker);
1430 drbd_flush_workqueue(&connection->sender_work);
1433 /* if still unconfigured, stops worker again. */
1434 static void conn_reconfig_done(struct drbd_connection *connection)
1437 spin_lock_irq(&connection->resource->req_lock);
1438 stop_threads = conn_all_vols_unconf(connection) &&
1439 connection->cstate == C_STANDALONE;
1440 spin_unlock_irq(&connection->resource->req_lock);
1442 /* ack_receiver thread and ack_sender workqueue are implicitly
1443 * stopped by receiver in conn_disconnect() */
1444 drbd_thread_stop(&connection->receiver);
1445 drbd_thread_stop(&connection->worker);
1449 /* Make sure IO is suspended before calling this function(). */
1450 static void drbd_suspend_al(struct drbd_device *device)
1454 if (!lc_try_lock(device->act_log)) {
1455 drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1459 drbd_al_shrink(device);
1460 spin_lock_irq(&device->resource->req_lock);
1461 if (device->state.conn < C_CONNECTED)
1462 s = !test_and_set_bit(AL_SUSPENDED, &device->flags);
1463 spin_unlock_irq(&device->resource->req_lock);
1464 lc_unlock(device->act_log);
1467 drbd_info(device, "Suspended AL updates\n");
1471 static bool should_set_defaults(struct genl_info *info)
1473 unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1474 return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1477 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1479 /* This is limited by 16 bit "slot" numbers,
1480 * and by available on-disk context storage.
1482 * Also (u16)~0 is special (denotes a "free" extent).
1484 * One transaction occupies one 4kB on-disk block,
1485 * we have n such blocks in the on disk ring buffer,
1486 * the "current" transaction may fail (n-1),
1487 * and there is 919 slot numbers context information per transaction.
1489 * 72 transaction blocks amounts to more than 2**16 context slots,
1490 * so cap there first.
1492 const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1493 const unsigned int sufficient_on_disk =
1494 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1495 /AL_CONTEXT_PER_TRANSACTION;
1497 unsigned int al_size_4k = bdev->md.al_size_4k;
1499 if (al_size_4k > sufficient_on_disk)
1502 return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1505 static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1507 return a->disk_barrier != b->disk_barrier ||
1508 a->disk_flushes != b->disk_flushes ||
1509 a->disk_drain != b->disk_drain;
1512 static void sanitize_disk_conf(struct drbd_device *device, struct disk_conf *disk_conf,
1513 struct drbd_backing_dev *nbc)
1515 struct request_queue * const q = nbc->backing_bdev->bd_disk->queue;
1517 if (disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1518 disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1519 if (disk_conf->al_extents > drbd_al_extents_max(nbc))
1520 disk_conf->al_extents = drbd_al_extents_max(nbc);
1522 if (!blk_queue_discard(q)) {
1523 if (disk_conf->rs_discard_granularity) {
1524 disk_conf->rs_discard_granularity = 0; /* disable feature */
1525 drbd_info(device, "rs_discard_granularity feature disabled\n");
1529 if (disk_conf->rs_discard_granularity) {
1530 int orig_value = disk_conf->rs_discard_granularity;
1533 if (q->limits.discard_granularity > disk_conf->rs_discard_granularity)
1534 disk_conf->rs_discard_granularity = q->limits.discard_granularity;
1536 remainder = disk_conf->rs_discard_granularity % q->limits.discard_granularity;
1537 disk_conf->rs_discard_granularity += remainder;
1539 if (disk_conf->rs_discard_granularity > q->limits.max_discard_sectors << 9)
1540 disk_conf->rs_discard_granularity = q->limits.max_discard_sectors << 9;
1542 if (disk_conf->rs_discard_granularity != orig_value)
1543 drbd_info(device, "rs_discard_granularity changed to %d\n",
1544 disk_conf->rs_discard_granularity);
1548 static int disk_opts_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1552 if (device->act_log &&
1553 device->act_log->nr_elements == dc->al_extents)
1556 drbd_suspend_io(device);
1557 /* If IO completion is currently blocked, we would likely wait
1558 * "forever" for the activity log to become unused. So we don't. */
1559 if (atomic_read(&device->ap_bio_cnt))
1562 wait_event(device->al_wait, lc_try_lock(device->act_log));
1563 drbd_al_shrink(device);
1564 err = drbd_check_al_size(device, dc);
1565 lc_unlock(device->act_log);
1566 wake_up(&device->al_wait);
1568 drbd_resume_io(device);
1572 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1574 struct drbd_config_context adm_ctx;
1575 enum drbd_ret_code retcode;
1576 struct drbd_device *device;
1577 struct disk_conf *new_disk_conf, *old_disk_conf;
1578 struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1581 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1582 if (!adm_ctx.reply_skb)
1584 if (retcode != NO_ERROR)
1587 device = adm_ctx.device;
1588 mutex_lock(&adm_ctx.resource->adm_mutex);
1590 /* we also need a disk
1591 * to change the options on */
1592 if (!get_ldev(device)) {
1593 retcode = ERR_NO_DISK;
1597 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1598 if (!new_disk_conf) {
1599 retcode = ERR_NOMEM;
1603 mutex_lock(&device->resource->conf_update);
1604 old_disk_conf = device->ldev->disk_conf;
1605 *new_disk_conf = *old_disk_conf;
1606 if (should_set_defaults(info))
1607 set_disk_conf_defaults(new_disk_conf);
1609 err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1610 if (err && err != -ENOMSG) {
1611 retcode = ERR_MANDATORY_TAG;
1612 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1616 if (!expect(new_disk_conf->resync_rate >= 1))
1617 new_disk_conf->resync_rate = 1;
1619 sanitize_disk_conf(device, new_disk_conf, device->ldev);
1621 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1622 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1624 fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1625 if (fifo_size != device->rs_plan_s->size) {
1626 new_plan = fifo_alloc(fifo_size);
1628 drbd_err(device, "kmalloc of fifo_buffer failed");
1629 retcode = ERR_NOMEM;
1634 err = disk_opts_check_al_size(device, new_disk_conf);
1636 /* Could be just "busy". Ignore?
1637 * Introduce dedicated error code? */
1638 drbd_msg_put_info(adm_ctx.reply_skb,
1639 "Try again without changing current al-extents setting");
1640 retcode = ERR_NOMEM;
1644 lock_all_resources();
1645 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1646 if (retcode == NO_ERROR) {
1647 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1648 drbd_resync_after_changed(device);
1650 unlock_all_resources();
1652 if (retcode != NO_ERROR)
1656 old_plan = device->rs_plan_s;
1657 rcu_assign_pointer(device->rs_plan_s, new_plan);
1660 mutex_unlock(&device->resource->conf_update);
1662 if (new_disk_conf->al_updates)
1663 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1665 device->ldev->md.flags |= MDF_AL_DISABLED;
1667 if (new_disk_conf->md_flushes)
1668 clear_bit(MD_NO_FUA, &device->flags);
1670 set_bit(MD_NO_FUA, &device->flags);
1672 if (write_ordering_changed(old_disk_conf, new_disk_conf))
1673 drbd_bump_write_ordering(device->resource, NULL, WO_BDEV_FLUSH);
1675 if (old_disk_conf->discard_zeroes_if_aligned != new_disk_conf->discard_zeroes_if_aligned
1676 || old_disk_conf->disable_write_same != new_disk_conf->disable_write_same)
1677 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
1679 drbd_md_sync(device);
1681 if (device->state.conn >= C_CONNECTED) {
1682 struct drbd_peer_device *peer_device;
1684 for_each_peer_device(peer_device, device)
1685 drbd_send_sync_param(peer_device);
1689 kfree(old_disk_conf);
1691 mod_timer(&device->request_timer, jiffies + HZ);
1695 mutex_unlock(&device->resource->conf_update);
1697 kfree(new_disk_conf);
1702 mutex_unlock(&adm_ctx.resource->adm_mutex);
1704 drbd_adm_finish(&adm_ctx, info, retcode);
1708 static struct block_device *open_backing_dev(struct drbd_device *device,
1709 const char *bdev_path, void *claim_ptr, bool do_bd_link)
1711 struct block_device *bdev;
1714 bdev = blkdev_get_by_path(bdev_path,
1715 FMODE_READ | FMODE_WRITE | FMODE_EXCL, claim_ptr);
1717 drbd_err(device, "open(\"%s\") failed with %ld\n",
1718 bdev_path, PTR_ERR(bdev));
1725 err = bd_link_disk_holder(bdev, device->vdisk);
1727 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1728 drbd_err(device, "bd_link_disk_holder(\"%s\", ...) failed with %d\n",
1730 bdev = ERR_PTR(err);
1735 static int open_backing_devices(struct drbd_device *device,
1736 struct disk_conf *new_disk_conf,
1737 struct drbd_backing_dev *nbc)
1739 struct block_device *bdev;
1741 bdev = open_backing_dev(device, new_disk_conf->backing_dev, device, true);
1743 return ERR_OPEN_DISK;
1744 nbc->backing_bdev = bdev;
1747 * meta_dev_idx >= 0: external fixed size, possibly multiple
1748 * drbd sharing one meta device. TODO in that case, paranoia
1749 * check that [md_bdev, meta_dev_idx] is not yet used by some
1750 * other drbd minor! (if you use drbd.conf + drbdadm, that
1751 * should check it for you already; but if you don't, or
1752 * someone fooled it, we need to double check here)
1754 bdev = open_backing_dev(device, new_disk_conf->meta_dev,
1755 /* claim ptr: device, if claimed exclusively; shared drbd_m_holder,
1756 * if potentially shared with other drbd minors */
1757 (new_disk_conf->meta_dev_idx < 0) ? (void*)device : (void*)drbd_m_holder,
1758 /* avoid double bd_claim_by_disk() for the same (source,target) tuple,
1759 * as would happen with internal metadata. */
1760 (new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_FLEX_INT &&
1761 new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_INTERNAL));
1763 return ERR_OPEN_MD_DISK;
1764 nbc->md_bdev = bdev;
1768 static void close_backing_dev(struct drbd_device *device, struct block_device *bdev,
1774 bd_unlink_disk_holder(bdev, device->vdisk);
1775 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1778 void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev)
1783 close_backing_dev(device, ldev->md_bdev, ldev->md_bdev != ldev->backing_bdev);
1784 close_backing_dev(device, ldev->backing_bdev, true);
1786 kfree(ldev->disk_conf);
1790 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1792 struct drbd_config_context adm_ctx;
1793 struct drbd_device *device;
1794 struct drbd_peer_device *peer_device;
1795 struct drbd_connection *connection;
1797 enum drbd_ret_code retcode;
1798 enum determine_dev_size dd;
1799 sector_t max_possible_sectors;
1800 sector_t min_md_device_sectors;
1801 struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1802 struct disk_conf *new_disk_conf = NULL;
1803 struct lru_cache *resync_lru = NULL;
1804 struct fifo_buffer *new_plan = NULL;
1805 union drbd_state ns, os;
1806 enum drbd_state_rv rv;
1807 struct net_conf *nc;
1809 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1810 if (!adm_ctx.reply_skb)
1812 if (retcode != NO_ERROR)
1815 device = adm_ctx.device;
1816 mutex_lock(&adm_ctx.resource->adm_mutex);
1817 peer_device = first_peer_device(device);
1818 connection = peer_device->connection;
1819 conn_reconfig_start(connection);
1821 /* if you want to reconfigure, please tear down first */
1822 if (device->state.disk > D_DISKLESS) {
1823 retcode = ERR_DISK_CONFIGURED;
1826 /* It may just now have detached because of IO error. Make sure
1827 * drbd_ldev_destroy is done already, we may end up here very fast,
1828 * e.g. if someone calls attach from the on-io-error handler,
1829 * to realize a "hot spare" feature (not that I'd recommend that) */
1830 wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1832 /* make sure there is no leftover from previous force-detach attempts */
1833 clear_bit(FORCE_DETACH, &device->flags);
1834 clear_bit(WAS_IO_ERROR, &device->flags);
1835 clear_bit(WAS_READ_ERROR, &device->flags);
1837 /* and no leftover from previously aborted resync or verify, either */
1838 device->rs_total = 0;
1839 device->rs_failed = 0;
1840 atomic_set(&device->rs_pending_cnt, 0);
1842 /* allocation not in the IO path, drbdsetup context */
1843 nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1845 retcode = ERR_NOMEM;
1848 spin_lock_init(&nbc->md.uuid_lock);
1850 new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1851 if (!new_disk_conf) {
1852 retcode = ERR_NOMEM;
1855 nbc->disk_conf = new_disk_conf;
1857 set_disk_conf_defaults(new_disk_conf);
1858 err = disk_conf_from_attrs(new_disk_conf, info);
1860 retcode = ERR_MANDATORY_TAG;
1861 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1865 if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1866 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1868 new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1870 retcode = ERR_NOMEM;
1874 if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1875 retcode = ERR_MD_IDX_INVALID;
1880 nc = rcu_dereference(connection->net_conf);
1882 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1884 retcode = ERR_STONITH_AND_PROT_A;
1890 retcode = open_backing_devices(device, new_disk_conf, nbc);
1891 if (retcode != NO_ERROR)
1894 if ((nbc->backing_bdev == nbc->md_bdev) !=
1895 (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1896 new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1897 retcode = ERR_MD_IDX_INVALID;
1901 resync_lru = lc_create("resync", drbd_bm_ext_cache,
1902 1, 61, sizeof(struct bm_extent),
1903 offsetof(struct bm_extent, lce));
1905 retcode = ERR_NOMEM;
1909 /* Read our meta data super block early.
1910 * This also sets other on-disk offsets. */
1911 retcode = drbd_md_read(device, nbc);
1912 if (retcode != NO_ERROR)
1915 sanitize_disk_conf(device, new_disk_conf, nbc);
1917 if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1918 drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1919 (unsigned long long) drbd_get_max_capacity(nbc),
1920 (unsigned long long) new_disk_conf->disk_size);
1921 retcode = ERR_DISK_TOO_SMALL;
1925 if (new_disk_conf->meta_dev_idx < 0) {
1926 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1927 /* at least one MB, otherwise it does not make sense */
1928 min_md_device_sectors = (2<<10);
1930 max_possible_sectors = DRBD_MAX_SECTORS;
1931 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1934 if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1935 retcode = ERR_MD_DISK_TOO_SMALL;
1936 drbd_warn(device, "refusing attach: md-device too small, "
1937 "at least %llu sectors needed for this meta-disk type\n",
1938 (unsigned long long) min_md_device_sectors);
1942 /* Make sure the new disk is big enough
1943 * (we may currently be R_PRIMARY with no local disk...) */
1944 if (drbd_get_max_capacity(nbc) <
1945 drbd_get_capacity(device->this_bdev)) {
1946 retcode = ERR_DISK_TOO_SMALL;
1950 nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1952 if (nbc->known_size > max_possible_sectors) {
1953 drbd_warn(device, "==> truncating very big lower level device "
1954 "to currently maximum possible %llu sectors <==\n",
1955 (unsigned long long) max_possible_sectors);
1956 if (new_disk_conf->meta_dev_idx >= 0)
1957 drbd_warn(device, "==>> using internal or flexible "
1958 "meta data may help <<==\n");
1961 drbd_suspend_io(device);
1962 /* also wait for the last barrier ack. */
1963 /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1964 * We need a way to either ignore barrier acks for barriers sent before a device
1965 * was attached, or a way to wait for all pending barrier acks to come in.
1966 * As barriers are counted per resource,
1967 * we'd need to suspend io on all devices of a resource.
1969 wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1970 /* and for any other previously queued work */
1971 drbd_flush_workqueue(&connection->sender_work);
1973 rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1974 retcode = rv; /* FIXME: Type mismatch. */
1975 drbd_resume_io(device);
1976 if (rv < SS_SUCCESS)
1979 if (!get_ldev_if_state(device, D_ATTACHING))
1980 goto force_diskless;
1982 if (!device->bitmap) {
1983 if (drbd_bm_init(device)) {
1984 retcode = ERR_NOMEM;
1985 goto force_diskless_dec;
1989 if (device->state.pdsk != D_UP_TO_DATE && device->ed_uuid &&
1990 (device->state.role == R_PRIMARY || device->state.peer == R_PRIMARY) &&
1991 (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1992 drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1993 (unsigned long long)device->ed_uuid);
1994 retcode = ERR_DATA_NOT_CURRENT;
1995 goto force_diskless_dec;
1998 /* Since we are diskless, fix the activity log first... */
1999 if (drbd_check_al_size(device, new_disk_conf)) {
2000 retcode = ERR_NOMEM;
2001 goto force_diskless_dec;
2004 /* Prevent shrinking of consistent devices ! */
2006 unsigned long long nsz = drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0);
2007 unsigned long long eff = nbc->md.la_size_sect;
2008 if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && nsz < eff) {
2009 if (nsz == nbc->disk_conf->disk_size) {
2010 drbd_warn(device, "truncating a consistent device during attach (%llu < %llu)\n", nsz, eff);
2012 drbd_warn(device, "refusing to truncate a consistent device (%llu < %llu)\n", nsz, eff);
2013 drbd_msg_sprintf_info(adm_ctx.reply_skb,
2014 "To-be-attached device has last effective > current size, and is consistent\n"
2015 "(%llu > %llu sectors). Refusing to attach.", eff, nsz);
2016 retcode = ERR_IMPLICIT_SHRINK;
2017 goto force_diskless_dec;
2022 lock_all_resources();
2023 retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
2024 if (retcode != NO_ERROR) {
2025 unlock_all_resources();
2026 goto force_diskless_dec;
2029 /* Reset the "barriers don't work" bits here, then force meta data to
2030 * be written, to ensure we determine if barriers are supported. */
2031 if (new_disk_conf->md_flushes)
2032 clear_bit(MD_NO_FUA, &device->flags);
2034 set_bit(MD_NO_FUA, &device->flags);
2036 /* Point of no return reached.
2037 * Devices and memory are no longer released by error cleanup below.
2038 * now device takes over responsibility, and the state engine should
2039 * clean it up somewhere. */
2040 D_ASSERT(device, device->ldev == NULL);
2042 device->resync = resync_lru;
2043 device->rs_plan_s = new_plan;
2046 new_disk_conf = NULL;
2049 drbd_resync_after_changed(device);
2050 drbd_bump_write_ordering(device->resource, device->ldev, WO_BDEV_FLUSH);
2051 unlock_all_resources();
2053 if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
2054 set_bit(CRASHED_PRIMARY, &device->flags);
2056 clear_bit(CRASHED_PRIMARY, &device->flags);
2058 if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
2059 !(device->state.role == R_PRIMARY && device->resource->susp_nod))
2060 set_bit(CRASHED_PRIMARY, &device->flags);
2062 device->send_cnt = 0;
2063 device->recv_cnt = 0;
2064 device->read_cnt = 0;
2065 device->writ_cnt = 0;
2067 drbd_reconsider_queue_parameters(device, device->ldev, NULL);
2069 /* If I am currently not R_PRIMARY,
2070 * but meta data primary indicator is set,
2071 * I just now recover from a hard crash,
2072 * and have been R_PRIMARY before that crash.
2074 * Now, if I had no connection before that crash
2075 * (have been degraded R_PRIMARY), chances are that
2076 * I won't find my peer now either.
2078 * In that case, and _only_ in that case,
2079 * we use the degr-wfc-timeout instead of the default,
2080 * so we can automatically recover from a crash of a
2081 * degraded but active "cluster" after a certain timeout.
2083 clear_bit(USE_DEGR_WFC_T, &device->flags);
2084 if (device->state.role != R_PRIMARY &&
2085 drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
2086 !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
2087 set_bit(USE_DEGR_WFC_T, &device->flags);
2089 dd = drbd_determine_dev_size(device, 0, NULL);
2090 if (dd <= DS_ERROR) {
2091 retcode = ERR_NOMEM_BITMAP;
2092 goto force_diskless_dec;
2093 } else if (dd == DS_GREW)
2094 set_bit(RESYNC_AFTER_NEG, &device->flags);
2096 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
2097 (test_bit(CRASHED_PRIMARY, &device->flags) &&
2098 drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
2099 drbd_info(device, "Assuming that all blocks are out of sync "
2100 "(aka FullSync)\n");
2101 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2102 "set_n_write from attaching", BM_LOCKED_MASK)) {
2103 retcode = ERR_IO_MD_DISK;
2104 goto force_diskless_dec;
2107 if (drbd_bitmap_io(device, &drbd_bm_read,
2108 "read from attaching", BM_LOCKED_MASK)) {
2109 retcode = ERR_IO_MD_DISK;
2110 goto force_diskless_dec;
2114 if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
2115 drbd_suspend_al(device); /* IO is still suspended here... */
2117 spin_lock_irq(&device->resource->req_lock);
2118 os = drbd_read_state(device);
2120 /* If MDF_CONSISTENT is not set go into inconsistent state,
2121 otherwise investigate MDF_WasUpToDate...
2122 If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
2123 otherwise into D_CONSISTENT state.
2125 if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
2126 if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
2127 ns.disk = D_CONSISTENT;
2129 ns.disk = D_OUTDATED;
2131 ns.disk = D_INCONSISTENT;
2134 if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
2135 ns.pdsk = D_OUTDATED;
2138 if (ns.disk == D_CONSISTENT &&
2139 (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
2140 ns.disk = D_UP_TO_DATE;
2142 /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
2143 MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
2144 this point, because drbd_request_state() modifies these
2147 if (rcu_dereference(device->ldev->disk_conf)->al_updates)
2148 device->ldev->md.flags &= ~MDF_AL_DISABLED;
2150 device->ldev->md.flags |= MDF_AL_DISABLED;
2154 /* In case we are C_CONNECTED postpone any decision on the new disk
2155 state after the negotiation phase. */
2156 if (device->state.conn == C_CONNECTED) {
2157 device->new_state_tmp.i = ns.i;
2159 ns.disk = D_NEGOTIATING;
2161 /* We expect to receive up-to-date UUIDs soon.
2162 To avoid a race in receive_state, free p_uuid while
2163 holding req_lock. I.e. atomic with the state change */
2164 kfree(device->p_uuid);
2165 device->p_uuid = NULL;
2168 rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
2169 spin_unlock_irq(&device->resource->req_lock);
2171 if (rv < SS_SUCCESS)
2172 goto force_diskless_dec;
2174 mod_timer(&device->request_timer, jiffies + HZ);
2176 if (device->state.role == R_PRIMARY)
2177 device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
2179 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
2181 drbd_md_mark_dirty(device);
2182 drbd_md_sync(device);
2184 kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
2186 conn_reconfig_done(connection);
2187 mutex_unlock(&adm_ctx.resource->adm_mutex);
2188 drbd_adm_finish(&adm_ctx, info, retcode);
2194 drbd_force_state(device, NS(disk, D_DISKLESS));
2195 drbd_md_sync(device);
2197 conn_reconfig_done(connection);
2199 close_backing_dev(device, nbc->md_bdev, nbc->md_bdev != nbc->backing_bdev);
2200 close_backing_dev(device, nbc->backing_bdev, true);
2203 kfree(new_disk_conf);
2204 lc_destroy(resync_lru);
2206 mutex_unlock(&adm_ctx.resource->adm_mutex);
2208 drbd_adm_finish(&adm_ctx, info, retcode);
2212 static int adm_detach(struct drbd_device *device, int force)
2215 set_bit(FORCE_DETACH, &device->flags);
2216 drbd_force_state(device, NS(disk, D_FAILED));
2220 return drbd_request_detach_interruptible(device);
2223 /* Detaching the disk is a process in multiple stages. First we need to lock
2224 * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
2225 * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
2226 * internal references as well.
2227 * Only then we have finally detached. */
2228 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
2230 struct drbd_config_context adm_ctx;
2231 enum drbd_ret_code retcode;
2232 struct detach_parms parms = { };
2235 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2236 if (!adm_ctx.reply_skb)
2238 if (retcode != NO_ERROR)
2241 if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
2242 err = detach_parms_from_attrs(&parms, info);
2244 retcode = ERR_MANDATORY_TAG;
2245 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2250 mutex_lock(&adm_ctx.resource->adm_mutex);
2251 retcode = adm_detach(adm_ctx.device, parms.force_detach);
2252 mutex_unlock(&adm_ctx.resource->adm_mutex);
2254 drbd_adm_finish(&adm_ctx, info, retcode);
2258 static bool conn_resync_running(struct drbd_connection *connection)
2260 struct drbd_peer_device *peer_device;
2265 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2266 struct drbd_device *device = peer_device->device;
2267 if (device->state.conn == C_SYNC_SOURCE ||
2268 device->state.conn == C_SYNC_TARGET ||
2269 device->state.conn == C_PAUSED_SYNC_S ||
2270 device->state.conn == C_PAUSED_SYNC_T) {
2280 static bool conn_ov_running(struct drbd_connection *connection)
2282 struct drbd_peer_device *peer_device;
2287 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2288 struct drbd_device *device = peer_device->device;
2289 if (device->state.conn == C_VERIFY_S ||
2290 device->state.conn == C_VERIFY_T) {
2300 static enum drbd_ret_code
2301 _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2303 struct drbd_peer_device *peer_device;
2306 if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2307 if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2308 return ERR_NEED_APV_100;
2310 if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2311 return ERR_NEED_APV_100;
2313 if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2314 return ERR_NEED_APV_100;
2317 if (!new_net_conf->two_primaries &&
2318 conn_highest_role(connection) == R_PRIMARY &&
2319 conn_highest_peer(connection) == R_PRIMARY)
2320 return ERR_NEED_ALLOW_TWO_PRI;
2322 if (new_net_conf->two_primaries &&
2323 (new_net_conf->wire_protocol != DRBD_PROT_C))
2324 return ERR_NOT_PROTO_C;
2326 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2327 struct drbd_device *device = peer_device->device;
2328 if (get_ldev(device)) {
2329 enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2331 if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2332 return ERR_STONITH_AND_PROT_A;
2334 if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2335 return ERR_DISCARD_IMPOSSIBLE;
2338 if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2339 return ERR_CONG_NOT_PROTO_A;
2344 static enum drbd_ret_code
2345 check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2347 enum drbd_ret_code rv;
2348 struct drbd_peer_device *peer_device;
2352 rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2355 /* connection->peer_devices protected by genl_lock() here */
2356 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2357 struct drbd_device *device = peer_device->device;
2358 if (!device->bitmap) {
2359 if (drbd_bm_init(device))
2368 struct crypto_shash *verify_tfm;
2369 struct crypto_shash *csums_tfm;
2370 struct crypto_shash *cram_hmac_tfm;
2371 struct crypto_shash *integrity_tfm;
2375 alloc_shash(struct crypto_shash **tfm, char *tfm_name, int err_alg)
2380 *tfm = crypto_alloc_shash(tfm_name, 0, 0);
2389 static enum drbd_ret_code
2390 alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2392 char hmac_name[CRYPTO_MAX_ALG_NAME];
2393 enum drbd_ret_code rv;
2395 rv = alloc_shash(&crypto->csums_tfm, new_net_conf->csums_alg,
2399 rv = alloc_shash(&crypto->verify_tfm, new_net_conf->verify_alg,
2403 rv = alloc_shash(&crypto->integrity_tfm, new_net_conf->integrity_alg,
2407 if (new_net_conf->cram_hmac_alg[0] != 0) {
2408 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
2409 new_net_conf->cram_hmac_alg);
2411 rv = alloc_shash(&crypto->cram_hmac_tfm, hmac_name,
2418 static void free_crypto(struct crypto *crypto)
2420 crypto_free_shash(crypto->cram_hmac_tfm);
2421 crypto_free_shash(crypto->integrity_tfm);
2422 crypto_free_shash(crypto->csums_tfm);
2423 crypto_free_shash(crypto->verify_tfm);
2426 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2428 struct drbd_config_context adm_ctx;
2429 enum drbd_ret_code retcode;
2430 struct drbd_connection *connection;
2431 struct net_conf *old_net_conf, *new_net_conf = NULL;
2433 int ovr; /* online verify running */
2434 int rsr; /* re-sync running */
2435 struct crypto crypto = { };
2437 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2438 if (!adm_ctx.reply_skb)
2440 if (retcode != NO_ERROR)
2443 connection = adm_ctx.connection;
2444 mutex_lock(&adm_ctx.resource->adm_mutex);
2446 new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
2447 if (!new_net_conf) {
2448 retcode = ERR_NOMEM;
2452 conn_reconfig_start(connection);
2454 mutex_lock(&connection->data.mutex);
2455 mutex_lock(&connection->resource->conf_update);
2456 old_net_conf = connection->net_conf;
2458 if (!old_net_conf) {
2459 drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect");
2460 retcode = ERR_INVALID_REQUEST;
2464 *new_net_conf = *old_net_conf;
2465 if (should_set_defaults(info))
2466 set_net_conf_defaults(new_net_conf);
2468 err = net_conf_from_attrs_for_change(new_net_conf, info);
2469 if (err && err != -ENOMSG) {
2470 retcode = ERR_MANDATORY_TAG;
2471 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2475 retcode = check_net_options(connection, new_net_conf);
2476 if (retcode != NO_ERROR)
2479 /* re-sync running */
2480 rsr = conn_resync_running(connection);
2481 if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2482 retcode = ERR_CSUMS_RESYNC_RUNNING;
2486 /* online verify running */
2487 ovr = conn_ov_running(connection);
2488 if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2489 retcode = ERR_VERIFY_RUNNING;
2493 retcode = alloc_crypto(&crypto, new_net_conf);
2494 if (retcode != NO_ERROR)
2497 rcu_assign_pointer(connection->net_conf, new_net_conf);
2500 crypto_free_shash(connection->csums_tfm);
2501 connection->csums_tfm = crypto.csums_tfm;
2502 crypto.csums_tfm = NULL;
2505 crypto_free_shash(connection->verify_tfm);
2506 connection->verify_tfm = crypto.verify_tfm;
2507 crypto.verify_tfm = NULL;
2510 crypto_free_shash(connection->integrity_tfm);
2511 connection->integrity_tfm = crypto.integrity_tfm;
2512 if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2513 /* Do this without trying to take connection->data.mutex again. */
2514 __drbd_send_protocol(connection, P_PROTOCOL_UPDATE);
2516 crypto_free_shash(connection->cram_hmac_tfm);
2517 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2519 mutex_unlock(&connection->resource->conf_update);
2520 mutex_unlock(&connection->data.mutex);
2522 kfree(old_net_conf);
2524 if (connection->cstate >= C_WF_REPORT_PARAMS) {
2525 struct drbd_peer_device *peer_device;
2528 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2529 drbd_send_sync_param(peer_device);
2535 mutex_unlock(&connection->resource->conf_update);
2536 mutex_unlock(&connection->data.mutex);
2537 free_crypto(&crypto);
2538 kfree(new_net_conf);
2540 conn_reconfig_done(connection);
2542 mutex_unlock(&adm_ctx.resource->adm_mutex);
2544 drbd_adm_finish(&adm_ctx, info, retcode);
2548 static void connection_to_info(struct connection_info *info,
2549 struct drbd_connection *connection)
2551 info->conn_connection_state = connection->cstate;
2552 info->conn_role = conn_highest_peer(connection);
2555 static void peer_device_to_info(struct peer_device_info *info,
2556 struct drbd_peer_device *peer_device)
2558 struct drbd_device *device = peer_device->device;
2560 info->peer_repl_state =
2561 max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn);
2562 info->peer_disk_state = device->state.pdsk;
2563 info->peer_resync_susp_user = device->state.user_isp;
2564 info->peer_resync_susp_peer = device->state.peer_isp;
2565 info->peer_resync_susp_dependency = device->state.aftr_isp;
2568 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2570 struct connection_info connection_info;
2571 enum drbd_notification_type flags;
2572 unsigned int peer_devices = 0;
2573 struct drbd_config_context adm_ctx;
2574 struct drbd_peer_device *peer_device;
2575 struct net_conf *old_net_conf, *new_net_conf = NULL;
2576 struct crypto crypto = { };
2577 struct drbd_resource *resource;
2578 struct drbd_connection *connection;
2579 enum drbd_ret_code retcode;
2583 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2585 if (!adm_ctx.reply_skb)
2587 if (retcode != NO_ERROR)
2589 if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2590 drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing");
2591 retcode = ERR_INVALID_REQUEST;
2595 /* No need for _rcu here. All reconfiguration is
2596 * strictly serialized on genl_lock(). We are protected against
2597 * concurrent reconfiguration/addition/deletion */
2598 for_each_resource(resource, &drbd_resources) {
2599 for_each_connection(connection, resource) {
2600 if (nla_len(adm_ctx.my_addr) == connection->my_addr_len &&
2601 !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr,
2602 connection->my_addr_len)) {
2603 retcode = ERR_LOCAL_ADDR;
2607 if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len &&
2608 !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr,
2609 connection->peer_addr_len)) {
2610 retcode = ERR_PEER_ADDR;
2616 mutex_lock(&adm_ctx.resource->adm_mutex);
2617 connection = first_connection(adm_ctx.resource);
2618 conn_reconfig_start(connection);
2620 if (connection->cstate > C_STANDALONE) {
2621 retcode = ERR_NET_CONFIGURED;
2625 /* allocation not in the IO path, drbdsetup / netlink process context */
2626 new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL);
2627 if (!new_net_conf) {
2628 retcode = ERR_NOMEM;
2632 set_net_conf_defaults(new_net_conf);
2634 err = net_conf_from_attrs(new_net_conf, info);
2635 if (err && err != -ENOMSG) {
2636 retcode = ERR_MANDATORY_TAG;
2637 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2641 retcode = check_net_options(connection, new_net_conf);
2642 if (retcode != NO_ERROR)
2645 retcode = alloc_crypto(&crypto, new_net_conf);
2646 if (retcode != NO_ERROR)
2649 ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2651 drbd_flush_workqueue(&connection->sender_work);
2653 mutex_lock(&adm_ctx.resource->conf_update);
2654 old_net_conf = connection->net_conf;
2656 retcode = ERR_NET_CONFIGURED;
2657 mutex_unlock(&adm_ctx.resource->conf_update);
2660 rcu_assign_pointer(connection->net_conf, new_net_conf);
2662 conn_free_crypto(connection);
2663 connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2664 connection->integrity_tfm = crypto.integrity_tfm;
2665 connection->csums_tfm = crypto.csums_tfm;
2666 connection->verify_tfm = crypto.verify_tfm;
2668 connection->my_addr_len = nla_len(adm_ctx.my_addr);
2669 memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2670 connection->peer_addr_len = nla_len(adm_ctx.peer_addr);
2671 memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2673 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2677 connection_to_info(&connection_info, connection);
2678 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2679 mutex_lock(¬ification_mutex);
2680 notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags);
2681 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2682 struct peer_device_info peer_device_info;
2684 peer_device_to_info(&peer_device_info, peer_device);
2685 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2686 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags);
2688 mutex_unlock(¬ification_mutex);
2689 mutex_unlock(&adm_ctx.resource->conf_update);
2692 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2693 struct drbd_device *device = peer_device->device;
2694 device->send_cnt = 0;
2695 device->recv_cnt = 0;
2699 retcode = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2701 conn_reconfig_done(connection);
2702 mutex_unlock(&adm_ctx.resource->adm_mutex);
2703 drbd_adm_finish(&adm_ctx, info, retcode);
2707 free_crypto(&crypto);
2708 kfree(new_net_conf);
2710 conn_reconfig_done(connection);
2711 mutex_unlock(&adm_ctx.resource->adm_mutex);
2713 drbd_adm_finish(&adm_ctx, info, retcode);
2717 static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2719 enum drbd_conns cstate;
2720 enum drbd_state_rv rv;
2723 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2724 force ? CS_HARD : 0);
2727 case SS_NOTHING_TO_DO:
2729 case SS_ALREADY_STANDALONE:
2731 case SS_PRIMARY_NOP:
2732 /* Our state checking code wants to see the peer outdated. */
2733 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2735 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2736 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2739 case SS_CW_FAILED_BY_PEER:
2740 spin_lock_irq(&connection->resource->req_lock);
2741 cstate = connection->cstate;
2742 spin_unlock_irq(&connection->resource->req_lock);
2743 if (cstate <= C_WF_CONNECTION)
2745 /* The peer probably wants to see us outdated. */
2746 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2747 disk, D_OUTDATED), 0);
2748 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2749 rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2754 /* no special handling necessary */
2757 if (rv >= SS_SUCCESS) {
2758 enum drbd_state_rv rv2;
2759 /* No one else can reconfigure the network while I am here.
2760 * The state handling only uses drbd_thread_stop_nowait(),
2761 * we want to really wait here until the receiver is no more.
2763 drbd_thread_stop(&connection->receiver);
2765 /* Race breaker. This additional state change request may be
2766 * necessary, if this was a forced disconnect during a receiver
2767 * restart. We may have "killed" the receiver thread just
2768 * after drbd_receiver() returned. Typically, we should be
2769 * C_STANDALONE already, now, and this becomes a no-op.
2771 rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2772 CS_VERBOSE | CS_HARD);
2773 if (rv2 < SS_SUCCESS)
2774 drbd_err(connection,
2775 "unexpected rv2=%d in conn_try_disconnect()\n",
2777 /* Unlike in DRBD 9, the state engine has generated
2778 * NOTIFY_DESTROY events before clearing connection->net_conf. */
2783 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2785 struct drbd_config_context adm_ctx;
2786 struct disconnect_parms parms;
2787 struct drbd_connection *connection;
2788 enum drbd_state_rv rv;
2789 enum drbd_ret_code retcode;
2792 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2793 if (!adm_ctx.reply_skb)
2795 if (retcode != NO_ERROR)
2798 connection = adm_ctx.connection;
2799 memset(&parms, 0, sizeof(parms));
2800 if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2801 err = disconnect_parms_from_attrs(&parms, info);
2803 retcode = ERR_MANDATORY_TAG;
2804 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2809 mutex_lock(&adm_ctx.resource->adm_mutex);
2810 rv = conn_try_disconnect(connection, parms.force_disconnect);
2811 if (rv < SS_SUCCESS)
2812 retcode = rv; /* FIXME: Type mismatch. */
2815 mutex_unlock(&adm_ctx.resource->adm_mutex);
2817 drbd_adm_finish(&adm_ctx, info, retcode);
2821 void resync_after_online_grow(struct drbd_device *device)
2823 int iass; /* I am sync source */
2825 drbd_info(device, "Resync of new storage after online grow\n");
2826 if (device->state.role != device->state.peer)
2827 iass = (device->state.role == R_PRIMARY);
2829 iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2832 drbd_start_resync(device, C_SYNC_SOURCE);
2834 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2837 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2839 struct drbd_config_context adm_ctx;
2840 struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2841 struct resize_parms rs;
2842 struct drbd_device *device;
2843 enum drbd_ret_code retcode;
2844 enum determine_dev_size dd;
2845 bool change_al_layout = false;
2846 enum dds_flags ddsf;
2850 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2851 if (!adm_ctx.reply_skb)
2853 if (retcode != NO_ERROR)
2856 mutex_lock(&adm_ctx.resource->adm_mutex);
2857 device = adm_ctx.device;
2858 if (!get_ldev(device)) {
2859 retcode = ERR_NO_DISK;
2863 memset(&rs, 0, sizeof(struct resize_parms));
2864 rs.al_stripes = device->ldev->md.al_stripes;
2865 rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2866 if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2867 err = resize_parms_from_attrs(&rs, info);
2869 retcode = ERR_MANDATORY_TAG;
2870 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2875 if (device->state.conn > C_CONNECTED) {
2876 retcode = ERR_RESIZE_RESYNC;
2880 if (device->state.role == R_SECONDARY &&
2881 device->state.peer == R_SECONDARY) {
2882 retcode = ERR_NO_PRIMARY;
2886 if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2887 retcode = ERR_NEED_APV_93;
2892 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2894 if (u_size != (sector_t)rs.resize_size) {
2895 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2896 if (!new_disk_conf) {
2897 retcode = ERR_NOMEM;
2902 if (device->ldev->md.al_stripes != rs.al_stripes ||
2903 device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2904 u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2906 if (al_size_k > (16 * 1024 * 1024)) {
2907 retcode = ERR_MD_LAYOUT_TOO_BIG;
2911 if (al_size_k < MD_32kB_SECT/2) {
2912 retcode = ERR_MD_LAYOUT_TOO_SMALL;
2916 if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2917 retcode = ERR_MD_LAYOUT_CONNECTED;
2921 change_al_layout = true;
2924 if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev))
2925 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
2927 if (new_disk_conf) {
2928 mutex_lock(&device->resource->conf_update);
2929 old_disk_conf = device->ldev->disk_conf;
2930 *new_disk_conf = *old_disk_conf;
2931 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2932 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2933 mutex_unlock(&device->resource->conf_update);
2935 kfree(old_disk_conf);
2936 new_disk_conf = NULL;
2939 ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2940 dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL);
2941 drbd_md_sync(device);
2943 if (dd == DS_ERROR) {
2944 retcode = ERR_NOMEM_BITMAP;
2946 } else if (dd == DS_ERROR_SPACE_MD) {
2947 retcode = ERR_MD_LAYOUT_NO_FIT;
2949 } else if (dd == DS_ERROR_SHRINK) {
2950 retcode = ERR_IMPLICIT_SHRINK;
2954 if (device->state.conn == C_CONNECTED) {
2956 set_bit(RESIZE_PENDING, &device->flags);
2958 drbd_send_uuids(first_peer_device(device));
2959 drbd_send_sizes(first_peer_device(device), 1, ddsf);
2963 mutex_unlock(&adm_ctx.resource->adm_mutex);
2965 drbd_adm_finish(&adm_ctx, info, retcode);
2970 kfree(new_disk_conf);
2974 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2976 struct drbd_config_context adm_ctx;
2977 enum drbd_ret_code retcode;
2978 struct res_opts res_opts;
2981 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2982 if (!adm_ctx.reply_skb)
2984 if (retcode != NO_ERROR)
2987 res_opts = adm_ctx.resource->res_opts;
2988 if (should_set_defaults(info))
2989 set_res_opts_defaults(&res_opts);
2991 err = res_opts_from_attrs(&res_opts, info);
2992 if (err && err != -ENOMSG) {
2993 retcode = ERR_MANDATORY_TAG;
2994 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2998 mutex_lock(&adm_ctx.resource->adm_mutex);
2999 err = set_resource_options(adm_ctx.resource, &res_opts);
3001 retcode = ERR_INVALID_REQUEST;
3003 retcode = ERR_NOMEM;
3005 mutex_unlock(&adm_ctx.resource->adm_mutex);
3008 drbd_adm_finish(&adm_ctx, info, retcode);
3012 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
3014 struct drbd_config_context adm_ctx;
3015 struct drbd_device *device;
3016 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3018 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3019 if (!adm_ctx.reply_skb)
3021 if (retcode != NO_ERROR)
3024 device = adm_ctx.device;
3025 if (!get_ldev(device)) {
3026 retcode = ERR_NO_DISK;
3030 mutex_lock(&adm_ctx.resource->adm_mutex);
3032 /* If there is still bitmap IO pending, probably because of a previous
3033 * resync just being finished, wait for it before requesting a new resync.
3034 * Also wait for it's after_state_ch(). */
3035 drbd_suspend_io(device);
3036 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
3037 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
3039 /* If we happen to be C_STANDALONE R_SECONDARY, just change to
3040 * D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
3041 * try to start a resync handshake as sync target for full sync.
3043 if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
3044 retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
3045 if (retcode >= SS_SUCCESS) {
3046 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
3047 "set_n_write from invalidate", BM_LOCKED_MASK))
3048 retcode = ERR_IO_MD_DISK;
3051 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
3052 drbd_resume_io(device);
3053 mutex_unlock(&adm_ctx.resource->adm_mutex);
3056 drbd_adm_finish(&adm_ctx, info, retcode);
3060 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
3061 union drbd_state mask, union drbd_state val)
3063 struct drbd_config_context adm_ctx;
3064 enum drbd_ret_code retcode;
3066 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3067 if (!adm_ctx.reply_skb)
3069 if (retcode != NO_ERROR)
3072 mutex_lock(&adm_ctx.resource->adm_mutex);
3073 retcode = drbd_request_state(adm_ctx.device, mask, val);
3074 mutex_unlock(&adm_ctx.resource->adm_mutex);
3076 drbd_adm_finish(&adm_ctx, info, retcode);
3080 static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local)
3084 rv = drbd_bmio_set_n_write(device);
3085 drbd_suspend_al(device);
3089 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
3091 struct drbd_config_context adm_ctx;
3092 int retcode; /* drbd_ret_code, drbd_state_rv */
3093 struct drbd_device *device;
3095 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3096 if (!adm_ctx.reply_skb)
3098 if (retcode != NO_ERROR)
3101 device = adm_ctx.device;
3102 if (!get_ldev(device)) {
3103 retcode = ERR_NO_DISK;
3107 mutex_lock(&adm_ctx.resource->adm_mutex);
3109 /* If there is still bitmap IO pending, probably because of a previous
3110 * resync just being finished, wait for it before requesting a new resync.
3111 * Also wait for it's after_state_ch(). */
3112 drbd_suspend_io(device);
3113 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
3114 drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
3116 /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
3117 * in the bitmap. Otherwise, try to start a resync handshake
3118 * as sync source for full sync.
3120 if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
3121 /* The peer will get a resync upon connect anyways. Just make that
3122 into a full resync. */
3123 retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
3124 if (retcode >= SS_SUCCESS) {
3125 if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al,
3126 "set_n_write from invalidate_peer",
3127 BM_LOCKED_SET_ALLOWED))
3128 retcode = ERR_IO_MD_DISK;
3131 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
3132 drbd_resume_io(device);
3133 mutex_unlock(&adm_ctx.resource->adm_mutex);
3136 drbd_adm_finish(&adm_ctx, info, retcode);
3140 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
3142 struct drbd_config_context adm_ctx;
3143 enum drbd_ret_code retcode;
3145 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3146 if (!adm_ctx.reply_skb)
3148 if (retcode != NO_ERROR)
3151 mutex_lock(&adm_ctx.resource->adm_mutex);
3152 if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
3153 retcode = ERR_PAUSE_IS_SET;
3154 mutex_unlock(&adm_ctx.resource->adm_mutex);
3156 drbd_adm_finish(&adm_ctx, info, retcode);
3160 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
3162 struct drbd_config_context adm_ctx;
3163 union drbd_dev_state s;
3164 enum drbd_ret_code retcode;
3166 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3167 if (!adm_ctx.reply_skb)
3169 if (retcode != NO_ERROR)
3172 mutex_lock(&adm_ctx.resource->adm_mutex);
3173 if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
3174 s = adm_ctx.device->state;
3175 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
3176 retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
3177 s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
3179 retcode = ERR_PAUSE_IS_CLEAR;
3182 mutex_unlock(&adm_ctx.resource->adm_mutex);
3184 drbd_adm_finish(&adm_ctx, info, retcode);
3188 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
3190 return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
3193 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
3195 struct drbd_config_context adm_ctx;
3196 struct drbd_device *device;
3197 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3199 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3200 if (!adm_ctx.reply_skb)
3202 if (retcode != NO_ERROR)
3205 mutex_lock(&adm_ctx.resource->adm_mutex);
3206 device = adm_ctx.device;
3207 if (test_bit(NEW_CUR_UUID, &device->flags)) {
3208 if (get_ldev_if_state(device, D_ATTACHING)) {
3209 drbd_uuid_new_current(device);
3212 /* This is effectively a multi-stage "forced down".
3213 * The NEW_CUR_UUID bit is supposedly only set, if we
3214 * lost the replication connection, and are configured
3215 * to freeze IO and wait for some fence-peer handler.
3216 * So we still don't have a replication connection.
3217 * And now we don't have a local disk either. After
3218 * resume, we will fail all pending and new IO, because
3219 * we don't have any data anymore. Which means we will
3220 * eventually be able to terminate all users of this
3221 * device, and then take it down. By bumping the
3222 * "effective" data uuid, we make sure that you really
3223 * need to tear down before you reconfigure, we will
3224 * the refuse to re-connect or re-attach (because no
3225 * matching real data uuid exists).
3228 get_random_bytes(&val, sizeof(u64));
3229 drbd_set_ed_uuid(device, val);
3230 drbd_warn(device, "Resumed without access to data; please tear down before attempting to re-configure.\n");
3232 clear_bit(NEW_CUR_UUID, &device->flags);
3234 drbd_suspend_io(device);
3235 retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
3236 if (retcode == SS_SUCCESS) {
3237 if (device->state.conn < C_CONNECTED)
3238 tl_clear(first_peer_device(device)->connection);
3239 if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
3240 tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO);
3242 drbd_resume_io(device);
3243 mutex_unlock(&adm_ctx.resource->adm_mutex);
3245 drbd_adm_finish(&adm_ctx, info, retcode);
3249 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
3251 return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
3254 static int nla_put_drbd_cfg_context(struct sk_buff *skb,
3255 struct drbd_resource *resource,
3256 struct drbd_connection *connection,
3257 struct drbd_device *device)
3260 nla = nla_nest_start_noflag(skb, DRBD_NLA_CFG_CONTEXT);
3262 goto nla_put_failure;
3264 nla_put_u32(skb, T_ctx_volume, device->vnr))
3265 goto nla_put_failure;
3266 if (nla_put_string(skb, T_ctx_resource_name, resource->name))
3267 goto nla_put_failure;
3269 if (connection->my_addr_len &&
3270 nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr))
3271 goto nla_put_failure;
3272 if (connection->peer_addr_len &&
3273 nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr))
3274 goto nla_put_failure;
3276 nla_nest_end(skb, nla);
3281 nla_nest_cancel(skb, nla);
3286 * The generic netlink dump callbacks are called outside the genl_lock(), so
3287 * they cannot use the simple attribute parsing code which uses global
3290 static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr)
3292 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3293 const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3296 nla = nla_find(nlmsg_attrdata(nlh, hdrlen), nlmsg_attrlen(nlh, hdrlen),
3297 DRBD_NLA_CFG_CONTEXT);
3300 return drbd_nla_find_nested(maxtype, nla, __nla_type(attr));
3303 static void resource_to_info(struct resource_info *, struct drbd_resource *);
3305 int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb)
3307 struct drbd_genlmsghdr *dh;
3308 struct drbd_resource *resource;
3309 struct resource_info resource_info;
3310 struct resource_statistics resource_statistics;
3315 for_each_resource_rcu(resource, &drbd_resources)
3316 if (resource == (struct drbd_resource *)cb->args[0])
3317 goto found_resource;
3318 err = 0; /* resource was probably deleted */
3321 resource = list_entry(&drbd_resources,
3322 struct drbd_resource, resources);
3325 list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) {
3332 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3333 cb->nlh->nlmsg_seq, &drbd_genl_family,
3334 NLM_F_MULTI, DRBD_ADM_GET_RESOURCES);
3339 dh->ret_code = NO_ERROR;
3340 err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL);
3343 err = res_opts_to_skb(skb, &resource->res_opts, !capable(CAP_SYS_ADMIN));
3346 resource_to_info(&resource_info, resource);
3347 err = resource_info_to_skb(skb, &resource_info, !capable(CAP_SYS_ADMIN));
3350 resource_statistics.res_stat_write_ordering = resource->write_ordering;
3351 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
3354 cb->args[0] = (long)resource;
3355 genlmsg_end(skb, dh);
3365 static void device_to_statistics(struct device_statistics *s,
3366 struct drbd_device *device)
3368 memset(s, 0, sizeof(*s));
3369 s->dev_upper_blocked = !may_inc_ap_bio(device);
3370 if (get_ldev(device)) {
3371 struct drbd_md *md = &device->ldev->md;
3372 u64 *history_uuids = (u64 *)s->history_uuids;
3373 struct request_queue *q;
3376 spin_lock_irq(&md->uuid_lock);
3377 s->dev_current_uuid = md->uuid[UI_CURRENT];
3378 BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1);
3379 for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++)
3380 history_uuids[n] = md->uuid[UI_HISTORY_START + n];
3381 for (; n < HISTORY_UUIDS; n++)
3382 history_uuids[n] = 0;
3383 s->history_uuids_len = HISTORY_UUIDS;
3384 spin_unlock_irq(&md->uuid_lock);
3386 s->dev_disk_flags = md->flags;
3387 q = bdev_get_queue(device->ldev->backing_bdev);
3388 s->dev_lower_blocked =
3389 bdi_congested(q->backing_dev_info,
3390 (1 << WB_async_congested) |
3391 (1 << WB_sync_congested));
3394 s->dev_size = drbd_get_capacity(device->this_bdev);
3395 s->dev_read = device->read_cnt;
3396 s->dev_write = device->writ_cnt;
3397 s->dev_al_writes = device->al_writ_cnt;
3398 s->dev_bm_writes = device->bm_writ_cnt;
3399 s->dev_upper_pending = atomic_read(&device->ap_bio_cnt);
3400 s->dev_lower_pending = atomic_read(&device->local_cnt);
3401 s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags);
3402 s->dev_exposed_data_uuid = device->ed_uuid;
3405 static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr)
3408 struct drbd_resource *resource =
3409 (struct drbd_resource *)cb->args[0];
3410 kref_put(&resource->kref, drbd_destroy_resource);
3416 int drbd_adm_dump_devices_done(struct netlink_callback *cb) {
3417 return put_resource_in_arg0(cb, 7);
3420 static void device_to_info(struct device_info *, struct drbd_device *);
3422 int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb)
3424 struct nlattr *resource_filter;
3425 struct drbd_resource *resource;
3426 struct drbd_device *uninitialized_var(device);
3427 int minor, err, retcode;
3428 struct drbd_genlmsghdr *dh;
3429 struct device_info device_info;
3430 struct device_statistics device_statistics;
3431 struct idr *idr_to_search;
3433 resource = (struct drbd_resource *)cb->args[0];
3434 if (!cb->args[0] && !cb->args[1]) {
3435 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3436 if (resource_filter) {
3437 retcode = ERR_RES_NOT_KNOWN;
3438 resource = drbd_find_resource(nla_data(resource_filter));
3441 cb->args[0] = (long)resource;
3446 minor = cb->args[1];
3447 idr_to_search = resource ? &resource->devices : &drbd_devices;
3448 device = idr_get_next(idr_to_search, &minor);
3453 idr_for_each_entry_continue(idr_to_search, device, minor) {
3455 goto put_result; /* only one iteration */
3458 goto out; /* no more devices */
3461 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3462 cb->nlh->nlmsg_seq, &drbd_genl_family,
3463 NLM_F_MULTI, DRBD_ADM_GET_DEVICES);
3467 dh->ret_code = retcode;
3469 if (retcode == NO_ERROR) {
3470 dh->minor = device->minor;
3471 err = nla_put_drbd_cfg_context(skb, device->resource, NULL, device);
3474 if (get_ldev(device)) {
3475 struct disk_conf *disk_conf =
3476 rcu_dereference(device->ldev->disk_conf);
3478 err = disk_conf_to_skb(skb, disk_conf, !capable(CAP_SYS_ADMIN));
3483 device_to_info(&device_info, device);
3484 err = device_info_to_skb(skb, &device_info, !capable(CAP_SYS_ADMIN));
3488 device_to_statistics(&device_statistics, device);
3489 err = device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
3492 cb->args[1] = minor + 1;
3494 genlmsg_end(skb, dh);
3504 int drbd_adm_dump_connections_done(struct netlink_callback *cb)
3506 return put_resource_in_arg0(cb, 6);
3509 enum { SINGLE_RESOURCE, ITERATE_RESOURCES };
3511 int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb)
3513 struct nlattr *resource_filter;
3514 struct drbd_resource *resource = NULL, *next_resource;
3515 struct drbd_connection *uninitialized_var(connection);
3516 int err = 0, retcode;
3517 struct drbd_genlmsghdr *dh;
3518 struct connection_info connection_info;
3519 struct connection_statistics connection_statistics;
3522 resource = (struct drbd_resource *)cb->args[0];
3524 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3525 if (resource_filter) {
3526 retcode = ERR_RES_NOT_KNOWN;
3527 resource = drbd_find_resource(nla_data(resource_filter));
3530 cb->args[0] = (long)resource;
3531 cb->args[1] = SINGLE_RESOURCE;
3535 if (list_empty(&drbd_resources))
3537 resource = list_first_entry(&drbd_resources, struct drbd_resource, resources);
3538 kref_get(&resource->kref);
3539 cb->args[0] = (long)resource;
3540 cb->args[1] = ITERATE_RESOURCES;
3545 mutex_lock(&resource->conf_update);
3548 for_each_connection_rcu(connection, resource)
3549 if (connection == (struct drbd_connection *)cb->args[2])
3550 goto found_connection;
3551 /* connection was probably deleted */
3552 goto no_more_connections;
3554 connection = list_entry(&resource->connections, struct drbd_connection, connections);
3557 list_for_each_entry_continue_rcu(connection, &resource->connections, connections) {
3558 if (!has_net_conf(connection))
3561 goto put_result; /* only one iteration */
3564 no_more_connections:
3565 if (cb->args[1] == ITERATE_RESOURCES) {
3566 for_each_resource_rcu(next_resource, &drbd_resources) {
3567 if (next_resource == resource)
3568 goto found_resource;
3570 /* resource was probably deleted */
3575 list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) {
3576 mutex_unlock(&resource->conf_update);
3577 kref_put(&resource->kref, drbd_destroy_resource);
3578 resource = next_resource;
3579 kref_get(&resource->kref);
3580 cb->args[0] = (long)resource;
3584 goto out; /* no more resources */
3587 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3588 cb->nlh->nlmsg_seq, &drbd_genl_family,
3589 NLM_F_MULTI, DRBD_ADM_GET_CONNECTIONS);
3593 dh->ret_code = retcode;
3595 if (retcode == NO_ERROR) {
3596 struct net_conf *net_conf;
3598 err = nla_put_drbd_cfg_context(skb, resource, connection, NULL);
3601 net_conf = rcu_dereference(connection->net_conf);
3603 err = net_conf_to_skb(skb, net_conf, !capable(CAP_SYS_ADMIN));
3607 connection_to_info(&connection_info, connection);
3608 err = connection_info_to_skb(skb, &connection_info, !capable(CAP_SYS_ADMIN));
3611 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
3612 err = connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
3615 cb->args[2] = (long)connection;
3617 genlmsg_end(skb, dh);
3623 mutex_unlock(&resource->conf_update);
3629 enum mdf_peer_flag {
3630 MDF_PEER_CONNECTED = 1 << 0,
3631 MDF_PEER_OUTDATED = 1 << 1,
3632 MDF_PEER_FENCING = 1 << 2,
3633 MDF_PEER_FULL_SYNC = 1 << 3,
3636 static void peer_device_to_statistics(struct peer_device_statistics *s,
3637 struct drbd_peer_device *peer_device)
3639 struct drbd_device *device = peer_device->device;
3641 memset(s, 0, sizeof(*s));
3642 s->peer_dev_received = device->recv_cnt;
3643 s->peer_dev_sent = device->send_cnt;
3644 s->peer_dev_pending = atomic_read(&device->ap_pending_cnt) +
3645 atomic_read(&device->rs_pending_cnt);
3646 s->peer_dev_unacked = atomic_read(&device->unacked_cnt);
3647 s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9);
3648 s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9);
3649 if (get_ldev(device)) {
3650 struct drbd_md *md = &device->ldev->md;
3652 spin_lock_irq(&md->uuid_lock);
3653 s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP];
3654 spin_unlock_irq(&md->uuid_lock);
3656 (drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ?
3657 MDF_PEER_CONNECTED : 0) +
3658 (drbd_md_test_flag(device->ldev, MDF_CONSISTENT) &&
3659 !drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ?
3660 MDF_PEER_OUTDATED : 0) +
3661 /* FIXME: MDF_PEER_FENCING? */
3662 (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ?
3663 MDF_PEER_FULL_SYNC : 0);
3668 int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb)
3670 return put_resource_in_arg0(cb, 9);
3673 int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb)
3675 struct nlattr *resource_filter;
3676 struct drbd_resource *resource;
3677 struct drbd_device *uninitialized_var(device);
3678 struct drbd_peer_device *peer_device = NULL;
3679 int minor, err, retcode;
3680 struct drbd_genlmsghdr *dh;
3681 struct idr *idr_to_search;
3683 resource = (struct drbd_resource *)cb->args[0];
3684 if (!cb->args[0] && !cb->args[1]) {
3685 resource_filter = find_cfg_context_attr(cb->nlh, T_ctx_resource_name);
3686 if (resource_filter) {
3687 retcode = ERR_RES_NOT_KNOWN;
3688 resource = drbd_find_resource(nla_data(resource_filter));
3692 cb->args[0] = (long)resource;
3696 minor = cb->args[1];
3697 idr_to_search = resource ? &resource->devices : &drbd_devices;
3698 device = idr_find(idr_to_search, minor);
3703 device = idr_get_next(idr_to_search, &minor);
3710 for_each_peer_device(peer_device, device)
3711 if (peer_device == (struct drbd_peer_device *)cb->args[2])
3712 goto found_peer_device;
3713 /* peer device was probably deleted */
3716 /* Make peer_device point to the list head (not the first entry). */
3717 peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices);
3720 list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) {
3721 if (!has_net_conf(peer_device->connection))
3724 goto put_result; /* only one iteration */
3729 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3730 cb->nlh->nlmsg_seq, &drbd_genl_family,
3731 NLM_F_MULTI, DRBD_ADM_GET_PEER_DEVICES);
3735 dh->ret_code = retcode;
3737 if (retcode == NO_ERROR) {
3738 struct peer_device_info peer_device_info;
3739 struct peer_device_statistics peer_device_statistics;
3742 err = nla_put_drbd_cfg_context(skb, device->resource, peer_device->connection, device);
3745 peer_device_to_info(&peer_device_info, peer_device);
3746 err = peer_device_info_to_skb(skb, &peer_device_info, !capable(CAP_SYS_ADMIN));
3749 peer_device_to_statistics(&peer_device_statistics, peer_device);
3750 err = peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
3753 cb->args[1] = minor;
3754 cb->args[2] = (long)peer_device;
3756 genlmsg_end(skb, dh);
3766 * Return the connection of @resource if @resource has exactly one connection.
3768 static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
3770 struct list_head *connections = &resource->connections;
3772 if (list_empty(connections) || connections->next->next != connections)
3774 return list_first_entry(&resource->connections, struct drbd_connection, connections);
3777 static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
3778 const struct sib_info *sib)
3780 struct drbd_resource *resource = device->resource;
3781 struct state_info *si = NULL; /* for sizeof(si->member); */
3785 int exclude_sensitive;
3787 /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
3788 * to. So we better exclude_sensitive information.
3790 * If sib == NULL, this is drbd_adm_get_status, executed synchronously
3791 * in the context of the requesting user process. Exclude sensitive
3792 * information, unless current has superuser.
3794 * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
3795 * relies on the current implementation of netlink_dump(), which
3796 * executes the dump callback successively from netlink_recvmsg(),
3797 * always in the context of the receiving process */
3798 exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
3800 got_ldev = get_ldev(device);
3802 /* We need to add connection name and volume number information still.
3803 * Minor number is in drbd_genlmsghdr. */
3804 if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device))
3805 goto nla_put_failure;
3807 if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive))
3808 goto nla_put_failure;
3812 struct disk_conf *disk_conf;
3814 disk_conf = rcu_dereference(device->ldev->disk_conf);
3815 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
3818 struct net_conf *nc;
3820 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3822 err = net_conf_to_skb(skb, nc, exclude_sensitive);
3826 goto nla_put_failure;
3828 nla = nla_nest_start_noflag(skb, DRBD_NLA_STATE_INFO);
3830 goto nla_put_failure;
3831 if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
3832 nla_put_u32(skb, T_current_state, device->state.i) ||
3833 nla_put_u64_0pad(skb, T_ed_uuid, device->ed_uuid) ||
3834 nla_put_u64_0pad(skb, T_capacity,
3835 drbd_get_capacity(device->this_bdev)) ||
3836 nla_put_u64_0pad(skb, T_send_cnt, device->send_cnt) ||
3837 nla_put_u64_0pad(skb, T_recv_cnt, device->recv_cnt) ||
3838 nla_put_u64_0pad(skb, T_read_cnt, device->read_cnt) ||
3839 nla_put_u64_0pad(skb, T_writ_cnt, device->writ_cnt) ||
3840 nla_put_u64_0pad(skb, T_al_writ_cnt, device->al_writ_cnt) ||
3841 nla_put_u64_0pad(skb, T_bm_writ_cnt, device->bm_writ_cnt) ||
3842 nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) ||
3843 nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) ||
3844 nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt)))
3845 goto nla_put_failure;
3850 spin_lock_irq(&device->ldev->md.uuid_lock);
3851 err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid);
3852 spin_unlock_irq(&device->ldev->md.uuid_lock);
3855 goto nla_put_failure;
3857 if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) ||
3858 nla_put_u64_0pad(skb, T_bits_total, drbd_bm_bits(device)) ||
3859 nla_put_u64_0pad(skb, T_bits_oos,
3860 drbd_bm_total_weight(device)))
3861 goto nla_put_failure;
3862 if (C_SYNC_SOURCE <= device->state.conn &&
3863 C_PAUSED_SYNC_T >= device->state.conn) {
3864 if (nla_put_u64_0pad(skb, T_bits_rs_total,
3865 device->rs_total) ||
3866 nla_put_u64_0pad(skb, T_bits_rs_failed,
3868 goto nla_put_failure;
3873 switch(sib->sib_reason) {
3874 case SIB_SYNC_PROGRESS:
3875 case SIB_GET_STATUS_REPLY:
3877 case SIB_STATE_CHANGE:
3878 if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
3879 nla_put_u32(skb, T_new_state, sib->ns.i))
3880 goto nla_put_failure;
3882 case SIB_HELPER_POST:
3883 if (nla_put_u32(skb, T_helper_exit_code,
3884 sib->helper_exit_code))
3885 goto nla_put_failure;
3887 case SIB_HELPER_PRE:
3888 if (nla_put_string(skb, T_helper, sib->helper_name))
3889 goto nla_put_failure;
3893 nla_nest_end(skb, nla);
3903 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3905 struct drbd_config_context adm_ctx;
3906 enum drbd_ret_code retcode;
3909 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3910 if (!adm_ctx.reply_skb)
3912 if (retcode != NO_ERROR)
3915 err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL);
3917 nlmsg_free(adm_ctx.reply_skb);
3921 drbd_adm_finish(&adm_ctx, info, retcode);
3925 static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3927 struct drbd_device *device;
3928 struct drbd_genlmsghdr *dh;
3929 struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3930 struct drbd_resource *resource = NULL;
3931 struct drbd_resource *tmp;
3932 unsigned volume = cb->args[1];
3934 /* Open coded, deferred, iteration:
3935 * for_each_resource_safe(resource, tmp, &drbd_resources) {
3936 * connection = "first connection of resource or undefined";
3937 * idr_for_each_entry(&resource->devices, device, i) {
3941 * where resource is cb->args[0];
3942 * and i is cb->args[1];
3944 * cb->args[2] indicates if we shall loop over all resources,
3945 * or just dump all volumes of a single resource.
3947 * This may miss entries inserted after this dump started,
3948 * or entries deleted before they are reached.
3950 * We need to make sure the device won't disappear while
3951 * we are looking at it, and revalidate our iterators
3952 * on each iteration.
3955 /* synchronize with conn_create()/drbd_destroy_connection() */
3957 /* revalidate iterator position */
3958 for_each_resource_rcu(tmp, &drbd_resources) {
3960 /* first iteration */
3972 device = idr_get_next(&resource->devices, &volume);
3974 /* No more volumes to dump on this resource.
3975 * Advance resource iterator. */
3976 pos = list_entry_rcu(resource->resources.next,
3977 struct drbd_resource, resources);
3978 /* Did we dump any volume of this resource yet? */
3980 /* If we reached the end of the list,
3981 * or only a single resource dump was requested,
3983 if (&pos->resources == &drbd_resources || cb->args[2])
3991 dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3992 cb->nlh->nlmsg_seq, &drbd_genl_family,
3993 NLM_F_MULTI, DRBD_ADM_GET_STATUS);
3998 /* This is a connection without a single volume.
3999 * Suprisingly enough, it may have a network
4001 struct drbd_connection *connection;
4004 dh->ret_code = NO_ERROR;
4005 connection = the_only_connection(resource);
4006 if (nla_put_drbd_cfg_context(skb, resource, connection, NULL))
4009 struct net_conf *nc;
4011 nc = rcu_dereference(connection->net_conf);
4012 if (nc && net_conf_to_skb(skb, nc, 1) != 0)
4018 D_ASSERT(device, device->vnr == volume);
4019 D_ASSERT(device, device->resource == resource);
4021 dh->minor = device_to_minor(device);
4022 dh->ret_code = NO_ERROR;
4024 if (nla_put_status_info(skb, device, NULL)) {
4026 genlmsg_cancel(skb, dh);
4030 genlmsg_end(skb, dh);
4035 /* where to start the next iteration */
4036 cb->args[0] = (long)pos;
4037 cb->args[1] = (pos == resource) ? volume + 1 : 0;
4039 /* No more resources/volumes/minors found results in an empty skb.
4040 * Which will terminate the dump. */
4045 * Request status of all resources, or of all volumes within a single resource.
4047 * This is a dump, as the answer may not fit in a single reply skb otherwise.
4048 * Which means we cannot use the family->attrbuf or other such members, because
4049 * dump is NOT protected by the genl_lock(). During dump, we only have access
4050 * to the incoming skb, and need to opencode "parsing" of the nlattr payload.
4052 * Once things are setup properly, we call into get_one_status().
4054 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
4056 const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
4058 const char *resource_name;
4059 struct drbd_resource *resource;
4062 /* Is this a followup call? */
4064 /* ... of a single resource dump,
4065 * and the resource iterator has been advanced already? */
4066 if (cb->args[2] && cb->args[2] != cb->args[0])
4067 return 0; /* DONE. */
4071 /* First call (from netlink_dump_start). We need to figure out
4072 * which resource(s) the user wants us to dump. */
4073 nla = nla_find(nlmsg_attrdata(cb->nlh, hdrlen),
4074 nlmsg_attrlen(cb->nlh, hdrlen),
4075 DRBD_NLA_CFG_CONTEXT);
4077 /* No explicit context given. Dump all. */
4080 maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
4081 nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
4083 return PTR_ERR(nla);
4084 /* context given, but no name present? */
4087 resource_name = nla_data(nla);
4088 if (!*resource_name)
4090 resource = drbd_find_resource(resource_name);
4094 kref_put(&resource->kref, drbd_destroy_resource); /* get_one_status() revalidates the resource */
4096 /* prime iterators, and set "filter" mode mark:
4097 * only dump this connection. */
4098 cb->args[0] = (long)resource;
4099 /* cb->args[1] = 0; passed in this way. */
4100 cb->args[2] = (long)resource;
4103 return get_one_status(skb, cb);
4106 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
4108 struct drbd_config_context adm_ctx;
4109 enum drbd_ret_code retcode;
4110 struct timeout_parms tp;
4113 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4114 if (!adm_ctx.reply_skb)
4116 if (retcode != NO_ERROR)
4120 adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
4121 test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED :
4124 err = timeout_parms_to_priv_skb(adm_ctx.reply_skb, &tp);
4126 nlmsg_free(adm_ctx.reply_skb);
4130 drbd_adm_finish(&adm_ctx, info, retcode);
4134 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
4136 struct drbd_config_context adm_ctx;
4137 struct drbd_device *device;
4138 enum drbd_ret_code retcode;
4139 struct start_ov_parms parms;
4141 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4142 if (!adm_ctx.reply_skb)
4144 if (retcode != NO_ERROR)
4147 device = adm_ctx.device;
4149 /* resume from last known position, if possible */
4150 parms.ov_start_sector = device->ov_start_sector;
4151 parms.ov_stop_sector = ULLONG_MAX;
4152 if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
4153 int err = start_ov_parms_from_attrs(&parms, info);
4155 retcode = ERR_MANDATORY_TAG;
4156 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4160 mutex_lock(&adm_ctx.resource->adm_mutex);
4162 /* w_make_ov_request expects position to be aligned */
4163 device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
4164 device->ov_stop_sector = parms.ov_stop_sector;
4166 /* If there is still bitmap IO pending, e.g. previous resync or verify
4167 * just being finished, wait for it before requesting a new resync. */
4168 drbd_suspend_io(device);
4169 wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
4170 retcode = drbd_request_state(device, NS(conn, C_VERIFY_S));
4171 drbd_resume_io(device);
4173 mutex_unlock(&adm_ctx.resource->adm_mutex);
4175 drbd_adm_finish(&adm_ctx, info, retcode);
4180 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
4182 struct drbd_config_context adm_ctx;
4183 struct drbd_device *device;
4184 enum drbd_ret_code retcode;
4185 int skip_initial_sync = 0;
4187 struct new_c_uuid_parms args;
4189 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4190 if (!adm_ctx.reply_skb)
4192 if (retcode != NO_ERROR)
4195 device = adm_ctx.device;
4196 memset(&args, 0, sizeof(args));
4197 if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
4198 err = new_c_uuid_parms_from_attrs(&args, info);
4200 retcode = ERR_MANDATORY_TAG;
4201 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4206 mutex_lock(&adm_ctx.resource->adm_mutex);
4207 mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */
4209 if (!get_ldev(device)) {
4210 retcode = ERR_NO_DISK;
4214 /* this is "skip initial sync", assume to be clean */
4215 if (device->state.conn == C_CONNECTED &&
4216 first_peer_device(device)->connection->agreed_pro_version >= 90 &&
4217 device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
4218 drbd_info(device, "Preparing to skip initial sync\n");
4219 skip_initial_sync = 1;
4220 } else if (device->state.conn != C_STANDALONE) {
4221 retcode = ERR_CONNECTED;
4225 drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
4226 drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
4228 if (args.clear_bm) {
4229 err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
4230 "clear_n_write from new_c_uuid", BM_LOCKED_MASK);
4232 drbd_err(device, "Writing bitmap failed with %d\n", err);
4233 retcode = ERR_IO_MD_DISK;
4235 if (skip_initial_sync) {
4236 drbd_send_uuids_skip_initial_sync(first_peer_device(device));
4237 _drbd_uuid_set(device, UI_BITMAP, 0);
4238 drbd_print_uuids(device, "cleared bitmap UUID");
4239 spin_lock_irq(&device->resource->req_lock);
4240 _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
4242 spin_unlock_irq(&device->resource->req_lock);
4246 drbd_md_sync(device);
4250 mutex_unlock(device->state_mutex);
4251 mutex_unlock(&adm_ctx.resource->adm_mutex);
4253 drbd_adm_finish(&adm_ctx, info, retcode);
4257 static enum drbd_ret_code
4258 drbd_check_resource_name(struct drbd_config_context *adm_ctx)
4260 const char *name = adm_ctx->resource_name;
4261 if (!name || !name[0]) {
4262 drbd_msg_put_info(adm_ctx->reply_skb, "resource name missing");
4263 return ERR_MANDATORY_TAG;
4265 /* if we want to use these in sysfs/configfs/debugfs some day,
4266 * we must not allow slashes */
4267 if (strchr(name, '/')) {
4268 drbd_msg_put_info(adm_ctx->reply_skb, "invalid resource name");
4269 return ERR_INVALID_REQUEST;
4274 static void resource_to_info(struct resource_info *info,
4275 struct drbd_resource *resource)
4277 info->res_role = conn_highest_role(first_connection(resource));
4278 info->res_susp = resource->susp;
4279 info->res_susp_nod = resource->susp_nod;
4280 info->res_susp_fen = resource->susp_fen;
4283 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
4285 struct drbd_connection *connection;
4286 struct drbd_config_context adm_ctx;
4287 enum drbd_ret_code retcode;
4288 struct res_opts res_opts;
4291 retcode = drbd_adm_prepare(&adm_ctx, skb, info, 0);
4292 if (!adm_ctx.reply_skb)
4294 if (retcode != NO_ERROR)
4297 set_res_opts_defaults(&res_opts);
4298 err = res_opts_from_attrs(&res_opts, info);
4299 if (err && err != -ENOMSG) {
4300 retcode = ERR_MANDATORY_TAG;
4301 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
4305 retcode = drbd_check_resource_name(&adm_ctx);
4306 if (retcode != NO_ERROR)
4309 if (adm_ctx.resource) {
4310 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
4311 retcode = ERR_INVALID_REQUEST;
4312 drbd_msg_put_info(adm_ctx.reply_skb, "resource exists");
4314 /* else: still NO_ERROR */
4318 /* not yet safe for genl_family.parallel_ops */
4319 mutex_lock(&resources_mutex);
4320 connection = conn_create(adm_ctx.resource_name, &res_opts);
4321 mutex_unlock(&resources_mutex);
4324 struct resource_info resource_info;
4326 mutex_lock(¬ification_mutex);
4327 resource_to_info(&resource_info, connection->resource);
4328 notify_resource_state(NULL, 0, connection->resource,
4329 &resource_info, NOTIFY_CREATE);
4330 mutex_unlock(¬ification_mutex);
4332 retcode = ERR_NOMEM;
4335 drbd_adm_finish(&adm_ctx, info, retcode);
4339 static void device_to_info(struct device_info *info,
4340 struct drbd_device *device)
4342 info->dev_disk_state = device->state.disk;
4346 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info)
4348 struct drbd_config_context adm_ctx;
4349 struct drbd_genlmsghdr *dh = info->userhdr;
4350 enum drbd_ret_code retcode;
4352 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4353 if (!adm_ctx.reply_skb)
4355 if (retcode != NO_ERROR)
4358 if (dh->minor > MINORMASK) {
4359 drbd_msg_put_info(adm_ctx.reply_skb, "requested minor out of range");
4360 retcode = ERR_INVALID_REQUEST;
4363 if (adm_ctx.volume > DRBD_VOLUME_MAX) {
4364 drbd_msg_put_info(adm_ctx.reply_skb, "requested volume id out of range");
4365 retcode = ERR_INVALID_REQUEST;
4369 /* drbd_adm_prepare made sure already
4370 * that first_peer_device(device)->connection and device->vnr match the request. */
4371 if (adm_ctx.device) {
4372 if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
4373 retcode = ERR_MINOR_OR_VOLUME_EXISTS;
4374 /* else: still NO_ERROR */
4378 mutex_lock(&adm_ctx.resource->adm_mutex);
4379 retcode = drbd_create_device(&adm_ctx, dh->minor);
4380 if (retcode == NO_ERROR) {
4381 struct drbd_device *device;
4382 struct drbd_peer_device *peer_device;
4383 struct device_info info;
4384 unsigned int peer_devices = 0;
4385 enum drbd_notification_type flags;
4387 device = minor_to_device(dh->minor);
4388 for_each_peer_device(peer_device, device) {
4389 if (!has_net_conf(peer_device->connection))
4394 device_to_info(&info, device);
4395 mutex_lock(¬ification_mutex);
4396 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4397 notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags);
4398 for_each_peer_device(peer_device, device) {
4399 struct peer_device_info peer_device_info;
4401 if (!has_net_conf(peer_device->connection))
4403 peer_device_to_info(&peer_device_info, peer_device);
4404 flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4405 notify_peer_device_state(NULL, 0, peer_device, &peer_device_info,
4406 NOTIFY_CREATE | flags);
4408 mutex_unlock(¬ification_mutex);
4410 mutex_unlock(&adm_ctx.resource->adm_mutex);
4412 drbd_adm_finish(&adm_ctx, info, retcode);
4416 static enum drbd_ret_code adm_del_minor(struct drbd_device *device)
4418 struct drbd_peer_device *peer_device;
4420 if (device->state.disk == D_DISKLESS &&
4421 /* no need to be device->state.conn == C_STANDALONE &&
4422 * we may want to delete a minor from a live replication group.
4424 device->state.role == R_SECONDARY) {
4425 struct drbd_connection *connection =
4426 first_connection(device->resource);
4428 _drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS),
4429 CS_VERBOSE + CS_WAIT_COMPLETE);
4431 /* If the state engine hasn't stopped the sender thread yet, we
4432 * need to flush the sender work queue before generating the
4433 * DESTROY events here. */
4434 if (get_t_state(&connection->worker) == RUNNING)
4435 drbd_flush_workqueue(&connection->sender_work);
4437 mutex_lock(¬ification_mutex);
4438 for_each_peer_device(peer_device, device) {
4439 if (!has_net_conf(peer_device->connection))
4441 notify_peer_device_state(NULL, 0, peer_device, NULL,
4442 NOTIFY_DESTROY | NOTIFY_CONTINUES);
4444 notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY);
4445 mutex_unlock(¬ification_mutex);
4447 drbd_delete_device(device);
4450 return ERR_MINOR_CONFIGURED;
4453 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info)
4455 struct drbd_config_context adm_ctx;
4456 enum drbd_ret_code retcode;
4458 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4459 if (!adm_ctx.reply_skb)
4461 if (retcode != NO_ERROR)
4464 mutex_lock(&adm_ctx.resource->adm_mutex);
4465 retcode = adm_del_minor(adm_ctx.device);
4466 mutex_unlock(&adm_ctx.resource->adm_mutex);
4468 drbd_adm_finish(&adm_ctx, info, retcode);
4472 static int adm_del_resource(struct drbd_resource *resource)
4474 struct drbd_connection *connection;
4476 for_each_connection(connection, resource) {
4477 if (connection->cstate > C_STANDALONE)
4478 return ERR_NET_CONFIGURED;
4480 if (!idr_is_empty(&resource->devices))
4481 return ERR_RES_IN_USE;
4483 /* The state engine has stopped the sender thread, so we don't
4484 * need to flush the sender work queue before generating the
4485 * DESTROY event here. */
4486 mutex_lock(¬ification_mutex);
4487 notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY);
4488 mutex_unlock(¬ification_mutex);
4490 mutex_lock(&resources_mutex);
4491 list_del_rcu(&resource->resources);
4492 mutex_unlock(&resources_mutex);
4493 /* Make sure all threads have actually stopped: state handling only
4494 * does drbd_thread_stop_nowait(). */
4495 list_for_each_entry(connection, &resource->connections, connections)
4496 drbd_thread_stop(&connection->worker);
4498 drbd_free_resource(resource);
4502 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
4504 struct drbd_config_context adm_ctx;
4505 struct drbd_resource *resource;
4506 struct drbd_connection *connection;
4507 struct drbd_device *device;
4508 int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
4511 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4512 if (!adm_ctx.reply_skb)
4514 if (retcode != NO_ERROR)
4517 resource = adm_ctx.resource;
4518 mutex_lock(&resource->adm_mutex);
4520 for_each_connection(connection, resource) {
4521 struct drbd_peer_device *peer_device;
4523 idr_for_each_entry(&connection->peer_devices, peer_device, i) {
4524 retcode = drbd_set_role(peer_device->device, R_SECONDARY, 0);
4525 if (retcode < SS_SUCCESS) {
4526 drbd_msg_put_info(adm_ctx.reply_skb, "failed to demote");
4531 retcode = conn_try_disconnect(connection, 0);
4532 if (retcode < SS_SUCCESS) {
4533 drbd_msg_put_info(adm_ctx.reply_skb, "failed to disconnect");
4539 idr_for_each_entry(&resource->devices, device, i) {
4540 retcode = adm_detach(device, 0);
4541 if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
4542 drbd_msg_put_info(adm_ctx.reply_skb, "failed to detach");
4547 /* delete volumes */
4548 idr_for_each_entry(&resource->devices, device, i) {
4549 retcode = adm_del_minor(device);
4550 if (retcode != NO_ERROR) {
4551 /* "can not happen" */
4552 drbd_msg_put_info(adm_ctx.reply_skb, "failed to delete volume");
4557 retcode = adm_del_resource(resource);
4559 mutex_unlock(&resource->adm_mutex);
4561 drbd_adm_finish(&adm_ctx, info, retcode);
4565 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
4567 struct drbd_config_context adm_ctx;
4568 struct drbd_resource *resource;
4569 enum drbd_ret_code retcode;
4571 retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4572 if (!adm_ctx.reply_skb)
4574 if (retcode != NO_ERROR)
4576 resource = adm_ctx.resource;
4578 mutex_lock(&resource->adm_mutex);
4579 retcode = adm_del_resource(resource);
4580 mutex_unlock(&resource->adm_mutex);
4582 drbd_adm_finish(&adm_ctx, info, retcode);
4586 void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib)
4588 struct sk_buff *msg;
4589 struct drbd_genlmsghdr *d_out;
4593 seq = atomic_inc_return(&drbd_genl_seq);
4594 msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4599 d_out = genlmsg_put(msg, 0, seq, &drbd_genl_family, 0, DRBD_EVENT);
4600 if (!d_out) /* cannot happen, but anyways. */
4601 goto nla_put_failure;
4602 d_out->minor = device_to_minor(device);
4603 d_out->ret_code = NO_ERROR;
4605 if (nla_put_status_info(msg, device, sib))
4606 goto nla_put_failure;
4607 genlmsg_end(msg, d_out);
4608 err = drbd_genl_multicast_events(msg, GFP_NOWAIT);
4609 /* msg has been consumed or freed in netlink_broadcast() */
4610 if (err && err != -ESRCH)
4618 drbd_err(device, "Error %d while broadcasting event. "
4619 "Event seq:%u sib_reason:%u\n",
4620 err, seq, sib->sib_reason);
4623 static int nla_put_notification_header(struct sk_buff *msg,
4624 enum drbd_notification_type type)
4626 struct drbd_notification_header nh = {
4630 return drbd_notification_header_to_skb(msg, &nh, true);
4633 void notify_resource_state(struct sk_buff *skb,
4635 struct drbd_resource *resource,
4636 struct resource_info *resource_info,
4637 enum drbd_notification_type type)
4639 struct resource_statistics resource_statistics;
4640 struct drbd_genlmsghdr *dh;
4641 bool multicast = false;
4645 seq = atomic_inc_return(¬ify_genl_seq);
4646 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4654 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_RESOURCE_STATE);
4656 goto nla_put_failure;
4658 dh->ret_code = NO_ERROR;
4659 if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) ||
4660 nla_put_notification_header(skb, type) ||
4661 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4662 resource_info_to_skb(skb, resource_info, true)))
4663 goto nla_put_failure;
4664 resource_statistics.res_stat_write_ordering = resource->write_ordering;
4665 err = resource_statistics_to_skb(skb, &resource_statistics, !capable(CAP_SYS_ADMIN));
4667 goto nla_put_failure;
4668 genlmsg_end(skb, dh);
4670 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4671 /* skb has been consumed or freed in netlink_broadcast() */
4672 if (err && err != -ESRCH)
4680 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4684 void notify_device_state(struct sk_buff *skb,
4686 struct drbd_device *device,
4687 struct device_info *device_info,
4688 enum drbd_notification_type type)
4690 struct device_statistics device_statistics;
4691 struct drbd_genlmsghdr *dh;
4692 bool multicast = false;
4696 seq = atomic_inc_return(¬ify_genl_seq);
4697 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4705 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_DEVICE_STATE);
4707 goto nla_put_failure;
4708 dh->minor = device->minor;
4709 dh->ret_code = NO_ERROR;
4710 if (nla_put_drbd_cfg_context(skb, device->resource, NULL, device) ||
4711 nla_put_notification_header(skb, type) ||
4712 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4713 device_info_to_skb(skb, device_info, true)))
4714 goto nla_put_failure;
4715 device_to_statistics(&device_statistics, device);
4716 device_statistics_to_skb(skb, &device_statistics, !capable(CAP_SYS_ADMIN));
4717 genlmsg_end(skb, dh);
4719 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4720 /* skb has been consumed or freed in netlink_broadcast() */
4721 if (err && err != -ESRCH)
4729 drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n",
4733 void notify_connection_state(struct sk_buff *skb,
4735 struct drbd_connection *connection,
4736 struct connection_info *connection_info,
4737 enum drbd_notification_type type)
4739 struct connection_statistics connection_statistics;
4740 struct drbd_genlmsghdr *dh;
4741 bool multicast = false;
4745 seq = atomic_inc_return(¬ify_genl_seq);
4746 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4754 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_CONNECTION_STATE);
4756 goto nla_put_failure;
4758 dh->ret_code = NO_ERROR;
4759 if (nla_put_drbd_cfg_context(skb, connection->resource, connection, NULL) ||
4760 nla_put_notification_header(skb, type) ||
4761 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4762 connection_info_to_skb(skb, connection_info, true)))
4763 goto nla_put_failure;
4764 connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
4765 connection_statistics_to_skb(skb, &connection_statistics, !capable(CAP_SYS_ADMIN));
4766 genlmsg_end(skb, dh);
4768 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4769 /* skb has been consumed or freed in netlink_broadcast() */
4770 if (err && err != -ESRCH)
4778 drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n",
4782 void notify_peer_device_state(struct sk_buff *skb,
4784 struct drbd_peer_device *peer_device,
4785 struct peer_device_info *peer_device_info,
4786 enum drbd_notification_type type)
4788 struct peer_device_statistics peer_device_statistics;
4789 struct drbd_resource *resource = peer_device->device->resource;
4790 struct drbd_genlmsghdr *dh;
4791 bool multicast = false;
4795 seq = atomic_inc_return(¬ify_genl_seq);
4796 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4804 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_PEER_DEVICE_STATE);
4806 goto nla_put_failure;
4808 dh->ret_code = NO_ERROR;
4809 if (nla_put_drbd_cfg_context(skb, resource, peer_device->connection, peer_device->device) ||
4810 nla_put_notification_header(skb, type) ||
4811 ((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4812 peer_device_info_to_skb(skb, peer_device_info, true)))
4813 goto nla_put_failure;
4814 peer_device_to_statistics(&peer_device_statistics, peer_device);
4815 peer_device_statistics_to_skb(skb, &peer_device_statistics, !capable(CAP_SYS_ADMIN));
4816 genlmsg_end(skb, dh);
4818 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4819 /* skb has been consumed or freed in netlink_broadcast() */
4820 if (err && err != -ESRCH)
4828 drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n",
4832 void notify_helper(enum drbd_notification_type type,
4833 struct drbd_device *device, struct drbd_connection *connection,
4834 const char *name, int status)
4836 struct drbd_resource *resource = device ? device->resource : connection->resource;
4837 struct drbd_helper_info helper_info;
4838 unsigned int seq = atomic_inc_return(¬ify_genl_seq);
4839 struct sk_buff *skb = NULL;
4840 struct drbd_genlmsghdr *dh;
4843 strlcpy(helper_info.helper_name, name, sizeof(helper_info.helper_name));
4844 helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name));
4845 helper_info.helper_status = status;
4847 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4853 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_HELPER);
4856 dh->minor = device ? device->minor : -1;
4857 dh->ret_code = NO_ERROR;
4858 mutex_lock(¬ification_mutex);
4859 if (nla_put_drbd_cfg_context(skb, resource, connection, device) ||
4860 nla_put_notification_header(skb, type) ||
4861 drbd_helper_info_to_skb(skb, &helper_info, true))
4863 genlmsg_end(skb, dh);
4864 err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4866 /* skb has been consumed or freed in netlink_broadcast() */
4867 if (err && err != -ESRCH)
4869 mutex_unlock(¬ification_mutex);
4873 mutex_unlock(¬ification_mutex);
4876 drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4880 static void notify_initial_state_done(struct sk_buff *skb, unsigned int seq)
4882 struct drbd_genlmsghdr *dh;
4886 dh = genlmsg_put(skb, 0, seq, &drbd_genl_family, 0, DRBD_INITIAL_STATE_DONE);
4888 goto nla_put_failure;
4890 dh->ret_code = NO_ERROR;
4891 if (nla_put_notification_header(skb, NOTIFY_EXISTS))
4892 goto nla_put_failure;
4893 genlmsg_end(skb, dh);
4898 pr_err("Error %d sending event. Event seq:%u\n", err, seq);
4901 static void free_state_changes(struct list_head *list)
4903 while (!list_empty(list)) {
4904 struct drbd_state_change *state_change =
4905 list_first_entry(list, struct drbd_state_change, list);
4906 list_del(&state_change->list);
4907 forget_state_change(state_change);
4911 static unsigned int notifications_for_state_change(struct drbd_state_change *state_change)
4914 state_change->n_connections +
4915 state_change->n_devices +
4916 state_change->n_devices * state_change->n_connections;
4919 static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4921 struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0];
4922 unsigned int seq = cb->args[2];
4924 enum drbd_notification_type flags = 0;
4926 /* There is no need for taking notification_mutex here: it doesn't
4927 matter if the initial state events mix with later state chage
4928 events; we can always tell the events apart by the NOTIFY_EXISTS
4932 if (cb->args[5] == 1) {
4933 notify_initial_state_done(skb, seq);
4937 if (cb->args[4] < cb->args[3])
4938 flags |= NOTIFY_CONTINUES;
4940 notify_resource_state_change(skb, seq, state_change->resource,
4941 NOTIFY_EXISTS | flags);
4945 if (n < state_change->n_connections) {
4946 notify_connection_state_change(skb, seq, &state_change->connections[n],
4947 NOTIFY_EXISTS | flags);
4950 n -= state_change->n_connections;
4951 if (n < state_change->n_devices) {
4952 notify_device_state_change(skb, seq, &state_change->devices[n],
4953 NOTIFY_EXISTS | flags);
4956 n -= state_change->n_devices;
4957 if (n < state_change->n_devices * state_change->n_connections) {
4958 notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n],
4959 NOTIFY_EXISTS | flags);
4964 if (cb->args[4] == cb->args[3]) {
4965 struct drbd_state_change *next_state_change =
4966 list_entry(state_change->list.next,
4967 struct drbd_state_change, list);
4968 cb->args[0] = (long)next_state_change;
4969 cb->args[3] = notifications_for_state_change(next_state_change);
4976 int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4978 struct drbd_resource *resource;
4981 if (cb->args[5] >= 1) {
4982 if (cb->args[5] > 1)
4983 return get_initial_state(skb, cb);
4985 struct drbd_state_change *state_change =
4986 (struct drbd_state_change *)cb->args[0];
4988 /* connect list to head */
4989 list_add(&head, &state_change->list);
4990 free_state_changes(&head);
4995 cb->args[5] = 2; /* number of iterations */
4996 mutex_lock(&resources_mutex);
4997 for_each_resource(resource, &drbd_resources) {
4998 struct drbd_state_change *state_change;
5000 state_change = remember_old_state(resource, GFP_KERNEL);
5001 if (!state_change) {
5002 if (!list_empty(&head))
5003 free_state_changes(&head);
5004 mutex_unlock(&resources_mutex);
5007 copy_old_to_new_state_change(state_change);
5008 list_add_tail(&state_change->list, &head);
5009 cb->args[5] += notifications_for_state_change(state_change);
5011 mutex_unlock(&resources_mutex);
5013 if (!list_empty(&head)) {
5014 struct drbd_state_change *state_change =
5015 list_entry(head.next, struct drbd_state_change, list);
5016 cb->args[0] = (long)state_change;
5017 cb->args[3] = notifications_for_state_change(state_change);
5018 list_del(&head); /* detach list from head */
5021 cb->args[2] = cb->nlh->nlmsg_seq;
5022 return get_initial_state(skb, cb);