1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	drbd_nl.c
4
5	This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6
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>.
10
11
12	*/
13
14	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
16	#include <linux/module.h>
17	#include <linux/drbd.h>
18	#include <linux/in.h>
19	#include <linux/fs.h>
20	#include <linux/file.h>
21	#include <linux/slab.h>
22	#include <linux/blkpg.h>
23	#include <linux/cpumask.h>
24	#include "drbd_int.h"
25	#include "drbd_protocol.h"
26	#include "drbd_req.h"
27	#include "drbd_state_change.h"
28	#include <asm/unaligned.h>
29	#include <linux/drbd_limits.h>
30	#include <linux/kthread.h>
31
32	#include <net/genetlink.h>
33
34	/* .doit */
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);
37
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);
40
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);
44
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);
65	/* .dumpit */
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);
75
76	#include <linux/drbd_genl_api.h>
77	#include "drbd_nla.h"
78	#include <linux/genl_magic_func.h>
79
80	static atomic_t drbd_genl_seq = ATOMIC_INIT(2); /* two. */
81	static atomic_t notify_genl_seq = ATOMIC_INIT(2); /* two. */
82
83	DEFINE_MUTEX(notification_mutex);
84
85	/* used bdev_open_by_path, to claim our meta data device(s) */
86	static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
87
88	static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
89	{
90	genlmsg_end(skb, hdr: genlmsg_data(gnlh: nlmsg_data(nlh: nlmsg_hdr(skb))));
91	if (genlmsg_reply(skb, info))
92	pr_err("error sending genl reply\n");
93	}
94
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)
98	{
99	struct nlattr *nla;
100	int err = -EMSGSIZE;
101
102	if (!info || !info[0])
103	return 0;
104
105	nla = nla_nest_start_noflag(skb, attrtype: DRBD_NLA_CFG_REPLY);
106	if (!nla)
107	return err;
108
109	err = nla_put_string(skb, attrtype: T_info_text, str: info);
110	if (err) {
111	nla_nest_cancel(skb, start: nla);
112	return err;
113	} else
114	nla_nest_end(skb, start: nla);
115	return 0;
116	}
117
118	__printf(2, 3)
119	static int drbd_msg_sprintf_info(struct sk_buff *skb, const char *fmt, ...)
120	{
121	va_list args;
122	struct nlattr *nla, *txt;
123	int err = -EMSGSIZE;
124	int len;
125
126	nla = nla_nest_start_noflag(skb, attrtype: DRBD_NLA_CFG_REPLY);
127	if (!nla)
128	return err;
129
130	txt = nla_reserve(skb, attrtype: T_info_text, attrlen: 256);
131	if (!txt) {
132	nla_nest_cancel(skb, start: nla);
133	return err;
134	}
135	va_start(args, fmt);
136	len = vscnprintf(buf: nla_data(nla: txt), size: 256, fmt, args);
137	va_end(args);
138
139	/* maybe: retry with larger reserve, if truncated */
140	txt->nla_len = nla_attr_size(payload: len+1);
141	nlmsg_trim(skb, mark: (char*)txt + NLA_ALIGN(txt->nla_len));
142	nla_nest_end(skb, start: nla);
143
144	return 0;
145	}
146
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.
151	*
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.
155	*/
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)
161	{
162	struct drbd_genlmsghdr *d_in = genl_info_userhdr(info);
163	const u8 cmd = info->genlhdr->cmd;
164	int err;
165
166	memset(adm_ctx, 0, sizeof(*adm_ctx));
167
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))
170	return -EPERM;
171
172	adm_ctx->reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
173	if (!adm_ctx->reply_skb) {
174	err = -ENOMEM;
175	goto fail;
176	}
177
178	adm_ctx->reply_dh = genlmsg_put_reply(skb: adm_ctx->reply_skb,
179	info, family: &drbd_genl_family, flags: 0, cmd);
180	/* put of a few bytes into a fresh skb of >= 4k will always succeed.
181	* but anyways */
182	if (!adm_ctx->reply_dh) {
183	err = -ENOMEM;
184	goto fail;
185	}
186
187	adm_ctx->reply_dh->minor = d_in->minor;
188	adm_ctx->reply_dh->ret_code = NO_ERROR;
189
190	adm_ctx->volume = VOLUME_UNSPECIFIED;
191	if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
192	struct nlattr *nla;
193	/* parse and validate only */
194	err = drbd_cfg_context_from_attrs(NULL, info);
195	if (err)
196	goto fail;
197
198	/* It was present, and valid,
199	* copy it over to the reply skb. */
200	err = nla_put_nohdr(skb: adm_ctx->reply_skb,
201	attrlen: info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
202	data: info->attrs[DRBD_NLA_CFG_CONTEXT]);
203	if (err)
204	goto fail;
205
206	/* and assign stuff to the adm_ctx */
207	nla = nested_attr_tb[__nla_type(T_ctx_volume)];
208	if (nla)
209	adm_ctx->volume = nla_get_u32(nla);
210	nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
211	if (nla)
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(nla: adm_ctx->my_addr) > sizeof(adm_ctx->connection->my_addr)) ||
217	(adm_ctx->peer_addr &&
218	nla_len(nla: adm_ctx->peer_addr) > sizeof(adm_ctx->connection->peer_addr))) {
219	err = -EINVAL;
220	goto fail;
221	}
222	}
223
224	adm_ctx->minor = d_in->minor;
225	adm_ctx->device = minor_to_device(minor: d_in->minor);
226
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. */
230	if (adm_ctx->device)
231	kref_get(kref: &adm_ctx->device->kref);
232
233	if (adm_ctx->resource_name) {
234	adm_ctx->resource = drbd_find_resource(name: adm_ctx->resource_name);
235	}
236
237	if (!adm_ctx->device && (flags & DRBD_ADM_NEED_MINOR)) {
238	drbd_msg_put_info(skb: adm_ctx->reply_skb, info: "unknown minor");
239	return ERR_MINOR_INVALID;
240	}
241	if (!adm_ctx->resource && (flags & DRBD_ADM_NEED_RESOURCE)) {
242	drbd_msg_put_info(skb: adm_ctx->reply_skb, info: "unknown resource");
243	if (adm_ctx->resource_name)
244	return ERR_RES_NOT_KNOWN;
245	return ERR_INVALID_REQUEST;
246	}
247
248	if (flags & DRBD_ADM_NEED_CONNECTION) {
249	if (adm_ctx->resource) {
250	drbd_msg_put_info(skb: adm_ctx->reply_skb, info: "no resource name expected");
251	return ERR_INVALID_REQUEST;
252	}
253	if (adm_ctx->device) {
254	drbd_msg_put_info(skb: adm_ctx->reply_skb, info: "no minor number expected");
255	return ERR_INVALID_REQUEST;
256	}
257	if (adm_ctx->my_addr && adm_ctx->peer_addr)
258	adm_ctx->connection = conn_get_by_addrs(my_addr: nla_data(nla: adm_ctx->my_addr),
259	my_addr_len: nla_len(nla: adm_ctx->my_addr),
260	peer_addr: nla_data(nla: adm_ctx->peer_addr),
261	peer_addr_len: nla_len(nla: adm_ctx->peer_addr));
262	if (!adm_ctx->connection) {
263	drbd_msg_put_info(skb: adm_ctx->reply_skb, info: "unknown connection");
264	return ERR_INVALID_REQUEST;
265	}
266	}
267
268	/* some more paranoia, if the request was over-determined */
269	if (adm_ctx->device && adm_ctx->resource &&
270	adm_ctx->device->resource != adm_ctx->resource) {
271	pr_warn("request: minor=%u, resource=%s; but that minor belongs to resource %s\n",
272	adm_ctx->minor, adm_ctx->resource->name,
273	adm_ctx->device->resource->name);
274	drbd_msg_put_info(skb: adm_ctx->reply_skb, info: "minor exists in different resource");
275	return ERR_INVALID_REQUEST;
276	}
277	if (adm_ctx->device &&
278	adm_ctx->volume != VOLUME_UNSPECIFIED &&
279	adm_ctx->volume != adm_ctx->device->vnr) {
280	pr_warn("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
281	adm_ctx->minor, adm_ctx->volume,
282	adm_ctx->device->vnr, adm_ctx->device->resource->name);
283	drbd_msg_put_info(skb: adm_ctx->reply_skb, info: "minor exists as different volume");
284	return ERR_INVALID_REQUEST;
285	}
286
287	/* still, provide adm_ctx->resource always, if possible. */
288	if (!adm_ctx->resource) {
289	adm_ctx->resource = adm_ctx->device ? adm_ctx->device->resource
290	: adm_ctx->connection ? adm_ctx->connection->resource : NULL;
291	if (adm_ctx->resource)
292	kref_get(kref: &adm_ctx->resource->kref);
293	}
294
295	return NO_ERROR;
296
297	fail:
298	nlmsg_free(skb: adm_ctx->reply_skb);
299	adm_ctx->reply_skb = NULL;
300	return err;
301	}
302
303	static int drbd_adm_finish(struct drbd_config_context *adm_ctx,
304	struct genl_info *info, int retcode)
305	{
306	if (adm_ctx->device) {
307	kref_put(kref: &adm_ctx->device->kref, release: drbd_destroy_device);
308	adm_ctx->device = NULL;
309	}
310	if (adm_ctx->connection) {
311	kref_put(kref: &adm_ctx->connection->kref, release: &drbd_destroy_connection);
312	adm_ctx->connection = NULL;
313	}
314	if (adm_ctx->resource) {
315	kref_put(kref: &adm_ctx->resource->kref, release: drbd_destroy_resource);
316	adm_ctx->resource = NULL;
317	}
318
319	if (!adm_ctx->reply_skb)
320	return -ENOMEM;
321
322	adm_ctx->reply_dh->ret_code = retcode;
323	drbd_adm_send_reply(skb: adm_ctx->reply_skb, info);
324	return 0;
325	}
326
327	static void setup_khelper_env(struct drbd_connection *connection, char **envp)
328	{
329	char *afs;
330
331	/* FIXME: A future version will not allow this case. */
332	if (connection->my_addr_len == 0 || connection->peer_addr_len == 0)
333	return;
334
335	switch (((struct sockaddr *)&connection->peer_addr)->sa_family) {
336	case AF_INET6:
337	afs = "ipv6";
338	snprintf(buf: envp[4], size: 60, fmt: "DRBD_PEER_ADDRESS=%pI6",
339	&((struct sockaddr_in6 *)&connection->peer_addr)->sin6_addr);
340	break;
341	case AF_INET:
342	afs = "ipv4";
343	snprintf(buf: envp[4], size: 60, fmt: "DRBD_PEER_ADDRESS=%pI4",
344	&((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
345	break;
346	default:
347	afs = "ssocks";
348	snprintf(buf: envp[4], size: 60, fmt: "DRBD_PEER_ADDRESS=%pI4",
349	&((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
350	}
351	snprintf(buf: envp[3], size: 20, fmt: "DRBD_PEER_AF=%s", afs);
352	}
353
354	int drbd_khelper(struct drbd_device *device, char *cmd)
355	{
356	char *envp[] = { "HOME=/",
357	"TERM=linux",
358	"PATH=/sbin:/usr/sbin:/bin:/usr/bin",
359	(char[20]) { }, /* address family */
360	(char[60]) { }, /* address */
361	NULL };
362	char mb[14];
363	char *argv[] = {drbd_usermode_helper, cmd, mb, NULL };
364	struct drbd_connection *connection = first_peer_device(device)->connection;
365	struct sib_info sib;
366	int ret;
367
368	if (current == connection->worker.task)
369	set_bit(nr: CALLBACK_PENDING, addr: &connection->flags);
370
371	snprintf(buf: mb, size: 14, fmt: "minor-%d", device_to_minor(device));
372	setup_khelper_env(connection, envp);
373
374	/* The helper may take some time.
375	* write out any unsynced meta data changes now */
376	drbd_md_sync(device);
377
378	drbd_info(device, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, mb);
379	sib.sib_reason = SIB_HELPER_PRE;
380	sib.helper_name = cmd;
381	drbd_bcast_event(device, sib: &sib);
382	notify_helper(NOTIFY_CALL, device, connection, cmd, 0);
383	ret = call_usermodehelper(path: drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
384	if (ret)
385	drbd_warn(device, "helper command: %s %s %s exit code %u (0x%x)\n",
386	drbd_usermode_helper, cmd, mb,
387	(ret >> 8) & 0xff, ret);
388	else
389	drbd_info(device, "helper command: %s %s %s exit code %u (0x%x)\n",
390	drbd_usermode_helper, cmd, mb,
391	(ret >> 8) & 0xff, ret);
392	sib.sib_reason = SIB_HELPER_POST;
393	sib.helper_exit_code = ret;
394	drbd_bcast_event(device, sib: &sib);
395	notify_helper(NOTIFY_RESPONSE, device, connection, cmd, ret);
396
397	if (current == connection->worker.task)
398	clear_bit(nr: CALLBACK_PENDING, addr: &connection->flags);
399
400	if (ret < 0) /* Ignore any ERRNOs we got. */
401	ret = 0;
402
403	return ret;
404	}
405
406	enum drbd_peer_state conn_khelper(struct drbd_connection *connection, char *cmd)
407	{
408	char *envp[] = { "HOME=/",
409	"TERM=linux",
410	"PATH=/sbin:/usr/sbin:/bin:/usr/bin",
411	(char[20]) { }, /* address family */
412	(char[60]) { }, /* address */
413	NULL };
414	char *resource_name = connection->resource->name;
415	char *argv[] = {drbd_usermode_helper, cmd, resource_name, NULL };
416	int ret;
417
418	setup_khelper_env(connection, envp);
419	conn_md_sync(connection);
420
421	drbd_info(connection, "helper command: %s %s %s\n", drbd_usermode_helper, cmd, resource_name);
422	/* TODO: conn_bcast_event() ?? */
423	notify_helper(NOTIFY_CALL, NULL, connection, cmd, 0);
424
425	ret = call_usermodehelper(path: drbd_usermode_helper, argv, envp, UMH_WAIT_PROC);
426	if (ret)
427	drbd_warn(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
428	drbd_usermode_helper, cmd, resource_name,
429	(ret >> 8) & 0xff, ret);
430	else
431	drbd_info(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
432	drbd_usermode_helper, cmd, resource_name,
433	(ret >> 8) & 0xff, ret);
434	/* TODO: conn_bcast_event() ?? */
435	notify_helper(NOTIFY_RESPONSE, NULL, connection, cmd, ret);
436
437	if (ret < 0) /* Ignore any ERRNOs we got. */
438	ret = 0;
439
440	return ret;
441	}
442
443	static enum drbd_fencing_p highest_fencing_policy(struct drbd_connection *connection)
444	{
445	enum drbd_fencing_p fp = FP_NOT_AVAIL;
446	struct drbd_peer_device *peer_device;
447	int vnr;
448
449	rcu_read_lock();
450	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
451	struct drbd_device *device = peer_device->device;
452	if (get_ldev_if_state(device, D_CONSISTENT)) {
453	struct disk_conf *disk_conf =
454	rcu_dereference(peer_device->device->ldev->disk_conf);
455	fp = max_t(enum drbd_fencing_p, fp, disk_conf->fencing);
456	put_ldev(device);
457	}
458	}
459	rcu_read_unlock();
460
461	return fp;
462	}
463
464	static bool resource_is_supended(struct drbd_resource *resource)
465	{
466	return resource->susp || resource->susp_fen || resource->susp_nod;
467	}
468
469	bool conn_try_outdate_peer(struct drbd_connection *connection)
470	{
471	struct drbd_resource * const resource = connection->resource;
472	unsigned int connect_cnt;
473	union drbd_state mask = { };
474	union drbd_state val = { };
475	enum drbd_fencing_p fp;
476	char *ex_to_string;
477	int r;
478
479	spin_lock_irq(lock: &resource->req_lock);
480	if (connection->cstate >= C_WF_REPORT_PARAMS) {
481	drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n");
482	spin_unlock_irq(lock: &resource->req_lock);
483	return false;
484	}
485
486	connect_cnt = connection->connect_cnt;
487	spin_unlock_irq(lock: &resource->req_lock);
488
489	fp = highest_fencing_policy(connection);
490	switch (fp) {
491	case FP_NOT_AVAIL:
492	drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n");
493	spin_lock_irq(lock: &resource->req_lock);
494	if (connection->cstate < C_WF_REPORT_PARAMS) {
495	_conn_request_state(connection,
496	mask: (union drbd_state) { { .susp_fen = 1 } },
497	val: (union drbd_state) { { .susp_fen = 0 } },
498	flags: CS_VERBOSE | CS_HARD | CS_DC_SUSP);
499	/* We are no longer suspended due to the fencing policy.
500	* We may still be suspended due to the on-no-data-accessible policy.
501	* If that was OND_IO_ERROR, fail pending requests. */
502	if (!resource_is_supended(resource))
503	_tl_restart(connection, what: CONNECTION_LOST_WHILE_PENDING);
504	}
505	/* Else: in case we raced with a connection handshake,
506	* let the handshake figure out if we maybe can RESEND,
507	* and do not resume/fail pending requests here.
508	* Worst case is we stay suspended for now, which may be
509	* resolved by either re-establishing the replication link, or
510	* the next link failure, or eventually the administrator. */
511	spin_unlock_irq(lock: &resource->req_lock);
512	return false;
513
514	case FP_DONT_CARE:
515	return true;
516	default: ;
517	}
518
519	r = conn_khelper(connection, cmd: "fence-peer");
520
521	switch ((r>>8) & 0xff) {
522	case P_INCONSISTENT: /* peer is inconsistent */
523	ex_to_string = "peer is inconsistent or worse";
524	mask.pdsk = D_MASK;
525	val.pdsk = D_INCONSISTENT;
526	break;
527	case P_OUTDATED: /* peer got outdated, or was already outdated */
528	ex_to_string = "peer was fenced";
529	mask.pdsk = D_MASK;
530	val.pdsk = D_OUTDATED;
531	break;
532	case P_DOWN: /* peer was down */
533	if (conn_highest_disk(connection) == D_UP_TO_DATE) {
534	/* we will(have) create(d) a new UUID anyways... */
535	ex_to_string = "peer is unreachable, assumed to be dead";
536	mask.pdsk = D_MASK;
537	val.pdsk = D_OUTDATED;
538	} else {
539	ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
540	}
541	break;
542	case P_PRIMARY: /* Peer is primary, voluntarily outdate myself.
543	* This is useful when an unconnected R_SECONDARY is asked to
544	* become R_PRIMARY, but finds the other peer being active. */
545	ex_to_string = "peer is active";
546	drbd_warn(connection, "Peer is primary, outdating myself.\n");
547	mask.disk = D_MASK;
548	val.disk = D_OUTDATED;
549	break;
550	case P_FENCING:
551	/* THINK: do we need to handle this
552	* like case 4, or more like case 5? */
553	if (fp != FP_STONITH)
554	drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n");
555	ex_to_string = "peer was stonithed";
556	mask.pdsk = D_MASK;
557	val.pdsk = D_OUTDATED;
558	break;
559	default:
560	/* The script is broken ... */
561	drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
562	return false; /* Eventually leave IO frozen */
563	}
564
565	drbd_info(connection, "fence-peer helper returned %d (%s)\n",
566	(r>>8) & 0xff, ex_to_string);
567
568	/* Not using
569	conn_request_state(connection, mask, val, CS_VERBOSE);
570	here, because we might were able to re-establish the connection in the
571	meantime. */
572	spin_lock_irq(lock: &resource->req_lock);
573	if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) {
574	if (connection->connect_cnt != connect_cnt)
575	/* In case the connection was established and droped
576	while the fence-peer handler was running, ignore it */
577	drbd_info(connection, "Ignoring fence-peer exit code\n");
578	else
579	_conn_request_state(connection, mask, val, flags: CS_VERBOSE);
580	}
581	spin_unlock_irq(lock: &resource->req_lock);
582
583	return conn_highest_pdsk(connection) <= D_OUTDATED;
584	}
585
586	static int _try_outdate_peer_async(void *data)
587	{
588	struct drbd_connection *connection = (struct drbd_connection *)data;
589
590	conn_try_outdate_peer(connection);
591
592	kref_put(kref: &connection->kref, release: drbd_destroy_connection);
593	return 0;
594	}
595
596	void conn_try_outdate_peer_async(struct drbd_connection *connection)
597	{
598	struct task_struct *opa;
599
600	kref_get(kref: &connection->kref);
601	/* We may have just sent a signal to this thread
602	* to get it out of some blocking network function.
603	* Clear signals; otherwise kthread_run(), which internally uses
604	* wait_on_completion_killable(), will mistake our pending signal
605	* for a new fatal signal and fail. */
606	flush_signals(current);
607	opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h");
608	if (IS_ERR(ptr: opa)) {
609	drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n");
610	kref_put(kref: &connection->kref, release: drbd_destroy_connection);
611	}
612	}
613
614	enum drbd_state_rv
615	drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force)
616	{
617	struct drbd_peer_device *const peer_device = first_peer_device(device);
618	struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL;
619	const int max_tries = 4;
620	enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
621	struct net_conf *nc;
622	int try = 0;
623	int forced = 0;
624	union drbd_state mask, val;
625
626	if (new_role == R_PRIMARY) {
627	struct drbd_connection *connection;
628
629	/* Detect dead peers as soon as possible. */
630
631	rcu_read_lock();
632	for_each_connection(connection, device->resource)
633	request_ping(connection);
634	rcu_read_unlock();
635	}
636
637	mutex_lock(device->state_mutex);
638
639	mask.i = 0; mask.role = R_MASK;
640	val.i = 0; val.role = new_role;
641
642	while (try++ < max_tries) {
643	rv = _drbd_request_state_holding_state_mutex(device, mask, val, CS_WAIT_COMPLETE);
644
645	/* in case we first succeeded to outdate,
646	* but now suddenly could establish a connection */
647	if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
648	val.pdsk = 0;
649	mask.pdsk = 0;
650	continue;
651	}
652
653	if (rv == SS_NO_UP_TO_DATE_DISK && force &&
654	(device->state.disk < D_UP_TO_DATE &&
655	device->state.disk >= D_INCONSISTENT)) {
656	mask.disk = D_MASK;
657	val.disk = D_UP_TO_DATE;
658	forced = 1;
659	continue;
660	}
661
662	if (rv == SS_NO_UP_TO_DATE_DISK &&
663	device->state.disk == D_CONSISTENT && mask.pdsk == 0) {
664	D_ASSERT(device, device->state.pdsk == D_UNKNOWN);
665
666	if (conn_try_outdate_peer(connection)) {
667	val.disk = D_UP_TO_DATE;
668	mask.disk = D_MASK;
669	}
670	continue;
671	}
672
673	if (rv == SS_NOTHING_TO_DO)
674	goto out;
675	if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
676	if (!conn_try_outdate_peer(connection) && force) {
677	drbd_warn(device, "Forced into split brain situation!\n");
678	mask.pdsk = D_MASK;
679	val.pdsk = D_OUTDATED;
680
681	}
682	continue;
683	}
684	if (rv == SS_TWO_PRIMARIES) {
685	/* Maybe the peer is detected as dead very soon...
686	retry at most once more in this case. */
687	if (try < max_tries) {
688	int timeo;
689	try = max_tries - 1;
690	rcu_read_lock();
691	nc = rcu_dereference(connection->net_conf);
692	timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
693	rcu_read_unlock();
694	schedule_timeout_interruptible(timeout: timeo);
695	}
696	continue;
697	}
698	if (rv < SS_SUCCESS) {
699	rv = _drbd_request_state(device, mask, val,
700	CS_VERBOSE + CS_WAIT_COMPLETE);
701	if (rv < SS_SUCCESS)
702	goto out;
703	}
704	break;
705	}
706
707	if (rv < SS_SUCCESS)
708	goto out;
709
710	if (forced)
711	drbd_warn(device, "Forced to consider local data as UpToDate!\n");
712
713	/* Wait until nothing is on the fly :) */
714	wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0);
715
716	/* FIXME also wait for all pending P_BARRIER_ACK? */
717
718	if (new_role == R_SECONDARY) {
719	if (get_ldev(device)) {
720	device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
721	put_ldev(device);
722	}
723	} else {
724	mutex_lock(&device->resource->conf_update);
725	nc = connection->net_conf;
726	if (nc)
727	nc->discard_my_data = 0; /* without copy; single bit op is atomic */
728	mutex_unlock(lock: &device->resource->conf_update);
729
730	if (get_ldev(device)) {
731	if (((device->state.conn < C_CONNECTED ||
732	device->state.pdsk <= D_FAILED)
733	&& device->ldev->md.uuid[UI_BITMAP] == 0) || forced)
734	drbd_uuid_new_current(device);
735
736	device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
737	put_ldev(device);
738	}
739	}
740
741	/* writeout of activity log covered areas of the bitmap
742	* to stable storage done in after state change already */
743
744	if (device->state.conn >= C_WF_REPORT_PARAMS) {
745	/* if this was forced, we should consider sync */
746	if (forced)
747	drbd_send_uuids(peer_device);
748	drbd_send_current_state(peer_device);
749	}
750
751	drbd_md_sync(device);
752	set_disk_ro(disk: device->vdisk, read_only: new_role == R_SECONDARY);
753	kobject_uevent(kobj: &disk_to_dev(device->vdisk)->kobj, action: KOBJ_CHANGE);
754	out:
755	mutex_unlock(lock: device->state_mutex);
756	return rv;
757	}
758
759	static const char *from_attrs_err_to_txt(int err)
760	{
761	return err == -ENOMSG ? "required attribute missing" :
762	err == -EOPNOTSUPP ? "unknown mandatory attribute" :
763	err == -EEXIST ? "can not change invariant setting" :
764	"invalid attribute value";
765	}
766
767	int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
768	{
769	struct drbd_config_context adm_ctx;
770	struct set_role_parms parms;
771	int err;
772	enum drbd_ret_code retcode;
773	enum drbd_state_rv rv;
774
775	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
776	if (!adm_ctx.reply_skb)
777	return retcode;
778	if (retcode != NO_ERROR)
779	goto out;
780
781	memset(&parms, 0, sizeof(parms));
782	if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
783	err = set_role_parms_from_attrs(s: &parms, info);
784	if (err) {
785	retcode = ERR_MANDATORY_TAG;
786	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
787	goto out;
788	}
789	}
790	genl_unlock();
791	mutex_lock(&adm_ctx.resource->adm_mutex);
792
793	if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
794	rv = drbd_set_role(device: adm_ctx.device, new_role: R_PRIMARY, force: parms.assume_uptodate);
795	else
796	rv = drbd_set_role(device: adm_ctx.device, new_role: R_SECONDARY, force: 0);
797
798	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
799	genl_lock();
800	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode: rv);
801	return 0;
802	out:
803	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
804	return 0;
805	}
806
807	/* Initializes the md.*_offset members, so we are able to find
808	* the on disk meta data.
809	*
810	* We currently have two possible layouts:
811	* external:
812	* |----------- md_size_sect ------------------|
813	* [ 4k superblock ][ activity log ][ Bitmap ]
814	* | al_offset == 8 |
815	* | bm_offset = al_offset + X |
816	* ==> bitmap sectors = md_size_sect - bm_offset
817	*
818	* internal:
819	* |----------- md_size_sect ------------------|
820	* [data.....][ Bitmap ][ activity log ][ 4k superblock ]
821	* | al_offset < 0 |
822	* | bm_offset = al_offset - Y |
823	* ==> bitmap sectors = Y = al_offset - bm_offset
824	*
825	* Activity log size used to be fixed 32kB,
826	* but is about to become configurable.
827	*/
828	static void drbd_md_set_sector_offsets(struct drbd_device *device,
829	struct drbd_backing_dev *bdev)
830	{
831	sector_t md_size_sect = 0;
832	unsigned int al_size_sect = bdev->md.al_size_4k * 8;
833
834	bdev->md.md_offset = drbd_md_ss(bdev);
835
836	switch (bdev->md.meta_dev_idx) {
837	default:
838	/* v07 style fixed size indexed meta data */
839	bdev->md.md_size_sect = MD_128MB_SECT;
840	bdev->md.al_offset = MD_4kB_SECT;
841	bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
842	break;
843	case DRBD_MD_INDEX_FLEX_EXT:
844	/* just occupy the full device; unit: sectors */
845	bdev->md.md_size_sect = drbd_get_capacity(bdev: bdev->md_bdev);
846	bdev->md.al_offset = MD_4kB_SECT;
847	bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
848	break;
849	case DRBD_MD_INDEX_INTERNAL:
850	case DRBD_MD_INDEX_FLEX_INT:
851	/* al size is still fixed */
852	bdev->md.al_offset = -al_size_sect;
853	/* we need (slightly less than) ~ this much bitmap sectors: */
854	md_size_sect = drbd_get_capacity(bdev: bdev->backing_bdev);
855	md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
856	md_size_sect = BM_SECT_TO_EXT(md_size_sect);
857	md_size_sect = ALIGN(md_size_sect, 8);
858
859	/* plus the "drbd meta data super block",
860	* and the activity log; */
861	md_size_sect += MD_4kB_SECT + al_size_sect;
862
863	bdev->md.md_size_sect = md_size_sect;
864	/* bitmap offset is adjusted by 'super' block size */
865	bdev->md.bm_offset = -md_size_sect + MD_4kB_SECT;
866	break;
867	}
868	}
869
870	/* input size is expected to be in KB */
871	char *ppsize(char *buf, unsigned long long size)
872	{
873	/* Needs 9 bytes at max including trailing NUL:
874	* -1ULL ==> "16384 EB" */
875	static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
876	int base = 0;
877	while (size >= 10000 && base < sizeof(units)-1) {
878	/* shift + round */
879	size = (size >> 10) + !!(size & (1<<9));
880	base++;
881	}
882	sprintf(buf, fmt: "%u %cB", (unsigned)size, units[base]);
883
884	return buf;
885	}
886
887	/* there is still a theoretical deadlock when called from receiver
888	* on an D_INCONSISTENT R_PRIMARY:
889	* remote READ does inc_ap_bio, receiver would need to receive answer
890	* packet from remote to dec_ap_bio again.
891	* receiver receive_sizes(), comes here,
892	* waits for ap_bio_cnt == 0. -> deadlock.
893	* but this cannot happen, actually, because:
894	* R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
895	* (not connected, or bad/no disk on peer):
896	* see drbd_fail_request_early, ap_bio_cnt is zero.
897	* R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
898	* peer may not initiate a resize.
899	*/
900	/* Note these are not to be confused with
901	* drbd_adm_suspend_io/drbd_adm_resume_io,
902	* which are (sub) state changes triggered by admin (drbdsetup),
903	* and can be long lived.
904	* This changes an device->flag, is triggered by drbd internals,
905	* and should be short-lived. */
906	/* It needs to be a counter, since multiple threads might
907	independently suspend and resume IO. */
908	void drbd_suspend_io(struct drbd_device *device)
909	{
910	atomic_inc(v: &device->suspend_cnt);
911	if (drbd_suspended(device))
912	return;
913	wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt));
914	}
915
916	void drbd_resume_io(struct drbd_device *device)
917	{
918	if (atomic_dec_and_test(v: &device->suspend_cnt))
919	wake_up(&device->misc_wait);
920	}
921
922	/*
923	* drbd_determine_dev_size() - Sets the right device size obeying all constraints
924	* @device: DRBD device.
925	*
926	* Returns 0 on success, negative return values indicate errors.
927	* You should call drbd_md_sync() after calling this function.
928	*/
929	enum determine_dev_size
930	drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local)
931	{
932	struct md_offsets_and_sizes {
933	u64 last_agreed_sect;
934	u64 md_offset;
935	s32 al_offset;
936	s32 bm_offset;
937	u32 md_size_sect;
938
939	u32 al_stripes;
940	u32 al_stripe_size_4k;
941	} prev;
942	sector_t u_size, size;
943	struct drbd_md *md = &device->ldev->md;
944	void *buffer;
945
946	int md_moved, la_size_changed;
947	enum determine_dev_size rv = DS_UNCHANGED;
948
949	/* We may change the on-disk offsets of our meta data below. Lock out
950	* anything that may cause meta data IO, to avoid acting on incomplete
951	* layout changes or scribbling over meta data that is in the process
952	* of being moved.
953	*
954	* Move is not exactly correct, btw, currently we have all our meta
955	* data in core memory, to "move" it we just write it all out, there
956	* are no reads. */
957	drbd_suspend_io(device);
958	buffer = drbd_md_get_buffer(device, intent: __func__); /* Lock meta-data IO */
959	if (!buffer) {
960	drbd_resume_io(device);
961	return DS_ERROR;
962	}
963
964	/* remember current offset and sizes */
965	prev.last_agreed_sect = md->la_size_sect;
966	prev.md_offset = md->md_offset;
967	prev.al_offset = md->al_offset;
968	prev.bm_offset = md->bm_offset;
969	prev.md_size_sect = md->md_size_sect;
970	prev.al_stripes = md->al_stripes;
971	prev.al_stripe_size_4k = md->al_stripe_size_4k;
972
973	if (rs) {
974	/* rs is non NULL if we should change the AL layout only */
975	md->al_stripes = rs->al_stripes;
976	md->al_stripe_size_4k = rs->al_stripe_size / 4;
977	md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4;
978	}
979
980	drbd_md_set_sector_offsets(device, bdev: device->ldev);
981
982	rcu_read_lock();
983	u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
984	rcu_read_unlock();
985	size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED);
986
987	if (size < prev.last_agreed_sect) {
988	if (rs && u_size == 0) {
989	/* Remove "rs &&" later. This check should always be active, but
990	right now the receiver expects the permissive behavior */
991	drbd_warn(device, "Implicit shrink not allowed. "
992	"Use --size=%llus for explicit shrink.\n",
993	(unsigned long long)size);
994	rv = DS_ERROR_SHRINK;
995	}
996	if (u_size > size)
997	rv = DS_ERROR_SPACE_MD;
998	if (rv != DS_UNCHANGED)
999	goto err_out;
1000	}
1001
1002	if (get_capacity(disk: device->vdisk) != size ||
1003	drbd_bm_capacity(device) != size) {
1004	int err;
1005	err = drbd_bm_resize(device, sectors: size, set_new_bits: !(flags & DDSF_NO_RESYNC));
1006	if (unlikely(err)) {
1007	/* currently there is only one error: ENOMEM! */
1008	size = drbd_bm_capacity(device);
1009	if (size == 0) {
1010	drbd_err(device, "OUT OF MEMORY! "
1011	"Could not allocate bitmap!\n");
1012	} else {
1013	drbd_err(device, "BM resizing failed. "
1014	"Leaving size unchanged\n");
1015	}
1016	rv = DS_ERROR;
1017	}
1018	/* racy, see comments above. */
1019	drbd_set_my_capacity(device, size);
1020	md->la_size_sect = size;
1021	}
1022	if (rv <= DS_ERROR)
1023	goto err_out;
1024
1025	la_size_changed = (prev.last_agreed_sect != md->la_size_sect);
1026
1027	md_moved = prev.md_offset != md->md_offset
1028	|| prev.md_size_sect != md->md_size_sect;
1029
1030	if (la_size_changed || md_moved || rs) {
1031	u32 prev_flags;
1032
1033	/* We do some synchronous IO below, which may take some time.
1034	* Clear the timer, to avoid scary "timer expired!" messages,
1035	* "Superblock" is written out at least twice below, anyways. */
1036	del_timer(timer: &device->md_sync_timer);
1037
1038	/* We won't change the "al-extents" setting, we just may need
1039	* to move the on-disk location of the activity log ringbuffer.
1040	* Lock for transaction is good enough, it may well be "dirty"
1041	* or even "starving". */
1042	wait_event(device->al_wait, lc_try_lock_for_transaction(device->act_log));
1043
1044	/* mark current on-disk bitmap and activity log as unreliable */
1045	prev_flags = md->flags;
1046	md->flags |= MDF_FULL_SYNC | MDF_AL_DISABLED;
1047	drbd_md_write(device, buffer);
1048
1049	drbd_al_initialize(device, buffer);
1050
1051	drbd_info(device, "Writing the whole bitmap, %s\n",
1052	la_size_changed && md_moved ? "size changed and md moved" :
1053	la_size_changed ? "size changed" : "md moved");
1054	/* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
1055	drbd_bitmap_io(device, io_fn: md_moved ? &drbd_bm_write_all : &drbd_bm_write,
1056	why: "size changed", flags: BM_LOCKED_MASK, NULL);
1057
1058	/* on-disk bitmap and activity log is authoritative again
1059	* (unless there was an IO error meanwhile...) */
1060	md->flags = prev_flags;
1061	drbd_md_write(device, buffer);
1062
1063	if (rs)
1064	drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
1065	md->al_stripes, md->al_stripe_size_4k * 4);
1066	}
1067
1068	if (size > prev.last_agreed_sect)
1069	rv = prev.last_agreed_sect ? DS_GREW : DS_GREW_FROM_ZERO;
1070	if (size < prev.last_agreed_sect)
1071	rv = DS_SHRUNK;
1072
1073	if (0) {
1074	err_out:
1075	/* restore previous offset and sizes */
1076	md->la_size_sect = prev.last_agreed_sect;
1077	md->md_offset = prev.md_offset;
1078	md->al_offset = prev.al_offset;
1079	md->bm_offset = prev.bm_offset;
1080	md->md_size_sect = prev.md_size_sect;
1081	md->al_stripes = prev.al_stripes;
1082	md->al_stripe_size_4k = prev.al_stripe_size_4k;
1083	md->al_size_4k = (u64)prev.al_stripes * prev.al_stripe_size_4k;
1084	}
1085	lc_unlock(lc: device->act_log);
1086	wake_up(&device->al_wait);
1087	drbd_md_put_buffer(device);
1088	drbd_resume_io(device);
1089
1090	return rv;
1091	}
1092
1093	sector_t
1094	drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev,
1095	sector_t u_size, int assume_peer_has_space)
1096	{
1097	sector_t p_size = device->p_size; /* partner's disk size. */
1098	sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
1099	sector_t m_size; /* my size */
1100	sector_t size = 0;
1101
1102	m_size = drbd_get_max_capacity(bdev);
1103
1104	if (device->state.conn < C_CONNECTED && assume_peer_has_space) {
1105	drbd_warn(device, "Resize while not connected was forced by the user!\n");
1106	p_size = m_size;
1107	}
1108
1109	if (p_size && m_size) {
1110	size = min_t(sector_t, p_size, m_size);
1111	} else {
1112	if (la_size_sect) {
1113	size = la_size_sect;
1114	if (m_size && m_size < size)
1115	size = m_size;
1116	if (p_size && p_size < size)
1117	size = p_size;
1118	} else {
1119	if (m_size)
1120	size = m_size;
1121	if (p_size)
1122	size = p_size;
1123	}
1124	}
1125
1126	if (size == 0)
1127	drbd_err(device, "Both nodes diskless!\n");
1128
1129	if (u_size) {
1130	if (u_size > size)
1131	drbd_err(device, "Requested disk size is too big (%lu > %lu)\n",
1132	(unsigned long)u_size>>1, (unsigned long)size>>1);
1133	else
1134	size = u_size;
1135	}
1136
1137	return size;
1138	}
1139
1140	/*
1141	* drbd_check_al_size() - Ensures that the AL is of the right size
1142	* @device: DRBD device.
1143	*
1144	* Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1145	* failed, and 0 on success. You should call drbd_md_sync() after you called
1146	* this function.
1147	*/
1148	static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1149	{
1150	struct lru_cache *n, *t;
1151	struct lc_element *e;
1152	unsigned int in_use;
1153	int i;
1154
1155	if (device->act_log &&
1156	device->act_log->nr_elements == dc->al_extents)
1157	return 0;
1158
1159	in_use = 0;
1160	t = device->act_log;
1161	n = lc_create(name: "act_log", cache: drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
1162	e_count: dc->al_extents, e_size: sizeof(struct lc_element), e_off: 0);
1163
1164	if (n == NULL) {
1165	drbd_err(device, "Cannot allocate act_log lru!\n");
1166	return -ENOMEM;
1167	}
1168	spin_lock_irq(lock: &device->al_lock);
1169	if (t) {
1170	for (i = 0; i < t->nr_elements; i++) {
1171	e = lc_element_by_index(lc: t, i);
1172	if (e->refcnt)
1173	drbd_err(device, "refcnt(%d)==%d\n",
1174	e->lc_number, e->refcnt);
1175	in_use += e->refcnt;
1176	}
1177	}
1178	if (!in_use)
1179	device->act_log = n;
1180	spin_unlock_irq(lock: &device->al_lock);
1181	if (in_use) {
1182	drbd_err(device, "Activity log still in use!\n");
1183	lc_destroy(lc: n);
1184	return -EBUSY;
1185	} else {
1186	lc_destroy(lc: t);
1187	}
1188	drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */
1189	return 0;
1190	}
1191
1192	static unsigned int drbd_max_peer_bio_size(struct drbd_device *device)
1193	{
1194	/*
1195	* We may ignore peer limits if the peer is modern enough. From 8.3.8
1196	* onwards the peer can use multiple BIOs for a single peer_request.
1197	*/
1198	if (device->state.conn < C_WF_REPORT_PARAMS)
1199	return device->peer_max_bio_size;
1200
1201	if (first_peer_device(device)->connection->agreed_pro_version < 94)
1202	return min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1203
1204	/*
1205	* Correct old drbd (up to 8.3.7) if it believes it can do more than
1206	* 32KiB.
1207	*/
1208	if (first_peer_device(device)->connection->agreed_pro_version == 94)
1209	return DRBD_MAX_SIZE_H80_PACKET;
1210
1211	/*
1212	* drbd 8.3.8 onwards, before 8.4.0
1213	*/
1214	if (first_peer_device(device)->connection->agreed_pro_version < 100)
1215	return DRBD_MAX_BIO_SIZE_P95;
1216	return DRBD_MAX_BIO_SIZE;
1217	}
1218
1219	static unsigned int drbd_max_discard_sectors(struct drbd_connection *connection)
1220	{
1221	/* when we introduced REQ_WRITE_SAME support, we also bumped
1222	* our maximum supported batch bio size used for discards. */
1223	if (connection->agreed_features & DRBD_FF_WSAME)
1224	return DRBD_MAX_BBIO_SECTORS;
1225	/* before, with DRBD <= 8.4.6, we only allowed up to one AL_EXTENT_SIZE. */
1226	return AL_EXTENT_SIZE >> 9;
1227	}
1228
1229	static bool drbd_discard_supported(struct drbd_connection *connection,
1230	struct drbd_backing_dev *bdev)
1231	{
1232	if (bdev && !bdev_max_discard_sectors(bdev: bdev->backing_bdev))
1233	return false;
1234
1235	if (connection->cstate >= C_CONNECTED &&
1236	!(connection->agreed_features & DRBD_FF_TRIM)) {
1237	drbd_info(connection,
1238	"peer DRBD too old, does not support TRIM: disabling discards\n");
1239	return false;
1240	}
1241
1242	return true;
1243	}
1244
1245	/* This is the workaround for "bio would need to, but cannot, be split" */
1246	static unsigned int drbd_backing_dev_max_segments(struct drbd_device *device)
1247	{
1248	unsigned int max_segments;
1249
1250	rcu_read_lock();
1251	max_segments = rcu_dereference(device->ldev->disk_conf)->max_bio_bvecs;
1252	rcu_read_unlock();
1253
1254	if (!max_segments)
1255	return BLK_MAX_SEGMENTS;
1256	return max_segments;
1257	}
1258
1259	void drbd_reconsider_queue_parameters(struct drbd_device *device,
1260	struct drbd_backing_dev *bdev, struct o_qlim *o)
1261	{
1262	struct drbd_connection *connection =
1263	first_peer_device(device)->connection;
1264	struct request_queue * const q = device->rq_queue;
1265	unsigned int now = queue_max_hw_sectors(q) << 9;
1266	struct queue_limits lim;
1267	struct request_queue *b = NULL;
1268	unsigned int new;
1269
1270	if (bdev) {
1271	b = bdev->backing_bdev->bd_disk->queue;
1272
1273	device->local_max_bio_size =
1274	queue_max_hw_sectors(q: b) << SECTOR_SHIFT;
1275	}
1276
1277	/*
1278	* We may later detach and re-attach on a disconnected Primary. Avoid
1279	* decreasing the value in this case.
1280	*
1281	* We want to store what we know the peer DRBD can handle, not what the
1282	* peer IO backend can handle.
1283	*/
1284	new = min3(DRBD_MAX_BIO_SIZE, device->local_max_bio_size,
1285	max(drbd_max_peer_bio_size(device), device->peer_max_bio_size));
1286	if (new != now) {
1287	if (device->state.role == R_PRIMARY && new < now)
1288	drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n",
1289	new, now);
1290	drbd_info(device, "max BIO size = %u\n", new);
1291	}
1292
1293	lim = queue_limits_start_update(q);
1294	if (bdev) {
1295	blk_set_stacking_limits(lim: &lim);
1296	lim.max_segments = drbd_backing_dev_max_segments(device);
1297	} else {
1298	lim.max_segments = BLK_MAX_SEGMENTS;
1299	}
1300
1301	lim.max_hw_sectors = new >> SECTOR_SHIFT;
1302	lim.seg_boundary_mask = PAGE_SIZE - 1;
1303
1304	/*
1305	* We don't care for the granularity, really.
1306	*
1307	* Stacking limits below should fix it for the local device. Whether or
1308	* not it is a suitable granularity on the remote device is not our
1309	* problem, really. If you care, you need to use devices with similar
1310	* topology on all peers.
1311	*/
1312	if (drbd_discard_supported(connection, bdev)) {
1313	lim.discard_granularity = 512;
1314	lim.max_hw_discard_sectors =
1315	drbd_max_discard_sectors(connection);
1316	} else {
1317	lim.discard_granularity = 0;
1318	lim.max_hw_discard_sectors = 0;
1319	}
1320
1321	if (bdev)
1322	blk_stack_limits(t: &lim, b: &b->limits, offset: 0);
1323
1324	/*
1325	* If we can handle "zeroes" efficiently on the protocol, we want to do
1326	* that, even if our backend does not announce max_write_zeroes_sectors
1327	* itself.
1328	*/
1329	if (connection->agreed_features & DRBD_FF_WZEROES)
1330	lim.max_write_zeroes_sectors = DRBD_MAX_BBIO_SECTORS;
1331	else
1332	lim.max_write_zeroes_sectors = 0;
1333
1334	if ((lim.discard_granularity >> SECTOR_SHIFT) >
1335	lim.max_hw_discard_sectors) {
1336	lim.discard_granularity = 0;
1337	lim.max_hw_discard_sectors = 0;
1338	}
1339
1340	if (queue_limits_commit_update(q, lim: &lim))
1341	drbd_err(device, "setting new queue limits failed\n");
1342	}
1343
1344	/* Starts the worker thread */
1345	static void conn_reconfig_start(struct drbd_connection *connection)
1346	{
1347	drbd_thread_start(thi: &connection->worker);
1348	drbd_flush_workqueue(work_queue: &connection->sender_work);
1349	}
1350
1351	/* if still unconfigured, stops worker again. */
1352	static void conn_reconfig_done(struct drbd_connection *connection)
1353	{
1354	bool stop_threads;
1355	spin_lock_irq(lock: &connection->resource->req_lock);
1356	stop_threads = conn_all_vols_unconf(connection) &&
1357	connection->cstate == C_STANDALONE;
1358	spin_unlock_irq(lock: &connection->resource->req_lock);
1359	if (stop_threads) {
1360	/* ack_receiver thread and ack_sender workqueue are implicitly
1361	* stopped by receiver in conn_disconnect() */
1362	drbd_thread_stop(thi: &connection->receiver);
1363	drbd_thread_stop(thi: &connection->worker);
1364	}
1365	}
1366
1367	/* Make sure IO is suspended before calling this function(). */
1368	static void drbd_suspend_al(struct drbd_device *device)
1369	{
1370	int s = 0;
1371
1372	if (!lc_try_lock(lc: device->act_log)) {
1373	drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1374	return;
1375	}
1376
1377	drbd_al_shrink(device);
1378	spin_lock_irq(lock: &device->resource->req_lock);
1379	if (device->state.conn < C_CONNECTED)
1380	s = !test_and_set_bit(nr: AL_SUSPENDED, addr: &device->flags);
1381	spin_unlock_irq(lock: &device->resource->req_lock);
1382	lc_unlock(lc: device->act_log);
1383
1384	if (s)
1385	drbd_info(device, "Suspended AL updates\n");
1386	}
1387
1388
1389	static bool should_set_defaults(struct genl_info *info)
1390	{
1391	struct drbd_genlmsghdr *dh = genl_info_userhdr(info);
1392
1393	return 0 != (dh->flags & DRBD_GENL_F_SET_DEFAULTS);
1394	}
1395
1396	static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1397	{
1398	/* This is limited by 16 bit "slot" numbers,
1399	* and by available on-disk context storage.
1400	*
1401	* Also (u16)~0 is special (denotes a "free" extent).
1402	*
1403	* One transaction occupies one 4kB on-disk block,
1404	* we have n such blocks in the on disk ring buffer,
1405	* the "current" transaction may fail (n-1),
1406	* and there is 919 slot numbers context information per transaction.
1407	*
1408	* 72 transaction blocks amounts to more than 2**16 context slots,
1409	* so cap there first.
1410	*/
1411	const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1412	const unsigned int sufficient_on_disk =
1413	(max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1414	/AL_CONTEXT_PER_TRANSACTION;
1415
1416	unsigned int al_size_4k = bdev->md.al_size_4k;
1417
1418	if (al_size_4k > sufficient_on_disk)
1419	return max_al_nr;
1420
1421	return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1422	}
1423
1424	static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1425	{
1426	return a->disk_barrier != b->disk_barrier ||
1427	a->disk_flushes != b->disk_flushes ||
1428	a->disk_drain != b->disk_drain;
1429	}
1430
1431	static void sanitize_disk_conf(struct drbd_device *device, struct disk_conf *disk_conf,
1432	struct drbd_backing_dev *nbc)
1433	{
1434	struct block_device *bdev = nbc->backing_bdev;
1435
1436	if (disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1437	disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1438	if (disk_conf->al_extents > drbd_al_extents_max(bdev: nbc))
1439	disk_conf->al_extents = drbd_al_extents_max(bdev: nbc);
1440
1441	if (!bdev_max_discard_sectors(bdev)) {
1442	if (disk_conf->rs_discard_granularity) {
1443	disk_conf->rs_discard_granularity = 0; /* disable feature */
1444	drbd_info(device, "rs_discard_granularity feature disabled\n");
1445	}
1446	}
1447
1448	if (disk_conf->rs_discard_granularity) {
1449	int orig_value = disk_conf->rs_discard_granularity;
1450	sector_t discard_size = bdev_max_discard_sectors(bdev) << 9;
1451	unsigned int discard_granularity = bdev_discard_granularity(bdev);
1452	int remainder;
1453
1454	if (discard_granularity > disk_conf->rs_discard_granularity)
1455	disk_conf->rs_discard_granularity = discard_granularity;
1456
1457	remainder = disk_conf->rs_discard_granularity %
1458	discard_granularity;
1459	disk_conf->rs_discard_granularity += remainder;
1460
1461	if (disk_conf->rs_discard_granularity > discard_size)
1462	disk_conf->rs_discard_granularity = discard_size;
1463
1464	if (disk_conf->rs_discard_granularity != orig_value)
1465	drbd_info(device, "rs_discard_granularity changed to %d\n",
1466	disk_conf->rs_discard_granularity);
1467	}
1468	}
1469
1470	static int disk_opts_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1471	{
1472	int err = -EBUSY;
1473
1474	if (device->act_log &&
1475	device->act_log->nr_elements == dc->al_extents)
1476	return 0;
1477
1478	drbd_suspend_io(device);
1479	/* If IO completion is currently blocked, we would likely wait
1480	* "forever" for the activity log to become unused. So we don't. */
1481	if (atomic_read(v: &device->ap_bio_cnt))
1482	goto out;
1483
1484	wait_event(device->al_wait, lc_try_lock(device->act_log));
1485	drbd_al_shrink(device);
1486	err = drbd_check_al_size(device, dc);
1487	lc_unlock(lc: device->act_log);
1488	wake_up(&device->al_wait);
1489	out:
1490	drbd_resume_io(device);
1491	return err;
1492	}
1493
1494	int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1495	{
1496	struct drbd_config_context adm_ctx;
1497	enum drbd_ret_code retcode;
1498	struct drbd_device *device;
1499	struct disk_conf *new_disk_conf, *old_disk_conf;
1500	struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1501	int err;
1502	unsigned int fifo_size;
1503
1504	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1505	if (!adm_ctx.reply_skb)
1506	return retcode;
1507	if (retcode != NO_ERROR)
1508	goto finish;
1509
1510	device = adm_ctx.device;
1511	mutex_lock(&adm_ctx.resource->adm_mutex);
1512
1513	/* we also need a disk
1514	* to change the options on */
1515	if (!get_ldev(device)) {
1516	retcode = ERR_NO_DISK;
1517	goto out;
1518	}
1519
1520	new_disk_conf = kmalloc(size: sizeof(struct disk_conf), GFP_KERNEL);
1521	if (!new_disk_conf) {
1522	retcode = ERR_NOMEM;
1523	goto fail;
1524	}
1525
1526	mutex_lock(&device->resource->conf_update);
1527	old_disk_conf = device->ldev->disk_conf;
1528	*new_disk_conf = *old_disk_conf;
1529	if (should_set_defaults(info))
1530	set_disk_conf_defaults(new_disk_conf);
1531
1532	err = disk_conf_from_attrs_for_change(s: new_disk_conf, info);
1533	if (err && err != -ENOMSG) {
1534	retcode = ERR_MANDATORY_TAG;
1535	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
1536	goto fail_unlock;
1537	}
1538
1539	if (!expect(device, new_disk_conf->resync_rate >= 1))
1540	new_disk_conf->resync_rate = 1;
1541
1542	sanitize_disk_conf(device, disk_conf: new_disk_conf, nbc: device->ldev);
1543
1544	if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1545	new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1546
1547	fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1548	if (fifo_size != device->rs_plan_s->size) {
1549	new_plan = fifo_alloc(fifo_size);
1550	if (!new_plan) {
1551	drbd_err(device, "kmalloc of fifo_buffer failed");
1552	retcode = ERR_NOMEM;
1553	goto fail_unlock;
1554	}
1555	}
1556
1557	err = disk_opts_check_al_size(device, dc: new_disk_conf);
1558	if (err) {
1559	/* Could be just "busy". Ignore?
1560	* Introduce dedicated error code? */
1561	drbd_msg_put_info(skb: adm_ctx.reply_skb,
1562	info: "Try again without changing current al-extents setting");
1563	retcode = ERR_NOMEM;
1564	goto fail_unlock;
1565	}
1566
1567	lock_all_resources();
1568	retcode = drbd_resync_after_valid(device, o_minor: new_disk_conf->resync_after);
1569	if (retcode == NO_ERROR) {
1570	rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1571	drbd_resync_after_changed(device);
1572	}
1573	unlock_all_resources();
1574
1575	if (retcode != NO_ERROR)
1576	goto fail_unlock;
1577
1578	if (new_plan) {
1579	old_plan = device->rs_plan_s;
1580	rcu_assign_pointer(device->rs_plan_s, new_plan);
1581	}
1582
1583	mutex_unlock(lock: &device->resource->conf_update);
1584
1585	if (new_disk_conf->al_updates)
1586	device->ldev->md.flags &= ~MDF_AL_DISABLED;
1587	else
1588	device->ldev->md.flags |= MDF_AL_DISABLED;
1589
1590	if (new_disk_conf->md_flushes)
1591	clear_bit(nr: MD_NO_FUA, addr: &device->flags);
1592	else
1593	set_bit(nr: MD_NO_FUA, addr: &device->flags);
1594
1595	if (write_ordering_changed(a: old_disk_conf, b: new_disk_conf))
1596	drbd_bump_write_ordering(resource: device->resource, NULL, wo: WO_BDEV_FLUSH);
1597
1598	if (old_disk_conf->discard_zeroes_if_aligned !=
1599	new_disk_conf->discard_zeroes_if_aligned)
1600	drbd_reconsider_queue_parameters(device, bdev: device->ldev, NULL);
1601
1602	drbd_md_sync(device);
1603
1604	if (device->state.conn >= C_CONNECTED) {
1605	struct drbd_peer_device *peer_device;
1606
1607	for_each_peer_device(peer_device, device)
1608	drbd_send_sync_param(peer_device);
1609	}
1610
1611	kvfree_rcu_mightsleep(old_disk_conf);
1612	kfree(objp: old_plan);
1613	mod_timer(timer: &device->request_timer, expires: jiffies + HZ);
1614	goto success;
1615
1616	fail_unlock:
1617	mutex_unlock(lock: &device->resource->conf_update);
1618	fail:
1619	kfree(objp: new_disk_conf);
1620	kfree(objp: new_plan);
1621	success:
1622	put_ldev(device);
1623	out:
1624	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
1625	finish:
1626	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
1627	return 0;
1628	}
1629
1630	static struct file *open_backing_dev(struct drbd_device *device,
1631	const char *bdev_path, void *claim_ptr, bool do_bd_link)
1632	{
1633	struct file *file;
1634	int err = 0;
1635
1636	file = bdev_file_open_by_path(path: bdev_path, BLK_OPEN_READ | BLK_OPEN_WRITE,
1637	holder: claim_ptr, NULL);
1638	if (IS_ERR(ptr: file)) {
1639	drbd_err(device, "open(\"%s\") failed with %ld\n",
1640	bdev_path, PTR_ERR(file));
1641	return file;
1642	}
1643
1644	if (!do_bd_link)
1645	return file;
1646
1647	err = bd_link_disk_holder(bdev: file_bdev(bdev_file: file), disk: device->vdisk);
1648	if (err) {
1649	fput(file);
1650	drbd_err(device, "bd_link_disk_holder(\"%s\", ...) failed with %d\n",
1651	bdev_path, err);
1652	file = ERR_PTR(error: err);
1653	}
1654	return file;
1655	}
1656
1657	static int open_backing_devices(struct drbd_device *device,
1658	struct disk_conf *new_disk_conf,
1659	struct drbd_backing_dev *nbc)
1660	{
1661	struct file *file;
1662
1663	file = open_backing_dev(device, bdev_path: new_disk_conf->backing_dev, claim_ptr: device,
1664	do_bd_link: true);
1665	if (IS_ERR(ptr: file))
1666	return ERR_OPEN_DISK;
1667	nbc->backing_bdev = file_bdev(bdev_file: file);
1668	nbc->backing_bdev_file = file;
1669
1670	/*
1671	* meta_dev_idx >= 0: external fixed size, possibly multiple
1672	* drbd sharing one meta device. TODO in that case, paranoia
1673	* check that [md_bdev, meta_dev_idx] is not yet used by some
1674	* other drbd minor! (if you use drbd.conf + drbdadm, that
1675	* should check it for you already; but if you don't, or
1676	* someone fooled it, we need to double check here)
1677	*/
1678	file = open_backing_dev(device, bdev_path: new_disk_conf->meta_dev,
1679	/* claim ptr: device, if claimed exclusively; shared drbd_m_holder,
1680	* if potentially shared with other drbd minors */
1681	claim_ptr: (new_disk_conf->meta_dev_idx < 0) ? (void*)device : (void*)drbd_m_holder,
1682	/* avoid double bd_claim_by_disk() for the same (source,target) tuple,
1683	* as would happen with internal metadata. */
1684	do_bd_link: (new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_FLEX_INT &&
1685	new_disk_conf->meta_dev_idx != DRBD_MD_INDEX_INTERNAL));
1686	if (IS_ERR(ptr: file))
1687	return ERR_OPEN_MD_DISK;
1688	nbc->md_bdev = file_bdev(bdev_file: file);
1689	nbc->f_md_bdev = file;
1690	return NO_ERROR;
1691	}
1692
1693	static void close_backing_dev(struct drbd_device *device,
1694	struct file *bdev_file, bool do_bd_unlink)
1695	{
1696	if (!bdev_file)
1697	return;
1698	if (do_bd_unlink)
1699	bd_unlink_disk_holder(bdev: file_bdev(bdev_file), disk: device->vdisk);
1700	fput(bdev_file);
1701	}
1702
1703	void drbd_backing_dev_free(struct drbd_device *device, struct drbd_backing_dev *ldev)
1704	{
1705	if (ldev == NULL)
1706	return;
1707
1708	close_backing_dev(device, bdev_file: ldev->f_md_bdev,
1709	do_bd_unlink: ldev->md_bdev != ldev->backing_bdev);
1710	close_backing_dev(device, bdev_file: ldev->backing_bdev_file, do_bd_unlink: true);
1711
1712	kfree(objp: ldev->disk_conf);
1713	kfree(objp: ldev);
1714	}
1715
1716	int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1717	{
1718	struct drbd_config_context adm_ctx;
1719	struct drbd_device *device;
1720	struct drbd_peer_device *peer_device;
1721	struct drbd_connection *connection;
1722	int err;
1723	enum drbd_ret_code retcode;
1724	enum determine_dev_size dd;
1725	sector_t max_possible_sectors;
1726	sector_t min_md_device_sectors;
1727	struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1728	struct disk_conf *new_disk_conf = NULL;
1729	struct lru_cache *resync_lru = NULL;
1730	struct fifo_buffer *new_plan = NULL;
1731	union drbd_state ns, os;
1732	enum drbd_state_rv rv;
1733	struct net_conf *nc;
1734
1735	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1736	if (!adm_ctx.reply_skb)
1737	return retcode;
1738	if (retcode != NO_ERROR)
1739	goto finish;
1740
1741	device = adm_ctx.device;
1742	mutex_lock(&adm_ctx.resource->adm_mutex);
1743	peer_device = first_peer_device(device);
1744	connection = peer_device->connection;
1745	conn_reconfig_start(connection);
1746
1747	/* if you want to reconfigure, please tear down first */
1748	if (device->state.disk > D_DISKLESS) {
1749	retcode = ERR_DISK_CONFIGURED;
1750	goto fail;
1751	}
1752	/* It may just now have detached because of IO error. Make sure
1753	* drbd_ldev_destroy is done already, we may end up here very fast,
1754	* e.g. if someone calls attach from the on-io-error handler,
1755	* to realize a "hot spare" feature (not that I'd recommend that) */
1756	wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1757
1758	/* make sure there is no leftover from previous force-detach attempts */
1759	clear_bit(nr: FORCE_DETACH, addr: &device->flags);
1760	clear_bit(nr: WAS_IO_ERROR, addr: &device->flags);
1761	clear_bit(nr: WAS_READ_ERROR, addr: &device->flags);
1762
1763	/* and no leftover from previously aborted resync or verify, either */
1764	device->rs_total = 0;
1765	device->rs_failed = 0;
1766	atomic_set(v: &device->rs_pending_cnt, i: 0);
1767
1768	/* allocation not in the IO path, drbdsetup context */
1769	nbc = kzalloc(size: sizeof(struct drbd_backing_dev), GFP_KERNEL);
1770	if (!nbc) {
1771	retcode = ERR_NOMEM;
1772	goto fail;
1773	}
1774	spin_lock_init(&nbc->md.uuid_lock);
1775
1776	new_disk_conf = kzalloc(size: sizeof(struct disk_conf), GFP_KERNEL);
1777	if (!new_disk_conf) {
1778	retcode = ERR_NOMEM;
1779	goto fail;
1780	}
1781	nbc->disk_conf = new_disk_conf;
1782
1783	set_disk_conf_defaults(new_disk_conf);
1784	err = disk_conf_from_attrs(s: new_disk_conf, info);
1785	if (err) {
1786	retcode = ERR_MANDATORY_TAG;
1787	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
1788	goto fail;
1789	}
1790
1791	if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1792	new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1793
1794	new_plan = fifo_alloc(fifo_size: (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1795	if (!new_plan) {
1796	retcode = ERR_NOMEM;
1797	goto fail;
1798	}
1799
1800	if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1801	retcode = ERR_MD_IDX_INVALID;
1802	goto fail;
1803	}
1804
1805	rcu_read_lock();
1806	nc = rcu_dereference(connection->net_conf);
1807	if (nc) {
1808	if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1809	rcu_read_unlock();
1810	retcode = ERR_STONITH_AND_PROT_A;
1811	goto fail;
1812	}
1813	}
1814	rcu_read_unlock();
1815
1816	retcode = open_backing_devices(device, new_disk_conf, nbc);
1817	if (retcode != NO_ERROR)
1818	goto fail;
1819
1820	if ((nbc->backing_bdev == nbc->md_bdev) !=
1821	(new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1822	new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1823	retcode = ERR_MD_IDX_INVALID;
1824	goto fail;
1825	}
1826
1827	resync_lru = lc_create(name: "resync", cache: drbd_bm_ext_cache,
1828	max_pending_changes: 1, e_count: 61, e_size: sizeof(struct bm_extent),
1829	offsetof(struct bm_extent, lce));
1830	if (!resync_lru) {
1831	retcode = ERR_NOMEM;
1832	goto fail;
1833	}
1834
1835	/* Read our meta data super block early.
1836	* This also sets other on-disk offsets. */
1837	retcode = drbd_md_read(device, bdev: nbc);
1838	if (retcode != NO_ERROR)
1839	goto fail;
1840
1841	sanitize_disk_conf(device, disk_conf: new_disk_conf, nbc);
1842
1843	if (drbd_get_max_capacity(bdev: nbc) < new_disk_conf->disk_size) {
1844	drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1845	(unsigned long long) drbd_get_max_capacity(nbc),
1846	(unsigned long long) new_disk_conf->disk_size);
1847	retcode = ERR_DISK_TOO_SMALL;
1848	goto fail;
1849	}
1850
1851	if (new_disk_conf->meta_dev_idx < 0) {
1852	max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1853	/* at least one MB, otherwise it does not make sense */
1854	min_md_device_sectors = (2<<10);
1855	} else {
1856	max_possible_sectors = DRBD_MAX_SECTORS;
1857	min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1858	}
1859
1860	if (drbd_get_capacity(bdev: nbc->md_bdev) < min_md_device_sectors) {
1861	retcode = ERR_MD_DISK_TOO_SMALL;
1862	drbd_warn(device, "refusing attach: md-device too small, "
1863	"at least %llu sectors needed for this meta-disk type\n",
1864	(unsigned long long) min_md_device_sectors);
1865	goto fail;
1866	}
1867
1868	/* Make sure the new disk is big enough
1869	* (we may currently be R_PRIMARY with no local disk...) */
1870	if (drbd_get_max_capacity(bdev: nbc) < get_capacity(disk: device->vdisk)) {
1871	retcode = ERR_DISK_TOO_SMALL;
1872	goto fail;
1873	}
1874
1875	nbc->known_size = drbd_get_capacity(bdev: nbc->backing_bdev);
1876
1877	if (nbc->known_size > max_possible_sectors) {
1878	drbd_warn(device, "==> truncating very big lower level device "
1879	"to currently maximum possible %llu sectors <==\n",
1880	(unsigned long long) max_possible_sectors);
1881	if (new_disk_conf->meta_dev_idx >= 0)
1882	drbd_warn(device, "==>> using internal or flexible "
1883	"meta data may help <<==\n");
1884	}
1885
1886	drbd_suspend_io(device);
1887	/* also wait for the last barrier ack. */
1888	/* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1889	* We need a way to either ignore barrier acks for barriers sent before a device
1890	* was attached, or a way to wait for all pending barrier acks to come in.
1891	* As barriers are counted per resource,
1892	* we'd need to suspend io on all devices of a resource.
1893	*/
1894	wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1895	/* and for any other previously queued work */
1896	drbd_flush_workqueue(work_queue: &connection->sender_work);
1897
1898	rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1899	retcode = (enum drbd_ret_code)rv;
1900	drbd_resume_io(device);
1901	if (rv < SS_SUCCESS)
1902	goto fail;
1903
1904	if (!get_ldev_if_state(device, D_ATTACHING))
1905	goto force_diskless;
1906
1907	if (!device->bitmap) {
1908	if (drbd_bm_init(device)) {
1909	retcode = ERR_NOMEM;
1910	goto force_diskless_dec;
1911	}
1912	}
1913
1914	if (device->state.pdsk != D_UP_TO_DATE && device->ed_uuid &&
1915	(device->state.role == R_PRIMARY || device->state.peer == R_PRIMARY) &&
1916	(device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1917	drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1918	(unsigned long long)device->ed_uuid);
1919	retcode = ERR_DATA_NOT_CURRENT;
1920	goto force_diskless_dec;
1921	}
1922
1923	/* Since we are diskless, fix the activity log first... */
1924	if (drbd_check_al_size(device, dc: new_disk_conf)) {
1925	retcode = ERR_NOMEM;
1926	goto force_diskless_dec;
1927	}
1928
1929	/* Prevent shrinking of consistent devices ! */
1930	{
1931	unsigned long long nsz = drbd_new_dev_size(device, bdev: nbc, u_size: nbc->disk_conf->disk_size, assume_peer_has_space: 0);
1932	unsigned long long eff = nbc->md.la_size_sect;
1933	if (drbd_md_test_flag(nbc, MDF_CONSISTENT) && nsz < eff) {
1934	if (nsz == nbc->disk_conf->disk_size) {
1935	drbd_warn(device, "truncating a consistent device during attach (%llu < %llu)\n", nsz, eff);
1936	} else {
1937	drbd_warn(device, "refusing to truncate a consistent device (%llu < %llu)\n", nsz, eff);
1938	drbd_msg_sprintf_info(skb: adm_ctx.reply_skb,
1939	fmt: "To-be-attached device has last effective > current size, and is consistent\n"
1940	"(%llu > %llu sectors). Refusing to attach.", eff, nsz);
1941	retcode = ERR_IMPLICIT_SHRINK;
1942	goto force_diskless_dec;
1943	}
1944	}
1945	}
1946
1947	lock_all_resources();
1948	retcode = drbd_resync_after_valid(device, o_minor: new_disk_conf->resync_after);
1949	if (retcode != NO_ERROR) {
1950	unlock_all_resources();
1951	goto force_diskless_dec;
1952	}
1953
1954	/* Reset the "barriers don't work" bits here, then force meta data to
1955	* be written, to ensure we determine if barriers are supported. */
1956	if (new_disk_conf->md_flushes)
1957	clear_bit(nr: MD_NO_FUA, addr: &device->flags);
1958	else
1959	set_bit(nr: MD_NO_FUA, addr: &device->flags);
1960
1961	/* Point of no return reached.
1962	* Devices and memory are no longer released by error cleanup below.
1963	* now device takes over responsibility, and the state engine should
1964	* clean it up somewhere. */
1965	D_ASSERT(device, device->ldev == NULL);
1966	device->ldev = nbc;
1967	device->resync = resync_lru;
1968	device->rs_plan_s = new_plan;
1969	nbc = NULL;
1970	resync_lru = NULL;
1971	new_disk_conf = NULL;
1972	new_plan = NULL;
1973
1974	drbd_resync_after_changed(device);
1975	drbd_bump_write_ordering(resource: device->resource, bdev: device->ldev, wo: WO_BDEV_FLUSH);
1976	unlock_all_resources();
1977
1978	if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
1979	set_bit(nr: CRASHED_PRIMARY, addr: &device->flags);
1980	else
1981	clear_bit(nr: CRASHED_PRIMARY, addr: &device->flags);
1982
1983	if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1984	!(device->state.role == R_PRIMARY && device->resource->susp_nod))
1985	set_bit(nr: CRASHED_PRIMARY, addr: &device->flags);
1986
1987	device->send_cnt = 0;
1988	device->recv_cnt = 0;
1989	device->read_cnt = 0;
1990	device->writ_cnt = 0;
1991
1992	drbd_reconsider_queue_parameters(device, bdev: device->ldev, NULL);
1993
1994	/* If I am currently not R_PRIMARY,
1995	* but meta data primary indicator is set,
1996	* I just now recover from a hard crash,
1997	* and have been R_PRIMARY before that crash.
1998	*
1999	* Now, if I had no connection before that crash
2000	* (have been degraded R_PRIMARY), chances are that
2001	* I won't find my peer now either.
2002	*
2003	* In that case, and _only_ in that case,
2004	* we use the degr-wfc-timeout instead of the default,
2005	* so we can automatically recover from a crash of a
2006	* degraded but active "cluster" after a certain timeout.
2007	*/
2008	clear_bit(nr: USE_DEGR_WFC_T, addr: &device->flags);
2009	if (device->state.role != R_PRIMARY &&
2010	drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
2011	!drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
2012	set_bit(nr: USE_DEGR_WFC_T, addr: &device->flags);
2013
2014	dd = drbd_determine_dev_size(device, flags: 0, NULL);
2015	if (dd <= DS_ERROR) {
2016	retcode = ERR_NOMEM_BITMAP;
2017	goto force_diskless_dec;
2018	} else if (dd == DS_GREW)
2019	set_bit(nr: RESYNC_AFTER_NEG, addr: &device->flags);
2020
2021	if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
2022	(test_bit(CRASHED_PRIMARY, &device->flags) &&
2023	drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
2024	drbd_info(device, "Assuming that all blocks are out of sync "
2025	"(aka FullSync)\n");
2026	if (drbd_bitmap_io(device, io_fn: &drbd_bmio_set_n_write,
2027	why: "set_n_write from attaching", flags: BM_LOCKED_MASK,
2028	NULL)) {
2029	retcode = ERR_IO_MD_DISK;
2030	goto force_diskless_dec;
2031	}
2032	} else {
2033	if (drbd_bitmap_io(device, io_fn: &drbd_bm_read,
2034	why: "read from attaching", flags: BM_LOCKED_MASK,
2035	NULL)) {
2036	retcode = ERR_IO_MD_DISK;
2037	goto force_diskless_dec;
2038	}
2039	}
2040
2041	if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
2042	drbd_suspend_al(device); /* IO is still suspended here... */
2043
2044	spin_lock_irq(lock: &device->resource->req_lock);
2045	os = drbd_read_state(device);
2046	ns = os;
2047	/* If MDF_CONSISTENT is not set go into inconsistent state,
2048	otherwise investigate MDF_WasUpToDate...
2049	If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
2050	otherwise into D_CONSISTENT state.
2051	*/
2052	if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
2053	if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
2054	ns.disk = D_CONSISTENT;
2055	else
2056	ns.disk = D_OUTDATED;
2057	} else {
2058	ns.disk = D_INCONSISTENT;
2059	}
2060
2061	if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
2062	ns.pdsk = D_OUTDATED;
2063
2064	rcu_read_lock();
2065	if (ns.disk == D_CONSISTENT &&
2066	(ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
2067	ns.disk = D_UP_TO_DATE;
2068
2069	/* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
2070	MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
2071	this point, because drbd_request_state() modifies these
2072	flags. */
2073
2074	if (rcu_dereference(device->ldev->disk_conf)->al_updates)
2075	device->ldev->md.flags &= ~MDF_AL_DISABLED;
2076	else
2077	device->ldev->md.flags |= MDF_AL_DISABLED;
2078
2079	rcu_read_unlock();
2080
2081	/* In case we are C_CONNECTED postpone any decision on the new disk
2082	state after the negotiation phase. */
2083	if (device->state.conn == C_CONNECTED) {
2084	device->new_state_tmp.i = ns.i;
2085	ns.i = os.i;
2086	ns.disk = D_NEGOTIATING;
2087
2088	/* We expect to receive up-to-date UUIDs soon.
2089	To avoid a race in receive_state, free p_uuid while
2090	holding req_lock. I.e. atomic with the state change */
2091	kfree(objp: device->p_uuid);
2092	device->p_uuid = NULL;
2093	}
2094
2095	rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
2096	spin_unlock_irq(lock: &device->resource->req_lock);
2097
2098	if (rv < SS_SUCCESS)
2099	goto force_diskless_dec;
2100
2101	mod_timer(timer: &device->request_timer, expires: jiffies + HZ);
2102
2103	if (device->state.role == R_PRIMARY)
2104	device->ldev->md.uuid[UI_CURRENT] |= (u64)1;
2105	else
2106	device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
2107
2108	drbd_md_mark_dirty(device);
2109	drbd_md_sync(device);
2110
2111	kobject_uevent(kobj: &disk_to_dev(device->vdisk)->kobj, action: KOBJ_CHANGE);
2112	put_ldev(device);
2113	conn_reconfig_done(connection);
2114	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2115	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
2116	return 0;
2117
2118	force_diskless_dec:
2119	put_ldev(device);
2120	force_diskless:
2121	drbd_force_state(device, NS(disk, D_DISKLESS));
2122	drbd_md_sync(device);
2123	fail:
2124	conn_reconfig_done(connection);
2125	if (nbc) {
2126	close_backing_dev(device, bdev_file: nbc->f_md_bdev,
2127	do_bd_unlink: nbc->md_bdev != nbc->backing_bdev);
2128	close_backing_dev(device, bdev_file: nbc->backing_bdev_file, do_bd_unlink: true);
2129	kfree(objp: nbc);
2130	}
2131	kfree(objp: new_disk_conf);
2132	lc_destroy(lc: resync_lru);
2133	kfree(objp: new_plan);
2134	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2135	finish:
2136	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
2137	return 0;
2138	}
2139
2140	static int adm_detach(struct drbd_device *device, int force)
2141	{
2142	if (force) {
2143	set_bit(nr: FORCE_DETACH, addr: &device->flags);
2144	drbd_force_state(device, NS(disk, D_FAILED));
2145	return SS_SUCCESS;
2146	}
2147
2148	return drbd_request_detach_interruptible(device);
2149	}
2150
2151	/* Detaching the disk is a process in multiple stages. First we need to lock
2152	* out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
2153	* Then we transition to D_DISKLESS, and wait for put_ldev() to return all
2154	* internal references as well.
2155	* Only then we have finally detached. */
2156	int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
2157	{
2158	struct drbd_config_context adm_ctx;
2159	enum drbd_ret_code retcode;
2160	struct detach_parms parms = { };
2161	int err;
2162
2163	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2164	if (!adm_ctx.reply_skb)
2165	return retcode;
2166	if (retcode != NO_ERROR)
2167	goto out;
2168
2169	if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
2170	err = detach_parms_from_attrs(s: &parms, info);
2171	if (err) {
2172	retcode = ERR_MANDATORY_TAG;
2173	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
2174	goto out;
2175	}
2176	}
2177
2178	mutex_lock(&adm_ctx.resource->adm_mutex);
2179	retcode = adm_detach(device: adm_ctx.device, force: parms.force_detach);
2180	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2181	out:
2182	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
2183	return 0;
2184	}
2185
2186	static bool conn_resync_running(struct drbd_connection *connection)
2187	{
2188	struct drbd_peer_device *peer_device;
2189	bool rv = false;
2190	int vnr;
2191
2192	rcu_read_lock();
2193	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2194	struct drbd_device *device = peer_device->device;
2195	if (device->state.conn == C_SYNC_SOURCE ||
2196	device->state.conn == C_SYNC_TARGET ||
2197	device->state.conn == C_PAUSED_SYNC_S ||
2198	device->state.conn == C_PAUSED_SYNC_T) {
2199	rv = true;
2200	break;
2201	}
2202	}
2203	rcu_read_unlock();
2204
2205	return rv;
2206	}
2207
2208	static bool conn_ov_running(struct drbd_connection *connection)
2209	{
2210	struct drbd_peer_device *peer_device;
2211	bool rv = false;
2212	int vnr;
2213
2214	rcu_read_lock();
2215	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2216	struct drbd_device *device = peer_device->device;
2217	if (device->state.conn == C_VERIFY_S ||
2218	device->state.conn == C_VERIFY_T) {
2219	rv = true;
2220	break;
2221	}
2222	}
2223	rcu_read_unlock();
2224
2225	return rv;
2226	}
2227
2228	static enum drbd_ret_code
2229	_check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2230	{
2231	struct drbd_peer_device *peer_device;
2232	int i;
2233
2234	if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2235	if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2236	return ERR_NEED_APV_100;
2237
2238	if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2239	return ERR_NEED_APV_100;
2240
2241	if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2242	return ERR_NEED_APV_100;
2243	}
2244
2245	if (!new_net_conf->two_primaries &&
2246	conn_highest_role(connection) == R_PRIMARY &&
2247	conn_highest_peer(connection) == R_PRIMARY)
2248	return ERR_NEED_ALLOW_TWO_PRI;
2249
2250	if (new_net_conf->two_primaries &&
2251	(new_net_conf->wire_protocol != DRBD_PROT_C))
2252	return ERR_NOT_PROTO_C;
2253
2254	idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2255	struct drbd_device *device = peer_device->device;
2256	if (get_ldev(device)) {
2257	enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2258	put_ldev(device);
2259	if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2260	return ERR_STONITH_AND_PROT_A;
2261	}
2262	if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2263	return ERR_DISCARD_IMPOSSIBLE;
2264	}
2265
2266	if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2267	return ERR_CONG_NOT_PROTO_A;
2268
2269	return NO_ERROR;
2270	}
2271
2272	static enum drbd_ret_code
2273	check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2274	{
2275	enum drbd_ret_code rv;
2276	struct drbd_peer_device *peer_device;
2277	int i;
2278
2279	rcu_read_lock();
2280	rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2281	rcu_read_unlock();
2282
2283	/* connection->peer_devices protected by genl_lock() here */
2284	idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2285	struct drbd_device *device = peer_device->device;
2286	if (!device->bitmap) {
2287	if (drbd_bm_init(device))
2288	return ERR_NOMEM;
2289	}
2290	}
2291
2292	return rv;
2293	}
2294
2295	struct crypto {
2296	struct crypto_shash *verify_tfm;
2297	struct crypto_shash *csums_tfm;
2298	struct crypto_shash *cram_hmac_tfm;
2299	struct crypto_shash *integrity_tfm;
2300	};
2301
2302	static int
2303	alloc_shash(struct crypto_shash **tfm, char *tfm_name, int err_alg)
2304	{
2305	if (!tfm_name[0])
2306	return NO_ERROR;
2307
2308	*tfm = crypto_alloc_shash(alg_name: tfm_name, type: 0, mask: 0);
2309	if (IS_ERR(ptr: *tfm)) {
2310	*tfm = NULL;
2311	return err_alg;
2312	}
2313
2314	return NO_ERROR;
2315	}
2316
2317	static enum drbd_ret_code
2318	alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2319	{
2320	char hmac_name[CRYPTO_MAX_ALG_NAME];
2321	enum drbd_ret_code rv;
2322
2323	rv = alloc_shash(tfm: &crypto->csums_tfm, tfm_name: new_net_conf->csums_alg,
2324	err_alg: ERR_CSUMS_ALG);
2325	if (rv != NO_ERROR)
2326	return rv;
2327	rv = alloc_shash(tfm: &crypto->verify_tfm, tfm_name: new_net_conf->verify_alg,
2328	err_alg: ERR_VERIFY_ALG);
2329	if (rv != NO_ERROR)
2330	return rv;
2331	rv = alloc_shash(tfm: &crypto->integrity_tfm, tfm_name: new_net_conf->integrity_alg,
2332	err_alg: ERR_INTEGRITY_ALG);
2333	if (rv != NO_ERROR)
2334	return rv;
2335	if (new_net_conf->cram_hmac_alg[0] != 0) {
2336	snprintf(buf: hmac_name, CRYPTO_MAX_ALG_NAME, fmt: "hmac(%s)",
2337	new_net_conf->cram_hmac_alg);
2338
2339	rv = alloc_shash(tfm: &crypto->cram_hmac_tfm, tfm_name: hmac_name,
2340	err_alg: ERR_AUTH_ALG);
2341	}
2342
2343	return rv;
2344	}
2345
2346	static void free_crypto(struct crypto *crypto)
2347	{
2348	crypto_free_shash(tfm: crypto->cram_hmac_tfm);
2349	crypto_free_shash(tfm: crypto->integrity_tfm);
2350	crypto_free_shash(tfm: crypto->csums_tfm);
2351	crypto_free_shash(tfm: crypto->verify_tfm);
2352	}
2353
2354	int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2355	{
2356	struct drbd_config_context adm_ctx;
2357	enum drbd_ret_code retcode;
2358	struct drbd_connection *connection;
2359	struct net_conf *old_net_conf, *new_net_conf = NULL;
2360	int err;
2361	int ovr; /* online verify running */
2362	int rsr; /* re-sync running */
2363	struct crypto crypto = { };
2364
2365	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2366	if (!adm_ctx.reply_skb)
2367	return retcode;
2368	if (retcode != NO_ERROR)
2369	goto finish;
2370
2371	connection = adm_ctx.connection;
2372	mutex_lock(&adm_ctx.resource->adm_mutex);
2373
2374	new_net_conf = kzalloc(size: sizeof(struct net_conf), GFP_KERNEL);
2375	if (!new_net_conf) {
2376	retcode = ERR_NOMEM;
2377	goto out;
2378	}
2379
2380	conn_reconfig_start(connection);
2381
2382	mutex_lock(&connection->data.mutex);
2383	mutex_lock(&connection->resource->conf_update);
2384	old_net_conf = connection->net_conf;
2385
2386	if (!old_net_conf) {
2387	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: "net conf missing, try connect");
2388	retcode = ERR_INVALID_REQUEST;
2389	goto fail;
2390	}
2391
2392	*new_net_conf = *old_net_conf;
2393	if (should_set_defaults(info))
2394	set_net_conf_defaults(new_net_conf);
2395
2396	err = net_conf_from_attrs_for_change(s: new_net_conf, info);
2397	if (err && err != -ENOMSG) {
2398	retcode = ERR_MANDATORY_TAG;
2399	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
2400	goto fail;
2401	}
2402
2403	retcode = check_net_options(connection, new_net_conf);
2404	if (retcode != NO_ERROR)
2405	goto fail;
2406
2407	/* re-sync running */
2408	rsr = conn_resync_running(connection);
2409	if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2410	retcode = ERR_CSUMS_RESYNC_RUNNING;
2411	goto fail;
2412	}
2413
2414	/* online verify running */
2415	ovr = conn_ov_running(connection);
2416	if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2417	retcode = ERR_VERIFY_RUNNING;
2418	goto fail;
2419	}
2420
2421	retcode = alloc_crypto(crypto: &crypto, new_net_conf);
2422	if (retcode != NO_ERROR)
2423	goto fail;
2424
2425	rcu_assign_pointer(connection->net_conf, new_net_conf);
2426
2427	if (!rsr) {
2428	crypto_free_shash(tfm: connection->csums_tfm);
2429	connection->csums_tfm = crypto.csums_tfm;
2430	crypto.csums_tfm = NULL;
2431	}
2432	if (!ovr) {
2433	crypto_free_shash(tfm: connection->verify_tfm);
2434	connection->verify_tfm = crypto.verify_tfm;
2435	crypto.verify_tfm = NULL;
2436	}
2437
2438	crypto_free_shash(tfm: connection->integrity_tfm);
2439	connection->integrity_tfm = crypto.integrity_tfm;
2440	if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2441	/* Do this without trying to take connection->data.mutex again. */
2442	__drbd_send_protocol(connection, cmd: P_PROTOCOL_UPDATE);
2443
2444	crypto_free_shash(tfm: connection->cram_hmac_tfm);
2445	connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2446
2447	mutex_unlock(lock: &connection->resource->conf_update);
2448	mutex_unlock(lock: &connection->data.mutex);
2449	kvfree_rcu_mightsleep(old_net_conf);
2450
2451	if (connection->cstate >= C_WF_REPORT_PARAMS) {
2452	struct drbd_peer_device *peer_device;
2453	int vnr;
2454
2455	idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2456	drbd_send_sync_param(peer_device);
2457	}
2458
2459	goto done;
2460
2461	fail:
2462	mutex_unlock(lock: &connection->resource->conf_update);
2463	mutex_unlock(lock: &connection->data.mutex);
2464	free_crypto(crypto: &crypto);
2465	kfree(objp: new_net_conf);
2466	done:
2467	conn_reconfig_done(connection);
2468	out:
2469	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2470	finish:
2471	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
2472	return 0;
2473	}
2474
2475	static void connection_to_info(struct connection_info *info,
2476	struct drbd_connection *connection)
2477	{
2478	info->conn_connection_state = connection->cstate;
2479	info->conn_role = conn_highest_peer(connection);
2480	}
2481
2482	static void peer_device_to_info(struct peer_device_info *info,
2483	struct drbd_peer_device *peer_device)
2484	{
2485	struct drbd_device *device = peer_device->device;
2486
2487	info->peer_repl_state =
2488	max_t(enum drbd_conns, C_WF_REPORT_PARAMS, device->state.conn);
2489	info->peer_disk_state = device->state.pdsk;
2490	info->peer_resync_susp_user = device->state.user_isp;
2491	info->peer_resync_susp_peer = device->state.peer_isp;
2492	info->peer_resync_susp_dependency = device->state.aftr_isp;
2493	}
2494
2495	int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2496	{
2497	struct connection_info connection_info;
2498	enum drbd_notification_type flags;
2499	unsigned int peer_devices = 0;
2500	struct drbd_config_context adm_ctx;
2501	struct drbd_peer_device *peer_device;
2502	struct net_conf *old_net_conf, *new_net_conf = NULL;
2503	struct crypto crypto = { };
2504	struct drbd_resource *resource;
2505	struct drbd_connection *connection;
2506	enum drbd_ret_code retcode;
2507	enum drbd_state_rv rv;
2508	int i;
2509	int err;
2510
2511	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2512
2513	if (!adm_ctx.reply_skb)
2514	return retcode;
2515	if (retcode != NO_ERROR)
2516	goto out;
2517	if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2518	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: "connection endpoint(s) missing");
2519	retcode = ERR_INVALID_REQUEST;
2520	goto out;
2521	}
2522
2523	/* No need for _rcu here. All reconfiguration is
2524	* strictly serialized on genl_lock(). We are protected against
2525	* concurrent reconfiguration/addition/deletion */
2526	for_each_resource(resource, &drbd_resources) {
2527	for_each_connection(connection, resource) {
2528	if (nla_len(nla: adm_ctx.my_addr) == connection->my_addr_len &&
2529	!memcmp(p: nla_data(nla: adm_ctx.my_addr), q: &connection->my_addr,
2530	size: connection->my_addr_len)) {
2531	retcode = ERR_LOCAL_ADDR;
2532	goto out;
2533	}
2534
2535	if (nla_len(nla: adm_ctx.peer_addr) == connection->peer_addr_len &&
2536	!memcmp(p: nla_data(nla: adm_ctx.peer_addr), q: &connection->peer_addr,
2537	size: connection->peer_addr_len)) {
2538	retcode = ERR_PEER_ADDR;
2539	goto out;
2540	}
2541	}
2542	}
2543
2544	mutex_lock(&adm_ctx.resource->adm_mutex);
2545	connection = first_connection(resource: adm_ctx.resource);
2546	conn_reconfig_start(connection);
2547
2548	if (connection->cstate > C_STANDALONE) {
2549	retcode = ERR_NET_CONFIGURED;
2550	goto fail;
2551	}
2552
2553	/* allocation not in the IO path, drbdsetup / netlink process context */
2554	new_net_conf = kzalloc(size: sizeof(*new_net_conf), GFP_KERNEL);
2555	if (!new_net_conf) {
2556	retcode = ERR_NOMEM;
2557	goto fail;
2558	}
2559
2560	set_net_conf_defaults(new_net_conf);
2561
2562	err = net_conf_from_attrs(s: new_net_conf, info);
2563	if (err && err != -ENOMSG) {
2564	retcode = ERR_MANDATORY_TAG;
2565	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
2566	goto fail;
2567	}
2568
2569	retcode = check_net_options(connection, new_net_conf);
2570	if (retcode != NO_ERROR)
2571	goto fail;
2572
2573	retcode = alloc_crypto(crypto: &crypto, new_net_conf);
2574	if (retcode != NO_ERROR)
2575	goto fail;
2576
2577	((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2578
2579	drbd_flush_workqueue(work_queue: &connection->sender_work);
2580
2581	mutex_lock(&adm_ctx.resource->conf_update);
2582	old_net_conf = connection->net_conf;
2583	if (old_net_conf) {
2584	retcode = ERR_NET_CONFIGURED;
2585	mutex_unlock(lock: &adm_ctx.resource->conf_update);
2586	goto fail;
2587	}
2588	rcu_assign_pointer(connection->net_conf, new_net_conf);
2589
2590	conn_free_crypto(connection);
2591	connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2592	connection->integrity_tfm = crypto.integrity_tfm;
2593	connection->csums_tfm = crypto.csums_tfm;
2594	connection->verify_tfm = crypto.verify_tfm;
2595
2596	connection->my_addr_len = nla_len(nla: adm_ctx.my_addr);
2597	memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2598	connection->peer_addr_len = nla_len(nla: adm_ctx.peer_addr);
2599	memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2600
2601	idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2602	peer_devices++;
2603	}
2604
2605	connection_to_info(info: &connection_info, connection);
2606	flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2607	mutex_lock(&notification_mutex);
2608	notify_connection_state(NULL, 0, connection, &connection_info, NOTIFY_CREATE | flags);
2609	idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2610	struct peer_device_info peer_device_info;
2611
2612	peer_device_to_info(info: &peer_device_info, peer_device);
2613	flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
2614	notify_peer_device_state(NULL, 0, peer_device, &peer_device_info, NOTIFY_CREATE | flags);
2615	}
2616	mutex_unlock(lock: &notification_mutex);
2617	mutex_unlock(lock: &adm_ctx.resource->conf_update);
2618
2619	rcu_read_lock();
2620	idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2621	struct drbd_device *device = peer_device->device;
2622	device->send_cnt = 0;
2623	device->recv_cnt = 0;
2624	}
2625	rcu_read_unlock();
2626
2627	rv = conn_request_state(connection, NS(conn, C_UNCONNECTED), flags: CS_VERBOSE);
2628
2629	conn_reconfig_done(connection);
2630	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2631	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode: rv);
2632	return 0;
2633
2634	fail:
2635	free_crypto(crypto: &crypto);
2636	kfree(objp: new_net_conf);
2637
2638	conn_reconfig_done(connection);
2639	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2640	out:
2641	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
2642	return 0;
2643	}
2644
2645	static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2646	{
2647	enum drbd_conns cstate;
2648	enum drbd_state_rv rv;
2649
2650	repeat:
2651	rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2652	flags: force ? CS_HARD : 0);
2653
2654	switch (rv) {
2655	case SS_NOTHING_TO_DO:
2656	break;
2657	case SS_ALREADY_STANDALONE:
2658	return SS_SUCCESS;
2659	case SS_PRIMARY_NOP:
2660	/* Our state checking code wants to see the peer outdated. */
2661	rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), flags: 0);
2662
2663	if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2664	rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), flags: CS_VERBOSE);
2665
2666	break;
2667	case SS_CW_FAILED_BY_PEER:
2668	spin_lock_irq(lock: &connection->resource->req_lock);
2669	cstate = connection->cstate;
2670	spin_unlock_irq(lock: &connection->resource->req_lock);
2671	if (cstate <= C_WF_CONNECTION)
2672	goto repeat;
2673	/* The peer probably wants to see us outdated. */
2674	rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2675	disk, D_OUTDATED), flags: 0);
2676	if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2677	rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2678	flags: CS_HARD);
2679	}
2680	break;
2681	default:;
2682	/* no special handling necessary */
2683	}
2684
2685	if (rv >= SS_SUCCESS) {
2686	enum drbd_state_rv rv2;
2687	/* No one else can reconfigure the network while I am here.
2688	* The state handling only uses drbd_thread_stop_nowait(),
2689	* we want to really wait here until the receiver is no more.
2690	*/
2691	drbd_thread_stop(thi: &connection->receiver);
2692
2693	/* Race breaker. This additional state change request may be
2694	* necessary, if this was a forced disconnect during a receiver
2695	* restart. We may have "killed" the receiver thread just
2696	* after drbd_receiver() returned. Typically, we should be
2697	* C_STANDALONE already, now, and this becomes a no-op.
2698	*/
2699	rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2700	flags: CS_VERBOSE | CS_HARD);
2701	if (rv2 < SS_SUCCESS)
2702	drbd_err(connection,
2703	"unexpected rv2=%d in conn_try_disconnect()\n",
2704	rv2);
2705	/* Unlike in DRBD 9, the state engine has generated
2706	* NOTIFY_DESTROY events before clearing connection->net_conf. */
2707	}
2708	return rv;
2709	}
2710
2711	int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2712	{
2713	struct drbd_config_context adm_ctx;
2714	struct disconnect_parms parms;
2715	struct drbd_connection *connection;
2716	enum drbd_state_rv rv;
2717	enum drbd_ret_code retcode;
2718	int err;
2719
2720	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2721	if (!adm_ctx.reply_skb)
2722	return retcode;
2723	if (retcode != NO_ERROR)
2724	goto fail;
2725
2726	connection = adm_ctx.connection;
2727	memset(&parms, 0, sizeof(parms));
2728	if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2729	err = disconnect_parms_from_attrs(s: &parms, info);
2730	if (err) {
2731	retcode = ERR_MANDATORY_TAG;
2732	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
2733	goto fail;
2734	}
2735	}
2736
2737	mutex_lock(&adm_ctx.resource->adm_mutex);
2738	rv = conn_try_disconnect(connection, force: parms.force_disconnect);
2739	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2740	if (rv < SS_SUCCESS) {
2741	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode: rv);
2742	return 0;
2743	}
2744	retcode = NO_ERROR;
2745	fail:
2746	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
2747	return 0;
2748	}
2749
2750	void resync_after_online_grow(struct drbd_device *device)
2751	{
2752	int iass; /* I am sync source */
2753
2754	drbd_info(device, "Resync of new storage after online grow\n");
2755	if (device->state.role != device->state.peer)
2756	iass = (device->state.role == R_PRIMARY);
2757	else
2758	iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2759
2760	if (iass)
2761	drbd_start_resync(device, side: C_SYNC_SOURCE);
2762	else
2763	_drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2764	}
2765
2766	int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2767	{
2768	struct drbd_config_context adm_ctx;
2769	struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2770	struct resize_parms rs;
2771	struct drbd_device *device;
2772	enum drbd_ret_code retcode;
2773	enum determine_dev_size dd;
2774	bool change_al_layout = false;
2775	enum dds_flags ddsf;
2776	sector_t u_size;
2777	int err;
2778
2779	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2780	if (!adm_ctx.reply_skb)
2781	return retcode;
2782	if (retcode != NO_ERROR)
2783	goto finish;
2784
2785	mutex_lock(&adm_ctx.resource->adm_mutex);
2786	device = adm_ctx.device;
2787	if (!get_ldev(device)) {
2788	retcode = ERR_NO_DISK;
2789	goto fail;
2790	}
2791
2792	memset(&rs, 0, sizeof(struct resize_parms));
2793	rs.al_stripes = device->ldev->md.al_stripes;
2794	rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2795	if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2796	err = resize_parms_from_attrs(s: &rs, info);
2797	if (err) {
2798	retcode = ERR_MANDATORY_TAG;
2799	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
2800	goto fail_ldev;
2801	}
2802	}
2803
2804	if (device->state.conn > C_CONNECTED) {
2805	retcode = ERR_RESIZE_RESYNC;
2806	goto fail_ldev;
2807	}
2808
2809	if (device->state.role == R_SECONDARY &&
2810	device->state.peer == R_SECONDARY) {
2811	retcode = ERR_NO_PRIMARY;
2812	goto fail_ldev;
2813	}
2814
2815	if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2816	retcode = ERR_NEED_APV_93;
2817	goto fail_ldev;
2818	}
2819
2820	rcu_read_lock();
2821	u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2822	rcu_read_unlock();
2823	if (u_size != (sector_t)rs.resize_size) {
2824	new_disk_conf = kmalloc(size: sizeof(struct disk_conf), GFP_KERNEL);
2825	if (!new_disk_conf) {
2826	retcode = ERR_NOMEM;
2827	goto fail_ldev;
2828	}
2829	}
2830
2831	if (device->ldev->md.al_stripes != rs.al_stripes ||
2832	device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2833	u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2834
2835	if (al_size_k > (16 * 1024 * 1024)) {
2836	retcode = ERR_MD_LAYOUT_TOO_BIG;
2837	goto fail_ldev;
2838	}
2839
2840	if (al_size_k < MD_32kB_SECT/2) {
2841	retcode = ERR_MD_LAYOUT_TOO_SMALL;
2842	goto fail_ldev;
2843	}
2844
2845	if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2846	retcode = ERR_MD_LAYOUT_CONNECTED;
2847	goto fail_ldev;
2848	}
2849
2850	change_al_layout = true;
2851	}
2852
2853	if (device->ldev->known_size != drbd_get_capacity(bdev: device->ldev->backing_bdev))
2854	device->ldev->known_size = drbd_get_capacity(bdev: device->ldev->backing_bdev);
2855
2856	if (new_disk_conf) {
2857	mutex_lock(&device->resource->conf_update);
2858	old_disk_conf = device->ldev->disk_conf;
2859	*new_disk_conf = *old_disk_conf;
2860	new_disk_conf->disk_size = (sector_t)rs.resize_size;
2861	rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2862	mutex_unlock(lock: &device->resource->conf_update);
2863	kvfree_rcu_mightsleep(old_disk_conf);
2864	new_disk_conf = NULL;
2865	}
2866
2867	ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2868	dd = drbd_determine_dev_size(device, flags: ddsf, rs: change_al_layout ? &rs : NULL);
2869	drbd_md_sync(device);
2870	put_ldev(device);
2871	if (dd == DS_ERROR) {
2872	retcode = ERR_NOMEM_BITMAP;
2873	goto fail;
2874	} else if (dd == DS_ERROR_SPACE_MD) {
2875	retcode = ERR_MD_LAYOUT_NO_FIT;
2876	goto fail;
2877	} else if (dd == DS_ERROR_SHRINK) {
2878	retcode = ERR_IMPLICIT_SHRINK;
2879	goto fail;
2880	}
2881
2882	if (device->state.conn == C_CONNECTED) {
2883	if (dd == DS_GREW)
2884	set_bit(nr: RESIZE_PENDING, addr: &device->flags);
2885
2886	drbd_send_uuids(first_peer_device(device));
2887	drbd_send_sizes(first_peer_device(device), trigger_reply: 1, flags: ddsf);
2888	}
2889
2890	fail:
2891	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2892	finish:
2893	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
2894	return 0;
2895
2896	fail_ldev:
2897	put_ldev(device);
2898	kfree(objp: new_disk_conf);
2899	goto fail;
2900	}
2901
2902	int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2903	{
2904	struct drbd_config_context adm_ctx;
2905	enum drbd_ret_code retcode;
2906	struct res_opts res_opts;
2907	int err;
2908
2909	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2910	if (!adm_ctx.reply_skb)
2911	return retcode;
2912	if (retcode != NO_ERROR)
2913	goto fail;
2914
2915	res_opts = adm_ctx.resource->res_opts;
2916	if (should_set_defaults(info))
2917	set_res_opts_defaults(&res_opts);
2918
2919	err = res_opts_from_attrs(s: &res_opts, info);
2920	if (err && err != -ENOMSG) {
2921	retcode = ERR_MANDATORY_TAG;
2922	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
2923	goto fail;
2924	}
2925
2926	mutex_lock(&adm_ctx.resource->adm_mutex);
2927	err = set_resource_options(resource: adm_ctx.resource, res_opts: &res_opts);
2928	if (err) {
2929	retcode = ERR_INVALID_REQUEST;
2930	if (err == -ENOMEM)
2931	retcode = ERR_NOMEM;
2932	}
2933	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2934
2935	fail:
2936	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
2937	return 0;
2938	}
2939
2940	int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2941	{
2942	struct drbd_config_context adm_ctx;
2943	struct drbd_device *device;
2944	int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2945
2946	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2947	if (!adm_ctx.reply_skb)
2948	return retcode;
2949	if (retcode != NO_ERROR)
2950	goto out;
2951
2952	device = adm_ctx.device;
2953	if (!get_ldev(device)) {
2954	retcode = ERR_NO_DISK;
2955	goto out;
2956	}
2957
2958	mutex_lock(&adm_ctx.resource->adm_mutex);
2959
2960	/* If there is still bitmap IO pending, probably because of a previous
2961	* resync just being finished, wait for it before requesting a new resync.
2962	* Also wait for it's after_state_ch(). */
2963	drbd_suspend_io(device);
2964	wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2965	drbd_flush_workqueue(work_queue: &first_peer_device(device)->connection->sender_work);
2966
2967	/* If we happen to be C_STANDALONE R_SECONDARY, just change to
2968	* D_INCONSISTENT, and set all bits in the bitmap. Otherwise,
2969	* try to start a resync handshake as sync target for full sync.
2970	*/
2971	if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
2972	retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
2973	if (retcode >= SS_SUCCESS) {
2974	if (drbd_bitmap_io(device, io_fn: &drbd_bmio_set_n_write,
2975	why: "set_n_write from invalidate", flags: BM_LOCKED_MASK, NULL))
2976	retcode = ERR_IO_MD_DISK;
2977	}
2978	} else
2979	retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
2980	drbd_resume_io(device);
2981	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
2982	put_ldev(device);
2983	out:
2984	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
2985	return 0;
2986	}
2987
2988	static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2989	union drbd_state mask, union drbd_state val)
2990	{
2991	struct drbd_config_context adm_ctx;
2992	enum drbd_ret_code retcode;
2993
2994	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2995	if (!adm_ctx.reply_skb)
2996	return retcode;
2997	if (retcode != NO_ERROR)
2998	goto out;
2999
3000	mutex_lock(&adm_ctx.resource->adm_mutex);
3001	retcode = drbd_request_state(device: adm_ctx.device, mask, val);
3002	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
3003	out:
3004	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
3005	return 0;
3006	}
3007
3008	static int drbd_bmio_set_susp_al(struct drbd_device *device,
3009	struct drbd_peer_device *peer_device) __must_hold(local)
3010	{
3011	int rv;
3012
3013	rv = drbd_bmio_set_n_write(device, peer_device);
3014	drbd_suspend_al(device);
3015	return rv;
3016	}
3017
3018	int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
3019	{
3020	struct drbd_config_context adm_ctx;
3021	int retcode; /* drbd_ret_code, drbd_state_rv */
3022	struct drbd_device *device;
3023
3024	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3025	if (!adm_ctx.reply_skb)
3026	return retcode;
3027	if (retcode != NO_ERROR)
3028	goto out;
3029
3030	device = adm_ctx.device;
3031	if (!get_ldev(device)) {
3032	retcode = ERR_NO_DISK;
3033	goto out;
3034	}
3035
3036	mutex_lock(&adm_ctx.resource->adm_mutex);
3037
3038	/* If there is still bitmap IO pending, probably because of a previous
3039	* resync just being finished, wait for it before requesting a new resync.
3040	* Also wait for it's after_state_ch(). */
3041	drbd_suspend_io(device);
3042	wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
3043	drbd_flush_workqueue(work_queue: &first_peer_device(device)->connection->sender_work);
3044
3045	/* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
3046	* in the bitmap. Otherwise, try to start a resync handshake
3047	* as sync source for full sync.
3048	*/
3049	if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
3050	/* The peer will get a resync upon connect anyways. Just make that
3051	into a full resync. */
3052	retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
3053	if (retcode >= SS_SUCCESS) {
3054	if (drbd_bitmap_io(device, io_fn: &drbd_bmio_set_susp_al,
3055	why: "set_n_write from invalidate_peer",
3056	flags: BM_LOCKED_SET_ALLOWED, NULL))
3057	retcode = ERR_IO_MD_DISK;
3058	}
3059	} else
3060	retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
3061	drbd_resume_io(device);
3062	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
3063	put_ldev(device);
3064	out:
3065	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
3066	return 0;
3067	}
3068
3069	int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
3070	{
3071	struct drbd_config_context adm_ctx;
3072	enum drbd_ret_code retcode;
3073
3074	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3075	if (!adm_ctx.reply_skb)
3076	return retcode;
3077	if (retcode != NO_ERROR)
3078	goto out;
3079
3080	mutex_lock(&adm_ctx.resource->adm_mutex);
3081	if (drbd_request_state(device: adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
3082	retcode = ERR_PAUSE_IS_SET;
3083	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
3084	out:
3085	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
3086	return 0;
3087	}
3088
3089	int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
3090	{
3091	struct drbd_config_context adm_ctx;
3092	union drbd_dev_state s;
3093	enum drbd_ret_code retcode;
3094
3095	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3096	if (!adm_ctx.reply_skb)
3097	return retcode;
3098	if (retcode != NO_ERROR)
3099	goto out;
3100
3101	mutex_lock(&adm_ctx.resource->adm_mutex);
3102	if (drbd_request_state(device: adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
3103	s = adm_ctx.device->state;
3104	if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
3105	retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
3106	s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
3107	} else {
3108	retcode = ERR_PAUSE_IS_CLEAR;
3109	}
3110	}
3111	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
3112	out:
3113	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
3114	return 0;
3115	}
3116
3117	int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
3118	{
3119	return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
3120	}
3121
3122	int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
3123	{
3124	struct drbd_config_context adm_ctx;
3125	struct drbd_device *device;
3126	int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
3127
3128	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3129	if (!adm_ctx.reply_skb)
3130	return retcode;
3131	if (retcode != NO_ERROR)
3132	goto out;
3133
3134	mutex_lock(&adm_ctx.resource->adm_mutex);
3135	device = adm_ctx.device;
3136	if (test_bit(NEW_CUR_UUID, &device->flags)) {
3137	if (get_ldev_if_state(device, D_ATTACHING)) {
3138	drbd_uuid_new_current(device);
3139	put_ldev(device);
3140	} else {
3141	/* This is effectively a multi-stage "forced down".
3142	* The NEW_CUR_UUID bit is supposedly only set, if we
3143	* lost the replication connection, and are configured
3144	* to freeze IO and wait for some fence-peer handler.
3145	* So we still don't have a replication connection.
3146	* And now we don't have a local disk either. After
3147	* resume, we will fail all pending and new IO, because
3148	* we don't have any data anymore. Which means we will
3149	* eventually be able to terminate all users of this
3150	* device, and then take it down. By bumping the
3151	* "effective" data uuid, we make sure that you really
3152	* need to tear down before you reconfigure, we will
3153	* the refuse to re-connect or re-attach (because no
3154	* matching real data uuid exists).
3155	*/
3156	u64 val;
3157	get_random_bytes(buf: &val, len: sizeof(u64));
3158	drbd_set_ed_uuid(device, val);
3159	drbd_warn(device, "Resumed without access to data; please tear down before attempting to re-configure.\n");
3160	}
3161	clear_bit(nr: NEW_CUR_UUID, addr: &device->flags);
3162	}
3163	drbd_suspend_io(device);
3164	retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
3165	if (retcode == SS_SUCCESS) {
3166	if (device->state.conn < C_CONNECTED)
3167	tl_clear(first_peer_device(device)->connection);
3168	if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
3169	tl_restart(connection: first_peer_device(device)->connection, what: FAIL_FROZEN_DISK_IO);
3170	}
3171	drbd_resume_io(device);
3172	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
3173	out:
3174	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
3175	return 0;
3176	}
3177
3178	int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
3179	{
3180	return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
3181	}
3182
3183	static int nla_put_drbd_cfg_context(struct sk_buff *skb,
3184	struct drbd_resource *resource,
3185	struct drbd_connection *connection,
3186	struct drbd_device *device)
3187	{
3188	struct nlattr *nla;
3189	nla = nla_nest_start_noflag(skb, attrtype: DRBD_NLA_CFG_CONTEXT);
3190	if (!nla)
3191	goto nla_put_failure;
3192	if (device &&
3193	nla_put_u32(skb, attrtype: T_ctx_volume, value: device->vnr))
3194	goto nla_put_failure;
3195	if (nla_put_string(skb, attrtype: T_ctx_resource_name, str: resource->name))
3196	goto nla_put_failure;
3197	if (connection) {
3198	if (connection->my_addr_len &&
3199	nla_put(skb, attrtype: T_ctx_my_addr, attrlen: connection->my_addr_len, data: &connection->my_addr))
3200	goto nla_put_failure;
3201	if (connection->peer_addr_len &&
3202	nla_put(skb, attrtype: T_ctx_peer_addr, attrlen: connection->peer_addr_len, data: &connection->peer_addr))
3203	goto nla_put_failure;
3204	}
3205	nla_nest_end(skb, start: nla);
3206	return 0;
3207
3208	nla_put_failure:
3209	if (nla)
3210	nla_nest_cancel(skb, start: nla);
3211	return -EMSGSIZE;
3212	}
3213
3214	/*
3215	* The generic netlink dump callbacks are called outside the genl_lock(), so
3216	* they cannot use the simple attribute parsing code which uses global
3217	* attribute tables.
3218	*/
3219	static struct nlattr *find_cfg_context_attr(const struct nlmsghdr *nlh, int attr)
3220	{
3221	const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3222	const int maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
3223	struct nlattr *nla;
3224
3225	nla = nla_find(head: nlmsg_attrdata(nlh, hdrlen), len: nlmsg_attrlen(nlh, hdrlen),
3226	attrtype: DRBD_NLA_CFG_CONTEXT);
3227	if (!nla)
3228	return NULL;
3229	return drbd_nla_find_nested(maxtype, nla, __nla_type(attr));
3230	}
3231
3232	static void resource_to_info(struct resource_info *, struct drbd_resource *);
3233
3234	int drbd_adm_dump_resources(struct sk_buff *skb, struct netlink_callback *cb)
3235	{
3236	struct drbd_genlmsghdr *dh;
3237	struct drbd_resource *resource;
3238	struct resource_info resource_info;
3239	struct resource_statistics resource_statistics;
3240	int err;
3241
3242	rcu_read_lock();
3243	if (cb->args[0]) {
3244	for_each_resource_rcu(resource, &drbd_resources)
3245	if (resource == (struct drbd_resource *)cb->args[0])
3246	goto found_resource;
3247	err = 0; /* resource was probably deleted */
3248	goto out;
3249	}
3250	resource = list_entry(&drbd_resources,
3251	struct drbd_resource, resources);
3252
3253	found_resource:
3254	list_for_each_entry_continue_rcu(resource, &drbd_resources, resources) {
3255	goto put_result;
3256	}
3257	err = 0;
3258	goto out;
3259
3260	put_result:
3261	dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3262	seq: cb->nlh->nlmsg_seq, family: &drbd_genl_family,
3263	NLM_F_MULTI, cmd: DRBD_ADM_GET_RESOURCES);
3264	err = -ENOMEM;
3265	if (!dh)
3266	goto out;
3267	dh->minor = -1U;
3268	dh->ret_code = NO_ERROR;
3269	err = nla_put_drbd_cfg_context(skb, resource, NULL, NULL);
3270	if (err)
3271	goto out;
3272	err = res_opts_to_skb(skb, s: &resource->res_opts, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3273	if (err)
3274	goto out;
3275	resource_to_info(&resource_info, resource);
3276	err = resource_info_to_skb(skb, s: &resource_info, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3277	if (err)
3278	goto out;
3279	resource_statistics.res_stat_write_ordering = resource->write_ordering;
3280	err = resource_statistics_to_skb(skb, s: &resource_statistics, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3281	if (err)
3282	goto out;
3283	cb->args[0] = (long)resource;
3284	genlmsg_end(skb, hdr: dh);
3285	err = 0;
3286
3287	out:
3288	rcu_read_unlock();
3289	if (err)
3290	return err;
3291	return skb->len;
3292	}
3293
3294	static void device_to_statistics(struct device_statistics *s,
3295	struct drbd_device *device)
3296	{
3297	memset(s, 0, sizeof(*s));
3298	s->dev_upper_blocked = !may_inc_ap_bio(device);
3299	if (get_ldev(device)) {
3300	struct drbd_md *md = &device->ldev->md;
3301	u64 *history_uuids = (u64 *)s->history_uuids;
3302	int n;
3303
3304	spin_lock_irq(lock: &md->uuid_lock);
3305	s->dev_current_uuid = md->uuid[UI_CURRENT];
3306	BUILD_BUG_ON(sizeof(s->history_uuids) < UI_HISTORY_END - UI_HISTORY_START + 1);
3307	for (n = 0; n < UI_HISTORY_END - UI_HISTORY_START + 1; n++)
3308	history_uuids[n] = md->uuid[UI_HISTORY_START + n];
3309	for (; n < HISTORY_UUIDS; n++)
3310	history_uuids[n] = 0;
3311	s->history_uuids_len = HISTORY_UUIDS;
3312	spin_unlock_irq(lock: &md->uuid_lock);
3313
3314	s->dev_disk_flags = md->flags;
3315	put_ldev(device);
3316	}
3317	s->dev_size = get_capacity(disk: device->vdisk);
3318	s->dev_read = device->read_cnt;
3319	s->dev_write = device->writ_cnt;
3320	s->dev_al_writes = device->al_writ_cnt;
3321	s->dev_bm_writes = device->bm_writ_cnt;
3322	s->dev_upper_pending = atomic_read(v: &device->ap_bio_cnt);
3323	s->dev_lower_pending = atomic_read(v: &device->local_cnt);
3324	s->dev_al_suspended = test_bit(AL_SUSPENDED, &device->flags);
3325	s->dev_exposed_data_uuid = device->ed_uuid;
3326	}
3327
3328	static int put_resource_in_arg0(struct netlink_callback *cb, int holder_nr)
3329	{
3330	if (cb->args[0]) {
3331	struct drbd_resource *resource =
3332	(struct drbd_resource *)cb->args[0];
3333	kref_put(kref: &resource->kref, release: drbd_destroy_resource);
3334	}
3335
3336	return 0;
3337	}
3338
3339	int drbd_adm_dump_devices_done(struct netlink_callback *cb) {
3340	return put_resource_in_arg0(cb, holder_nr: 7);
3341	}
3342
3343	static void device_to_info(struct device_info *, struct drbd_device *);
3344
3345	int drbd_adm_dump_devices(struct sk_buff *skb, struct netlink_callback *cb)
3346	{
3347	struct nlattr *resource_filter;
3348	struct drbd_resource *resource;
3349	struct drbd_device *device;
3350	int minor, err, retcode;
3351	struct drbd_genlmsghdr *dh;
3352	struct device_info device_info;
3353	struct device_statistics device_statistics;
3354	struct idr *idr_to_search;
3355
3356	resource = (struct drbd_resource *)cb->args[0];
3357	if (!cb->args[0] && !cb->args[1]) {
3358	resource_filter = find_cfg_context_attr(nlh: cb->nlh, attr: T_ctx_resource_name);
3359	if (resource_filter) {
3360	retcode = ERR_RES_NOT_KNOWN;
3361	resource = drbd_find_resource(name: nla_data(nla: resource_filter));
3362	if (!resource)
3363	goto put_result;
3364	cb->args[0] = (long)resource;
3365	}
3366	}
3367
3368	rcu_read_lock();
3369	minor = cb->args[1];
3370	idr_to_search = resource ? &resource->devices : &drbd_devices;
3371	device = idr_get_next(idr_to_search, nextid: &minor);
3372	if (!device) {
3373	err = 0;
3374	goto out;
3375	}
3376	idr_for_each_entry_continue(idr_to_search, device, minor) {
3377	retcode = NO_ERROR;
3378	goto put_result; /* only one iteration */
3379	}
3380	err = 0;
3381	goto out; /* no more devices */
3382
3383	put_result:
3384	dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3385	seq: cb->nlh->nlmsg_seq, family: &drbd_genl_family,
3386	NLM_F_MULTI, cmd: DRBD_ADM_GET_DEVICES);
3387	err = -ENOMEM;
3388	if (!dh)
3389	goto out;
3390	dh->ret_code = retcode;
3391	dh->minor = -1U;
3392	if (retcode == NO_ERROR) {
3393	dh->minor = device->minor;
3394	err = nla_put_drbd_cfg_context(skb, resource: device->resource, NULL, device);
3395	if (err)
3396	goto out;
3397	if (get_ldev(device)) {
3398	struct disk_conf *disk_conf =
3399	rcu_dereference(device->ldev->disk_conf);
3400
3401	err = disk_conf_to_skb(skb, s: disk_conf, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3402	put_ldev(device);
3403	if (err)
3404	goto out;
3405	}
3406	device_to_info(&device_info, device);
3407	err = device_info_to_skb(skb, s: &device_info, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3408	if (err)
3409	goto out;
3410
3411	device_to_statistics(s: &device_statistics, device);
3412	err = device_statistics_to_skb(skb, s: &device_statistics, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3413	if (err)
3414	goto out;
3415	cb->args[1] = minor + 1;
3416	}
3417	genlmsg_end(skb, hdr: dh);
3418	err = 0;
3419
3420	out:
3421	rcu_read_unlock();
3422	if (err)
3423	return err;
3424	return skb->len;
3425	}
3426
3427	int drbd_adm_dump_connections_done(struct netlink_callback *cb)
3428	{
3429	return put_resource_in_arg0(cb, holder_nr: 6);
3430	}
3431
3432	enum { SINGLE_RESOURCE, ITERATE_RESOURCES };
3433
3434	int drbd_adm_dump_connections(struct sk_buff *skb, struct netlink_callback *cb)
3435	{
3436	struct nlattr *resource_filter;
3437	struct drbd_resource *resource = NULL, *next_resource;
3438	struct drbd_connection *connection;
3439	int err = 0, retcode;
3440	struct drbd_genlmsghdr *dh;
3441	struct connection_info connection_info;
3442	struct connection_statistics connection_statistics;
3443
3444	rcu_read_lock();
3445	resource = (struct drbd_resource *)cb->args[0];
3446	if (!cb->args[0]) {
3447	resource_filter = find_cfg_context_attr(nlh: cb->nlh, attr: T_ctx_resource_name);
3448	if (resource_filter) {
3449	retcode = ERR_RES_NOT_KNOWN;
3450	resource = drbd_find_resource(name: nla_data(nla: resource_filter));
3451	if (!resource)
3452	goto put_result;
3453	cb->args[0] = (long)resource;
3454	cb->args[1] = SINGLE_RESOURCE;
3455	}
3456	}
3457	if (!resource) {
3458	if (list_empty(head: &drbd_resources))
3459	goto out;
3460	resource = list_first_entry(&drbd_resources, struct drbd_resource, resources);
3461	kref_get(kref: &resource->kref);
3462	cb->args[0] = (long)resource;
3463	cb->args[1] = ITERATE_RESOURCES;
3464	}
3465
3466	next_resource:
3467	rcu_read_unlock();
3468	mutex_lock(&resource->conf_update);
3469	rcu_read_lock();
3470	if (cb->args[2]) {
3471	for_each_connection_rcu(connection, resource)
3472	if (connection == (struct drbd_connection *)cb->args[2])
3473	goto found_connection;
3474	/* connection was probably deleted */
3475	goto no_more_connections;
3476	}
3477	connection = list_entry(&resource->connections, struct drbd_connection, connections);
3478
3479	found_connection:
3480	list_for_each_entry_continue_rcu(connection, &resource->connections, connections) {
3481	if (!has_net_conf(connection))
3482	continue;
3483	retcode = NO_ERROR;
3484	goto put_result; /* only one iteration */
3485	}
3486
3487	no_more_connections:
3488	if (cb->args[1] == ITERATE_RESOURCES) {
3489	for_each_resource_rcu(next_resource, &drbd_resources) {
3490	if (next_resource == resource)
3491	goto found_resource;
3492	}
3493	/* resource was probably deleted */
3494	}
3495	goto out;
3496
3497	found_resource:
3498	list_for_each_entry_continue_rcu(next_resource, &drbd_resources, resources) {
3499	mutex_unlock(lock: &resource->conf_update);
3500	kref_put(kref: &resource->kref, release: drbd_destroy_resource);
3501	resource = next_resource;
3502	kref_get(kref: &resource->kref);
3503	cb->args[0] = (long)resource;
3504	cb->args[2] = 0;
3505	goto next_resource;
3506	}
3507	goto out; /* no more resources */
3508
3509	put_result:
3510	dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3511	seq: cb->nlh->nlmsg_seq, family: &drbd_genl_family,
3512	NLM_F_MULTI, cmd: DRBD_ADM_GET_CONNECTIONS);
3513	err = -ENOMEM;
3514	if (!dh)
3515	goto out;
3516	dh->ret_code = retcode;
3517	dh->minor = -1U;
3518	if (retcode == NO_ERROR) {
3519	struct net_conf *net_conf;
3520
3521	err = nla_put_drbd_cfg_context(skb, resource, connection, NULL);
3522	if (err)
3523	goto out;
3524	net_conf = rcu_dereference(connection->net_conf);
3525	if (net_conf) {
3526	err = net_conf_to_skb(skb, s: net_conf, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3527	if (err)
3528	goto out;
3529	}
3530	connection_to_info(info: &connection_info, connection);
3531	err = connection_info_to_skb(skb, s: &connection_info, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3532	if (err)
3533	goto out;
3534	connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
3535	err = connection_statistics_to_skb(skb, s: &connection_statistics, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3536	if (err)
3537	goto out;
3538	cb->args[2] = (long)connection;
3539	}
3540	genlmsg_end(skb, hdr: dh);
3541	err = 0;
3542
3543	out:
3544	rcu_read_unlock();
3545	if (resource)
3546	mutex_unlock(lock: &resource->conf_update);
3547	if (err)
3548	return err;
3549	return skb->len;
3550	}
3551
3552	enum mdf_peer_flag {
3553	MDF_PEER_CONNECTED = 1 << 0,
3554	MDF_PEER_OUTDATED = 1 << 1,
3555	MDF_PEER_FENCING = 1 << 2,
3556	MDF_PEER_FULL_SYNC = 1 << 3,
3557	};
3558
3559	static void peer_device_to_statistics(struct peer_device_statistics *s,
3560	struct drbd_peer_device *peer_device)
3561	{
3562	struct drbd_device *device = peer_device->device;
3563
3564	memset(s, 0, sizeof(*s));
3565	s->peer_dev_received = device->recv_cnt;
3566	s->peer_dev_sent = device->send_cnt;
3567	s->peer_dev_pending = atomic_read(v: &device->ap_pending_cnt) +
3568	atomic_read(v: &device->rs_pending_cnt);
3569	s->peer_dev_unacked = atomic_read(v: &device->unacked_cnt);
3570	s->peer_dev_out_of_sync = drbd_bm_total_weight(device) << (BM_BLOCK_SHIFT - 9);
3571	s->peer_dev_resync_failed = device->rs_failed << (BM_BLOCK_SHIFT - 9);
3572	if (get_ldev(device)) {
3573	struct drbd_md *md = &device->ldev->md;
3574
3575	spin_lock_irq(lock: &md->uuid_lock);
3576	s->peer_dev_bitmap_uuid = md->uuid[UI_BITMAP];
3577	spin_unlock_irq(lock: &md->uuid_lock);
3578	s->peer_dev_flags =
3579	(drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND) ?
3580	MDF_PEER_CONNECTED : 0) +
3581	(drbd_md_test_flag(device->ldev, MDF_CONSISTENT) &&
3582	!drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE) ?
3583	MDF_PEER_OUTDATED : 0) +
3584	/* FIXME: MDF_PEER_FENCING? */
3585	(drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ?
3586	MDF_PEER_FULL_SYNC : 0);
3587	put_ldev(device);
3588	}
3589	}
3590
3591	int drbd_adm_dump_peer_devices_done(struct netlink_callback *cb)
3592	{
3593	return put_resource_in_arg0(cb, holder_nr: 9);
3594	}
3595
3596	int drbd_adm_dump_peer_devices(struct sk_buff *skb, struct netlink_callback *cb)
3597	{
3598	struct nlattr *resource_filter;
3599	struct drbd_resource *resource;
3600	struct drbd_device *device;
3601	struct drbd_peer_device *peer_device = NULL;
3602	int minor, err, retcode;
3603	struct drbd_genlmsghdr *dh;
3604	struct idr *idr_to_search;
3605
3606	resource = (struct drbd_resource *)cb->args[0];
3607	if (!cb->args[0] && !cb->args[1]) {
3608	resource_filter = find_cfg_context_attr(nlh: cb->nlh, attr: T_ctx_resource_name);
3609	if (resource_filter) {
3610	retcode = ERR_RES_NOT_KNOWN;
3611	resource = drbd_find_resource(name: nla_data(nla: resource_filter));
3612	if (!resource)
3613	goto put_result;
3614	}
3615	cb->args[0] = (long)resource;
3616	}
3617
3618	rcu_read_lock();
3619	minor = cb->args[1];
3620	idr_to_search = resource ? &resource->devices : &drbd_devices;
3621	device = idr_find(idr_to_search, id: minor);
3622	if (!device) {
3623	next_device:
3624	minor++;
3625	cb->args[2] = 0;
3626	device = idr_get_next(idr_to_search, nextid: &minor);
3627	if (!device) {
3628	err = 0;
3629	goto out;
3630	}
3631	}
3632	if (cb->args[2]) {
3633	for_each_peer_device(peer_device, device)
3634	if (peer_device == (struct drbd_peer_device *)cb->args[2])
3635	goto found_peer_device;
3636	/* peer device was probably deleted */
3637	goto next_device;
3638	}
3639	/* Make peer_device point to the list head (not the first entry). */
3640	peer_device = list_entry(&device->peer_devices, struct drbd_peer_device, peer_devices);
3641
3642	found_peer_device:
3643	list_for_each_entry_continue_rcu(peer_device, &device->peer_devices, peer_devices) {
3644	if (!has_net_conf(connection: peer_device->connection))
3645	continue;
3646	retcode = NO_ERROR;
3647	goto put_result; /* only one iteration */
3648	}
3649	goto next_device;
3650
3651	put_result:
3652	dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3653	seq: cb->nlh->nlmsg_seq, family: &drbd_genl_family,
3654	NLM_F_MULTI, cmd: DRBD_ADM_GET_PEER_DEVICES);
3655	err = -ENOMEM;
3656	if (!dh)
3657	goto out;
3658	dh->ret_code = retcode;
3659	dh->minor = -1U;
3660	if (retcode == NO_ERROR) {
3661	struct peer_device_info peer_device_info;
3662	struct peer_device_statistics peer_device_statistics;
3663
3664	dh->minor = minor;
3665	err = nla_put_drbd_cfg_context(skb, resource: device->resource, connection: peer_device->connection, device);
3666	if (err)
3667	goto out;
3668	peer_device_to_info(info: &peer_device_info, peer_device);
3669	err = peer_device_info_to_skb(skb, s: &peer_device_info, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3670	if (err)
3671	goto out;
3672	peer_device_to_statistics(s: &peer_device_statistics, peer_device);
3673	err = peer_device_statistics_to_skb(skb, s: &peer_device_statistics, exclude_sensitive: !capable(CAP_SYS_ADMIN));
3674	if (err)
3675	goto out;
3676	cb->args[1] = minor;
3677	cb->args[2] = (long)peer_device;
3678	}
3679	genlmsg_end(skb, hdr: dh);
3680	err = 0;
3681
3682	out:
3683	rcu_read_unlock();
3684	if (err)
3685	return err;
3686	return skb->len;
3687	}
3688	/*
3689	* Return the connection of @resource if @resource has exactly one connection.
3690	*/
3691	static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
3692	{
3693	struct list_head *connections = &resource->connections;
3694
3695	if (list_empty(head: connections) || connections->next->next != connections)
3696	return NULL;
3697	return list_first_entry(&resource->connections, struct drbd_connection, connections);
3698	}
3699
3700	static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
3701	const struct sib_info *sib)
3702	{
3703	struct drbd_resource *resource = device->resource;
3704	struct state_info *si = NULL; /* for sizeof(si->member); */
3705	struct nlattr *nla;
3706	int got_ldev;
3707	int err = 0;
3708	int exclude_sensitive;
3709
3710	/* If sib != NULL, this is drbd_bcast_event, which anyone can listen
3711	* to. So we better exclude_sensitive information.
3712	*
3713	* If sib == NULL, this is drbd_adm_get_status, executed synchronously
3714	* in the context of the requesting user process. Exclude sensitive
3715	* information, unless current has superuser.
3716	*
3717	* NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
3718	* relies on the current implementation of netlink_dump(), which
3719	* executes the dump callback successively from netlink_recvmsg(),
3720	* always in the context of the receiving process */
3721	exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
3722
3723	got_ldev = get_ldev(device);
3724
3725	/* We need to add connection name and volume number information still.
3726	* Minor number is in drbd_genlmsghdr. */
3727	if (nla_put_drbd_cfg_context(skb, resource, connection: the_only_connection(resource), device))
3728	goto nla_put_failure;
3729
3730	if (res_opts_to_skb(skb, s: &device->resource->res_opts, exclude_sensitive))
3731	goto nla_put_failure;
3732
3733	rcu_read_lock();
3734	if (got_ldev) {
3735	struct disk_conf *disk_conf;
3736
3737	disk_conf = rcu_dereference(device->ldev->disk_conf);
3738	err = disk_conf_to_skb(skb, s: disk_conf, exclude_sensitive);
3739	}
3740	if (!err) {
3741	struct net_conf *nc;
3742
3743	nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3744	if (nc)
3745	err = net_conf_to_skb(skb, s: nc, exclude_sensitive);
3746	}
3747	rcu_read_unlock();
3748	if (err)
3749	goto nla_put_failure;
3750
3751	nla = nla_nest_start_noflag(skb, attrtype: DRBD_NLA_STATE_INFO);
3752	if (!nla)
3753	goto nla_put_failure;
3754	if (nla_put_u32(skb, attrtype: T_sib_reason, value: sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
3755	nla_put_u32(skb, attrtype: T_current_state, value: device->state.i) ||
3756	nla_put_u64_0pad(skb, attrtype: T_ed_uuid, value: device->ed_uuid) ||
3757	nla_put_u64_0pad(skb, attrtype: T_capacity, value: get_capacity(disk: device->vdisk)) ||
3758	nla_put_u64_0pad(skb, attrtype: T_send_cnt, value: device->send_cnt) ||
3759	nla_put_u64_0pad(skb, attrtype: T_recv_cnt, value: device->recv_cnt) ||
3760	nla_put_u64_0pad(skb, attrtype: T_read_cnt, value: device->read_cnt) ||
3761	nla_put_u64_0pad(skb, attrtype: T_writ_cnt, value: device->writ_cnt) ||
3762	nla_put_u64_0pad(skb, attrtype: T_al_writ_cnt, value: device->al_writ_cnt) ||
3763	nla_put_u64_0pad(skb, attrtype: T_bm_writ_cnt, value: device->bm_writ_cnt) ||
3764	nla_put_u32(skb, attrtype: T_ap_bio_cnt, value: atomic_read(v: &device->ap_bio_cnt)) ||
3765	nla_put_u32(skb, attrtype: T_ap_pending_cnt, value: atomic_read(v: &device->ap_pending_cnt)) ||
3766	nla_put_u32(skb, attrtype: T_rs_pending_cnt, value: atomic_read(v: &device->rs_pending_cnt)))
3767	goto nla_put_failure;
3768
3769	if (got_ldev) {
3770	int err;
3771
3772	spin_lock_irq(lock: &device->ldev->md.uuid_lock);
3773	err = nla_put(skb, attrtype: T_uuids, attrlen: sizeof(si->uuids), data: device->ldev->md.uuid);
3774	spin_unlock_irq(lock: &device->ldev->md.uuid_lock);
3775
3776	if (err)
3777	goto nla_put_failure;
3778
3779	if (nla_put_u32(skb, attrtype: T_disk_flags, value: device->ldev->md.flags) ||
3780	nla_put_u64_0pad(skb, attrtype: T_bits_total, value: drbd_bm_bits(device)) ||
3781	nla_put_u64_0pad(skb, attrtype: T_bits_oos,
3782	value: drbd_bm_total_weight(device)))
3783	goto nla_put_failure;
3784	if (C_SYNC_SOURCE <= device->state.conn &&
3785	C_PAUSED_SYNC_T >= device->state.conn) {
3786	if (nla_put_u64_0pad(skb, attrtype: T_bits_rs_total,
3787	value: device->rs_total) ||
3788	nla_put_u64_0pad(skb, attrtype: T_bits_rs_failed,
3789	value: device->rs_failed))
3790	goto nla_put_failure;
3791	}
3792	}
3793
3794	if (sib) {
3795	switch(sib->sib_reason) {
3796	case SIB_SYNC_PROGRESS:
3797	case SIB_GET_STATUS_REPLY:
3798	break;
3799	case SIB_STATE_CHANGE:
3800	if (nla_put_u32(skb, attrtype: T_prev_state, value: sib->os.i) ||
3801	nla_put_u32(skb, attrtype: T_new_state, value: sib->ns.i))
3802	goto nla_put_failure;
3803	break;
3804	case SIB_HELPER_POST:
3805	if (nla_put_u32(skb, attrtype: T_helper_exit_code,
3806	value: sib->helper_exit_code))
3807	goto nla_put_failure;
3808	fallthrough;
3809	case SIB_HELPER_PRE:
3810	if (nla_put_string(skb, attrtype: T_helper, str: sib->helper_name))
3811	goto nla_put_failure;
3812	break;
3813	}
3814	}
3815	nla_nest_end(skb, start: nla);
3816
3817	if (0)
3818	nla_put_failure:
3819	err = -EMSGSIZE;
3820	if (got_ldev)
3821	put_ldev(device);
3822	return err;
3823	}
3824
3825	int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3826	{
3827	struct drbd_config_context adm_ctx;
3828	enum drbd_ret_code retcode;
3829	int err;
3830
3831	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3832	if (!adm_ctx.reply_skb)
3833	return retcode;
3834	if (retcode != NO_ERROR)
3835	goto out;
3836
3837	err = nla_put_status_info(skb: adm_ctx.reply_skb, device: adm_ctx.device, NULL);
3838	if (err) {
3839	nlmsg_free(skb: adm_ctx.reply_skb);
3840	return err;
3841	}
3842	out:
3843	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
3844	return 0;
3845	}
3846
3847	static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3848	{
3849	struct drbd_device *device;
3850	struct drbd_genlmsghdr *dh;
3851	struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3852	struct drbd_resource *resource = NULL;
3853	struct drbd_resource *tmp;
3854	unsigned volume = cb->args[1];
3855
3856	/* Open coded, deferred, iteration:
3857	* for_each_resource_safe(resource, tmp, &drbd_resources) {
3858	* connection = "first connection of resource or undefined";
3859	* idr_for_each_entry(&resource->devices, device, i) {
3860	* ...
3861	* }
3862	* }
3863	* where resource is cb->args[0];
3864	* and i is cb->args[1];
3865	*
3866	* cb->args[2] indicates if we shall loop over all resources,
3867	* or just dump all volumes of a single resource.
3868	*
3869	* This may miss entries inserted after this dump started,
3870	* or entries deleted before they are reached.
3871	*
3872	* We need to make sure the device won't disappear while
3873	* we are looking at it, and revalidate our iterators
3874	* on each iteration.
3875	*/
3876
3877	/* synchronize with conn_create()/drbd_destroy_connection() */
3878	rcu_read_lock();
3879	/* revalidate iterator position */
3880	for_each_resource_rcu(tmp, &drbd_resources) {
3881	if (pos == NULL) {
3882	/* first iteration */
3883	pos = tmp;
3884	resource = pos;
3885	break;
3886	}
3887	if (tmp == pos) {
3888	resource = pos;
3889	break;
3890	}
3891	}
3892	if (resource) {
3893	next_resource:
3894	device = idr_get_next(&resource->devices, nextid: &volume);
3895	if (!device) {
3896	/* No more volumes to dump on this resource.
3897	* Advance resource iterator. */
3898	pos = list_entry_rcu(resource->resources.next,
3899	struct drbd_resource, resources);
3900	/* Did we dump any volume of this resource yet? */
3901	if (volume != 0) {
3902	/* If we reached the end of the list,
3903	* or only a single resource dump was requested,
3904	* we are done. */
3905	if (&pos->resources == &drbd_resources || cb->args[2])
3906	goto out;
3907	volume = 0;
3908	resource = pos;
3909	goto next_resource;
3910	}
3911	}
3912
3913	dh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
3914	seq: cb->nlh->nlmsg_seq, family: &drbd_genl_family,
3915	NLM_F_MULTI, cmd: DRBD_ADM_GET_STATUS);
3916	if (!dh)
3917	goto out;
3918
3919	if (!device) {
3920	/* This is a connection without a single volume.
3921	* Suprisingly enough, it may have a network
3922	* configuration. */
3923	struct drbd_connection *connection;
3924
3925	dh->minor = -1U;
3926	dh->ret_code = NO_ERROR;
3927	connection = the_only_connection(resource);
3928	if (nla_put_drbd_cfg_context(skb, resource, connection, NULL))
3929	goto cancel;
3930	if (connection) {
3931	struct net_conf *nc;
3932
3933	nc = rcu_dereference(connection->net_conf);
3934	if (nc && net_conf_to_skb(skb, s: nc, exclude_sensitive: 1) != 0)
3935	goto cancel;
3936	}
3937	goto done;
3938	}
3939
3940	D_ASSERT(device, device->vnr == volume);
3941	D_ASSERT(device, device->resource == resource);
3942
3943	dh->minor = device_to_minor(device);
3944	dh->ret_code = NO_ERROR;
3945
3946	if (nla_put_status_info(skb, device, NULL)) {
3947	cancel:
3948	genlmsg_cancel(skb, hdr: dh);
3949	goto out;
3950	}
3951	done:
3952	genlmsg_end(skb, hdr: dh);
3953	}
3954
3955	out:
3956	rcu_read_unlock();
3957	/* where to start the next iteration */
3958	cb->args[0] = (long)pos;
3959	cb->args[1] = (pos == resource) ? volume + 1 : 0;
3960
3961	/* No more resources/volumes/minors found results in an empty skb.
3962	* Which will terminate the dump. */
3963	return skb->len;
3964	}
3965
3966	/*
3967	* Request status of all resources, or of all volumes within a single resource.
3968	*
3969	* This is a dump, as the answer may not fit in a single reply skb otherwise.
3970	* Which means we cannot use the family->attrbuf or other such members, because
3971	* dump is NOT protected by the genl_lock(). During dump, we only have access
3972	* to the incoming skb, and need to opencode "parsing" of the nlattr payload.
3973	*
3974	* Once things are setup properly, we call into get_one_status().
3975	*/
3976	int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb)
3977	{
3978	const unsigned hdrlen = GENL_HDRLEN + GENL_MAGIC_FAMILY_HDRSZ;
3979	struct nlattr *nla;
3980	const char *resource_name;
3981	struct drbd_resource *resource;
3982	int maxtype;
3983
3984	/* Is this a followup call? */
3985	if (cb->args[0]) {
3986	/* ... of a single resource dump,
3987	* and the resource iterator has been advanced already? */
3988	if (cb->args[2] && cb->args[2] != cb->args[0])
3989	return 0; /* DONE. */
3990	goto dump;
3991	}
3992
3993	/* First call (from netlink_dump_start). We need to figure out
3994	* which resource(s) the user wants us to dump. */
3995	nla = nla_find(head: nlmsg_attrdata(nlh: cb->nlh, hdrlen),
3996	len: nlmsg_attrlen(nlh: cb->nlh, hdrlen),
3997	attrtype: DRBD_NLA_CFG_CONTEXT);
3998
3999	/* No explicit context given. Dump all. */
4000	if (!nla)
4001	goto dump;
4002	maxtype = ARRAY_SIZE(drbd_cfg_context_nl_policy) - 1;
4003	nla = drbd_nla_find_nested(maxtype, nla, __nla_type(T_ctx_resource_name));
4004	if (IS_ERR(ptr: nla))
4005	return PTR_ERR(ptr: nla);
4006	/* context given, but no name present? */
4007	if (!nla)
4008	return -EINVAL;
4009	resource_name = nla_data(nla);
4010	if (!*resource_name)
4011	return -ENODEV;
4012	resource = drbd_find_resource(name: resource_name);
4013	if (!resource)
4014	return -ENODEV;
4015
4016	kref_put(kref: &resource->kref, release: drbd_destroy_resource); /* get_one_status() revalidates the resource */
4017
4018	/* prime iterators, and set "filter" mode mark:
4019	* only dump this connection. */
4020	cb->args[0] = (long)resource;
4021	/* cb->args[1] = 0; passed in this way. */
4022	cb->args[2] = (long)resource;
4023
4024	dump:
4025	return get_one_status(skb, cb);
4026	}
4027
4028	int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info)
4029	{
4030	struct drbd_config_context adm_ctx;
4031	enum drbd_ret_code retcode;
4032	struct timeout_parms tp;
4033	int err;
4034
4035	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4036	if (!adm_ctx.reply_skb)
4037	return retcode;
4038	if (retcode != NO_ERROR)
4039	goto out;
4040
4041	tp.timeout_type =
4042	adm_ctx.device->state.pdsk == D_OUTDATED ? UT_PEER_OUTDATED :
4043	test_bit(USE_DEGR_WFC_T, &adm_ctx.device->flags) ? UT_DEGRADED :
4044	UT_DEFAULT;
4045
4046	err = timeout_parms_to_priv_skb(skb: adm_ctx.reply_skb, s: &tp);
4047	if (err) {
4048	nlmsg_free(skb: adm_ctx.reply_skb);
4049	return err;
4050	}
4051	out:
4052	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
4053	return 0;
4054	}
4055
4056	int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info)
4057	{
4058	struct drbd_config_context adm_ctx;
4059	struct drbd_device *device;
4060	enum drbd_ret_code retcode;
4061	struct start_ov_parms parms;
4062
4063	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4064	if (!adm_ctx.reply_skb)
4065	return retcode;
4066	if (retcode != NO_ERROR)
4067	goto out;
4068
4069	device = adm_ctx.device;
4070
4071	/* resume from last known position, if possible */
4072	parms.ov_start_sector = device->ov_start_sector;
4073	parms.ov_stop_sector = ULLONG_MAX;
4074	if (info->attrs[DRBD_NLA_START_OV_PARMS]) {
4075	int err = start_ov_parms_from_attrs(s: &parms, info);
4076	if (err) {
4077	retcode = ERR_MANDATORY_TAG;
4078	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
4079	goto out;
4080	}
4081	}
4082	mutex_lock(&adm_ctx.resource->adm_mutex);
4083
4084	/* w_make_ov_request expects position to be aligned */
4085	device->ov_start_sector = parms.ov_start_sector & ~(BM_SECT_PER_BIT-1);
4086	device->ov_stop_sector = parms.ov_stop_sector;
4087
4088	/* If there is still bitmap IO pending, e.g. previous resync or verify
4089	* just being finished, wait for it before requesting a new resync. */
4090	drbd_suspend_io(device);
4091	wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
4092	retcode = drbd_request_state(device, NS(conn, C_VERIFY_S));
4093	drbd_resume_io(device);
4094
4095	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
4096	out:
4097	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
4098	return 0;
4099	}
4100
4101
4102	int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
4103	{
4104	struct drbd_config_context adm_ctx;
4105	struct drbd_device *device;
4106	enum drbd_ret_code retcode;
4107	int skip_initial_sync = 0;
4108	int err;
4109	struct new_c_uuid_parms args;
4110
4111	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4112	if (!adm_ctx.reply_skb)
4113	return retcode;
4114	if (retcode != NO_ERROR)
4115	goto out_nolock;
4116
4117	device = adm_ctx.device;
4118	memset(&args, 0, sizeof(args));
4119	if (info->attrs[DRBD_NLA_NEW_C_UUID_PARMS]) {
4120	err = new_c_uuid_parms_from_attrs(s: &args, info);
4121	if (err) {
4122	retcode = ERR_MANDATORY_TAG;
4123	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
4124	goto out_nolock;
4125	}
4126	}
4127
4128	mutex_lock(&adm_ctx.resource->adm_mutex);
4129	mutex_lock(device->state_mutex); /* Protects us against serialized state changes. */
4130
4131	if (!get_ldev(device)) {
4132	retcode = ERR_NO_DISK;
4133	goto out;
4134	}
4135
4136	/* this is "skip initial sync", assume to be clean */
4137	if (device->state.conn == C_CONNECTED &&
4138	first_peer_device(device)->connection->agreed_pro_version >= 90 &&
4139	device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
4140	drbd_info(device, "Preparing to skip initial sync\n");
4141	skip_initial_sync = 1;
4142	} else if (device->state.conn != C_STANDALONE) {
4143	retcode = ERR_CONNECTED;
4144	goto out_dec;
4145	}
4146
4147	drbd_uuid_set(device, idx: UI_BITMAP, val: 0); /* Rotate UI_BITMAP to History 1, etc... */
4148	drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
4149
4150	if (args.clear_bm) {
4151	err = drbd_bitmap_io(device, io_fn: &drbd_bmio_clear_n_write,
4152	why: "clear_n_write from new_c_uuid", flags: BM_LOCKED_MASK, NULL);
4153	if (err) {
4154	drbd_err(device, "Writing bitmap failed with %d\n", err);
4155	retcode = ERR_IO_MD_DISK;
4156	}
4157	if (skip_initial_sync) {
4158	drbd_send_uuids_skip_initial_sync(first_peer_device(device));
4159	_drbd_uuid_set(device, idx: UI_BITMAP, val: 0);
4160	drbd_print_uuids(device, text: "cleared bitmap UUID");
4161	spin_lock_irq(lock: &device->resource->req_lock);
4162	_drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
4163	CS_VERBOSE, NULL);
4164	spin_unlock_irq(lock: &device->resource->req_lock);
4165	}
4166	}
4167
4168	drbd_md_sync(device);
4169	out_dec:
4170	put_ldev(device);
4171	out:
4172	mutex_unlock(lock: device->state_mutex);
4173	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
4174	out_nolock:
4175	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
4176	return 0;
4177	}
4178
4179	static enum drbd_ret_code
4180	drbd_check_resource_name(struct drbd_config_context *adm_ctx)
4181	{
4182	const char *name = adm_ctx->resource_name;
4183	if (!name || !name[0]) {
4184	drbd_msg_put_info(skb: adm_ctx->reply_skb, info: "resource name missing");
4185	return ERR_MANDATORY_TAG;
4186	}
4187	/* if we want to use these in sysfs/configfs/debugfs some day,
4188	* we must not allow slashes */
4189	if (strchr(name, '/')) {
4190	drbd_msg_put_info(skb: adm_ctx->reply_skb, info: "invalid resource name");
4191	return ERR_INVALID_REQUEST;
4192	}
4193	return NO_ERROR;
4194	}
4195
4196	static void resource_to_info(struct resource_info *info,
4197	struct drbd_resource *resource)
4198	{
4199	info->res_role = conn_highest_role(connection: first_connection(resource));
4200	info->res_susp = resource->susp;
4201	info->res_susp_nod = resource->susp_nod;
4202	info->res_susp_fen = resource->susp_fen;
4203	}
4204
4205	int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info)
4206	{
4207	struct drbd_connection *connection;
4208	struct drbd_config_context adm_ctx;
4209	enum drbd_ret_code retcode;
4210	struct res_opts res_opts;
4211	int err;
4212
4213	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, flags: 0);
4214	if (!adm_ctx.reply_skb)
4215	return retcode;
4216	if (retcode != NO_ERROR)
4217	goto out;
4218
4219	set_res_opts_defaults(&res_opts);
4220	err = res_opts_from_attrs(s: &res_opts, info);
4221	if (err && err != -ENOMSG) {
4222	retcode = ERR_MANDATORY_TAG;
4223	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: from_attrs_err_to_txt(err));
4224	goto out;
4225	}
4226
4227	retcode = drbd_check_resource_name(adm_ctx: &adm_ctx);
4228	if (retcode != NO_ERROR)
4229	goto out;
4230
4231	if (adm_ctx.resource) {
4232	if (info->nlhdr->nlmsg_flags & NLM_F_EXCL) {
4233	retcode = ERR_INVALID_REQUEST;
4234	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: "resource exists");
4235	}
4236	/* else: still NO_ERROR */
4237	goto out;
4238	}
4239
4240	/* not yet safe for genl_family.parallel_ops */
4241	mutex_lock(&resources_mutex);
4242	connection = conn_create(name: adm_ctx.resource_name, res_opts: &res_opts);
4243	mutex_unlock(lock: &resources_mutex);
4244
4245	if (connection) {
4246	struct resource_info resource_info;
4247
4248	mutex_lock(&notification_mutex);
4249	resource_to_info(info: &resource_info, resource: connection->resource);
4250	notify_resource_state(NULL, 0, connection->resource,
4251	&resource_info, NOTIFY_CREATE);
4252	mutex_unlock(lock: &notification_mutex);
4253	} else
4254	retcode = ERR_NOMEM;
4255
4256	out:
4257	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
4258	return 0;
4259	}
4260
4261	static void device_to_info(struct device_info *info,
4262	struct drbd_device *device)
4263	{
4264	info->dev_disk_state = device->state.disk;
4265	}
4266
4267
4268	int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info)
4269	{
4270	struct drbd_config_context adm_ctx;
4271	struct drbd_genlmsghdr *dh = genl_info_userhdr(info);
4272	enum drbd_ret_code retcode;
4273
4274	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4275	if (!adm_ctx.reply_skb)
4276	return retcode;
4277	if (retcode != NO_ERROR)
4278	goto out;
4279
4280	if (dh->minor > MINORMASK) {
4281	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: "requested minor out of range");
4282	retcode = ERR_INVALID_REQUEST;
4283	goto out;
4284	}
4285	if (adm_ctx.volume > DRBD_VOLUME_MAX) {
4286	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: "requested volume id out of range");
4287	retcode = ERR_INVALID_REQUEST;
4288	goto out;
4289	}
4290
4291	/* drbd_adm_prepare made sure already
4292	* that first_peer_device(device)->connection and device->vnr match the request. */
4293	if (adm_ctx.device) {
4294	if (info->nlhdr->nlmsg_flags & NLM_F_EXCL)
4295	retcode = ERR_MINOR_OR_VOLUME_EXISTS;
4296	/* else: still NO_ERROR */
4297	goto out;
4298	}
4299
4300	mutex_lock(&adm_ctx.resource->adm_mutex);
4301	retcode = drbd_create_device(adm_ctx: &adm_ctx, minor: dh->minor);
4302	if (retcode == NO_ERROR) {
4303	struct drbd_device *device;
4304	struct drbd_peer_device *peer_device;
4305	struct device_info info;
4306	unsigned int peer_devices = 0;
4307	enum drbd_notification_type flags;
4308
4309	device = minor_to_device(minor: dh->minor);
4310	for_each_peer_device(peer_device, device) {
4311	if (!has_net_conf(connection: peer_device->connection))
4312	continue;
4313	peer_devices++;
4314	}
4315
4316	device_to_info(info: &info, device);
4317	mutex_lock(&notification_mutex);
4318	flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4319	notify_device_state(NULL, 0, device, &info, NOTIFY_CREATE | flags);
4320	for_each_peer_device(peer_device, device) {
4321	struct peer_device_info peer_device_info;
4322
4323	if (!has_net_conf(connection: peer_device->connection))
4324	continue;
4325	peer_device_to_info(info: &peer_device_info, peer_device);
4326	flags = (peer_devices--) ? NOTIFY_CONTINUES : 0;
4327	notify_peer_device_state(NULL, 0, peer_device, &peer_device_info,
4328	NOTIFY_CREATE | flags);
4329	}
4330	mutex_unlock(lock: &notification_mutex);
4331	}
4332	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
4333	out:
4334	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
4335	return 0;
4336	}
4337
4338	static enum drbd_ret_code adm_del_minor(struct drbd_device *device)
4339	{
4340	struct drbd_peer_device *peer_device;
4341
4342	if (device->state.disk == D_DISKLESS &&
4343	/* no need to be device->state.conn == C_STANDALONE &&
4344	* we may want to delete a minor from a live replication group.
4345	*/
4346	device->state.role == R_SECONDARY) {
4347	struct drbd_connection *connection =
4348	first_connection(resource: device->resource);
4349
4350	_drbd_request_state(device, NS(conn, C_WF_REPORT_PARAMS),
4351	CS_VERBOSE + CS_WAIT_COMPLETE);
4352
4353	/* If the state engine hasn't stopped the sender thread yet, we
4354	* need to flush the sender work queue before generating the
4355	* DESTROY events here. */
4356	if (get_t_state(thi: &connection->worker) == RUNNING)
4357	drbd_flush_workqueue(work_queue: &connection->sender_work);
4358
4359	mutex_lock(&notification_mutex);
4360	for_each_peer_device(peer_device, device) {
4361	if (!has_net_conf(connection: peer_device->connection))
4362	continue;
4363	notify_peer_device_state(NULL, 0, peer_device, NULL,
4364	NOTIFY_DESTROY | NOTIFY_CONTINUES);
4365	}
4366	notify_device_state(NULL, 0, device, NULL, NOTIFY_DESTROY);
4367	mutex_unlock(lock: &notification_mutex);
4368
4369	drbd_delete_device(device);
4370	return NO_ERROR;
4371	} else
4372	return ERR_MINOR_CONFIGURED;
4373	}
4374
4375	int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info)
4376	{
4377	struct drbd_config_context adm_ctx;
4378	enum drbd_ret_code retcode;
4379
4380	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
4381	if (!adm_ctx.reply_skb)
4382	return retcode;
4383	if (retcode != NO_ERROR)
4384	goto out;
4385
4386	mutex_lock(&adm_ctx.resource->adm_mutex);
4387	retcode = adm_del_minor(device: adm_ctx.device);
4388	mutex_unlock(lock: &adm_ctx.resource->adm_mutex);
4389	out:
4390	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
4391	return 0;
4392	}
4393
4394	static int adm_del_resource(struct drbd_resource *resource)
4395	{
4396	struct drbd_connection *connection;
4397
4398	for_each_connection(connection, resource) {
4399	if (connection->cstate > C_STANDALONE)
4400	return ERR_NET_CONFIGURED;
4401	}
4402	if (!idr_is_empty(idr: &resource->devices))
4403	return ERR_RES_IN_USE;
4404
4405	/* The state engine has stopped the sender thread, so we don't
4406	* need to flush the sender work queue before generating the
4407	* DESTROY event here. */
4408	mutex_lock(&notification_mutex);
4409	notify_resource_state(NULL, 0, resource, NULL, NOTIFY_DESTROY);
4410	mutex_unlock(lock: &notification_mutex);
4411
4412	mutex_lock(&resources_mutex);
4413	list_del_rcu(entry: &resource->resources);
4414	mutex_unlock(lock: &resources_mutex);
4415	/* Make sure all threads have actually stopped: state handling only
4416	* does drbd_thread_stop_nowait(). */
4417	list_for_each_entry(connection, &resource->connections, connections)
4418	drbd_thread_stop(thi: &connection->worker);
4419	synchronize_rcu();
4420	drbd_free_resource(resource);
4421	return NO_ERROR;
4422	}
4423
4424	int drbd_adm_down(struct sk_buff *skb, struct genl_info *info)
4425	{
4426	struct drbd_config_context adm_ctx;
4427	struct drbd_resource *resource;
4428	struct drbd_connection *connection;
4429	struct drbd_device *device;
4430	int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
4431	unsigned i;
4432
4433	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4434	if (!adm_ctx.reply_skb)
4435	return retcode;
4436	if (retcode != NO_ERROR)
4437	goto finish;
4438
4439	resource = adm_ctx.resource;
4440	mutex_lock(&resource->adm_mutex);
4441	/* demote */
4442	for_each_connection(connection, resource) {
4443	struct drbd_peer_device *peer_device;
4444
4445	idr_for_each_entry(&connection->peer_devices, peer_device, i) {
4446	retcode = drbd_set_role(device: peer_device->device, new_role: R_SECONDARY, force: 0);
4447	if (retcode < SS_SUCCESS) {
4448	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: "failed to demote");
4449	goto out;
4450	}
4451	}
4452
4453	retcode = conn_try_disconnect(connection, force: 0);
4454	if (retcode < SS_SUCCESS) {
4455	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: "failed to disconnect");
4456	goto out;
4457	}
4458	}
4459
4460	/* detach */
4461	idr_for_each_entry(&resource->devices, device, i) {
4462	retcode = adm_detach(device, force: 0);
4463	if (retcode < SS_SUCCESS || retcode > NO_ERROR) {
4464	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: "failed to detach");
4465	goto out;
4466	}
4467	}
4468
4469	/* delete volumes */
4470	idr_for_each_entry(&resource->devices, device, i) {
4471	retcode = adm_del_minor(device);
4472	if (retcode != NO_ERROR) {
4473	/* "can not happen" */
4474	drbd_msg_put_info(skb: adm_ctx.reply_skb, info: "failed to delete volume");
4475	goto out;
4476	}
4477	}
4478
4479	retcode = adm_del_resource(resource);
4480	out:
4481	mutex_unlock(lock: &resource->adm_mutex);
4482	finish:
4483	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
4484	return 0;
4485	}
4486
4487	int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info)
4488	{
4489	struct drbd_config_context adm_ctx;
4490	struct drbd_resource *resource;
4491	enum drbd_ret_code retcode;
4492
4493	retcode = drbd_adm_prepare(adm_ctx: &adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
4494	if (!adm_ctx.reply_skb)
4495	return retcode;
4496	if (retcode != NO_ERROR)
4497	goto finish;
4498	resource = adm_ctx.resource;
4499
4500	mutex_lock(&resource->adm_mutex);
4501	retcode = adm_del_resource(resource);
4502	mutex_unlock(lock: &resource->adm_mutex);
4503	finish:
4504	drbd_adm_finish(adm_ctx: &adm_ctx, info, retcode);
4505	return 0;
4506	}
4507
4508	void drbd_bcast_event(struct drbd_device *device, const struct sib_info *sib)
4509	{
4510	struct sk_buff *msg;
4511	struct drbd_genlmsghdr *d_out;
4512	unsigned seq;
4513	int err = -ENOMEM;
4514
4515	seq = atomic_inc_return(v: &drbd_genl_seq);
4516	msg = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4517	if (!msg)
4518	goto failed;
4519
4520	err = -EMSGSIZE;
4521	d_out = genlmsg_put(skb: msg, portid: 0, seq, family: &drbd_genl_family, flags: 0, cmd: DRBD_EVENT);
4522	if (!d_out) /* cannot happen, but anyways. */
4523	goto nla_put_failure;
4524	d_out->minor = device_to_minor(device);
4525	d_out->ret_code = NO_ERROR;
4526
4527	if (nla_put_status_info(skb: msg, device, sib))
4528	goto nla_put_failure;
4529	genlmsg_end(skb: msg, hdr: d_out);
4530	err = drbd_genl_multicast_events(skb: msg, GFP_NOWAIT);
4531	/* msg has been consumed or freed in netlink_broadcast() */
4532	if (err && err != -ESRCH)
4533	goto failed;
4534
4535	return;
4536
4537	nla_put_failure:
4538	nlmsg_free(skb: msg);
4539	failed:
4540	drbd_err(device, "Error %d while broadcasting event. "
4541	"Event seq:%u sib_reason:%u\n",
4542	err, seq, sib->sib_reason);
4543	}
4544
4545	static int nla_put_notification_header(struct sk_buff *msg,
4546	enum drbd_notification_type type)
4547	{
4548	struct drbd_notification_header nh = {
4549	.nh_type = type,
4550	};
4551
4552	return drbd_notification_header_to_skb(skb: msg, s: &nh, exclude_sensitive: true);
4553	}
4554
4555	int notify_resource_state(struct sk_buff *skb,
4556	unsigned int seq,
4557	struct drbd_resource *resource,
4558	struct resource_info *resource_info,
4559	enum drbd_notification_type type)
4560	{
4561	struct resource_statistics resource_statistics;
4562	struct drbd_genlmsghdr *dh;
4563	bool multicast = false;
4564	int err;
4565
4566	if (!skb) {
4567	seq = atomic_inc_return(v: &notify_genl_seq);
4568	skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4569	err = -ENOMEM;
4570	if (!skb)
4571	goto failed;
4572	multicast = true;
4573	}
4574
4575	err = -EMSGSIZE;
4576	dh = genlmsg_put(skb, portid: 0, seq, family: &drbd_genl_family, flags: 0, cmd: DRBD_RESOURCE_STATE);
4577	if (!dh)
4578	goto nla_put_failure;
4579	dh->minor = -1U;
4580	dh->ret_code = NO_ERROR;
4581	if (nla_put_drbd_cfg_context(skb, resource, NULL, NULL) ||
4582	nla_put_notification_header(msg: skb, type) ||
4583	((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4584	resource_info_to_skb(skb, s: resource_info, exclude_sensitive: true)))
4585	goto nla_put_failure;
4586	resource_statistics.res_stat_write_ordering = resource->write_ordering;
4587	err = resource_statistics_to_skb(skb, s: &resource_statistics, exclude_sensitive: !capable(CAP_SYS_ADMIN));
4588	if (err)
4589	goto nla_put_failure;
4590	genlmsg_end(skb, hdr: dh);
4591	if (multicast) {
4592	err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4593	/* skb has been consumed or freed in netlink_broadcast() */
4594	if (err && err != -ESRCH)
4595	goto failed;
4596	}
4597	return 0;
4598
4599	nla_put_failure:
4600	nlmsg_free(skb);
4601	failed:
4602	drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4603	err, seq);
4604	return err;
4605	}
4606
4607	int notify_device_state(struct sk_buff *skb,
4608	unsigned int seq,
4609	struct drbd_device *device,
4610	struct device_info *device_info,
4611	enum drbd_notification_type type)
4612	{
4613	struct device_statistics device_statistics;
4614	struct drbd_genlmsghdr *dh;
4615	bool multicast = false;
4616	int err;
4617
4618	if (!skb) {
4619	seq = atomic_inc_return(v: &notify_genl_seq);
4620	skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4621	err = -ENOMEM;
4622	if (!skb)
4623	goto failed;
4624	multicast = true;
4625	}
4626
4627	err = -EMSGSIZE;
4628	dh = genlmsg_put(skb, portid: 0, seq, family: &drbd_genl_family, flags: 0, cmd: DRBD_DEVICE_STATE);
4629	if (!dh)
4630	goto nla_put_failure;
4631	dh->minor = device->minor;
4632	dh->ret_code = NO_ERROR;
4633	if (nla_put_drbd_cfg_context(skb, resource: device->resource, NULL, device) ||
4634	nla_put_notification_header(msg: skb, type) ||
4635	((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4636	device_info_to_skb(skb, s: device_info, exclude_sensitive: true)))
4637	goto nla_put_failure;
4638	device_to_statistics(s: &device_statistics, device);
4639	device_statistics_to_skb(skb, s: &device_statistics, exclude_sensitive: !capable(CAP_SYS_ADMIN));
4640	genlmsg_end(skb, hdr: dh);
4641	if (multicast) {
4642	err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4643	/* skb has been consumed or freed in netlink_broadcast() */
4644	if (err && err != -ESRCH)
4645	goto failed;
4646	}
4647	return 0;
4648
4649	nla_put_failure:
4650	nlmsg_free(skb);
4651	failed:
4652	drbd_err(device, "Error %d while broadcasting event. Event seq:%u\n",
4653	err, seq);
4654	return err;
4655	}
4656
4657	int notify_connection_state(struct sk_buff *skb,
4658	unsigned int seq,
4659	struct drbd_connection *connection,
4660	struct connection_info *connection_info,
4661	enum drbd_notification_type type)
4662	{
4663	struct connection_statistics connection_statistics;
4664	struct drbd_genlmsghdr *dh;
4665	bool multicast = false;
4666	int err;
4667
4668	if (!skb) {
4669	seq = atomic_inc_return(v: &notify_genl_seq);
4670	skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4671	err = -ENOMEM;
4672	if (!skb)
4673	goto failed;
4674	multicast = true;
4675	}
4676
4677	err = -EMSGSIZE;
4678	dh = genlmsg_put(skb, portid: 0, seq, family: &drbd_genl_family, flags: 0, cmd: DRBD_CONNECTION_STATE);
4679	if (!dh)
4680	goto nla_put_failure;
4681	dh->minor = -1U;
4682	dh->ret_code = NO_ERROR;
4683	if (nla_put_drbd_cfg_context(skb, resource: connection->resource, connection, NULL) ||
4684	nla_put_notification_header(msg: skb, type) ||
4685	((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4686	connection_info_to_skb(skb, s: connection_info, exclude_sensitive: true)))
4687	goto nla_put_failure;
4688	connection_statistics.conn_congested = test_bit(NET_CONGESTED, &connection->flags);
4689	connection_statistics_to_skb(skb, s: &connection_statistics, exclude_sensitive: !capable(CAP_SYS_ADMIN));
4690	genlmsg_end(skb, hdr: dh);
4691	if (multicast) {
4692	err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4693	/* skb has been consumed or freed in netlink_broadcast() */
4694	if (err && err != -ESRCH)
4695	goto failed;
4696	}
4697	return 0;
4698
4699	nla_put_failure:
4700	nlmsg_free(skb);
4701	failed:
4702	drbd_err(connection, "Error %d while broadcasting event. Event seq:%u\n",
4703	err, seq);
4704	return err;
4705	}
4706
4707	int notify_peer_device_state(struct sk_buff *skb,
4708	unsigned int seq,
4709	struct drbd_peer_device *peer_device,
4710	struct peer_device_info *peer_device_info,
4711	enum drbd_notification_type type)
4712	{
4713	struct peer_device_statistics peer_device_statistics;
4714	struct drbd_resource *resource = peer_device->device->resource;
4715	struct drbd_genlmsghdr *dh;
4716	bool multicast = false;
4717	int err;
4718
4719	if (!skb) {
4720	seq = atomic_inc_return(v: &notify_genl_seq);
4721	skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4722	err = -ENOMEM;
4723	if (!skb)
4724	goto failed;
4725	multicast = true;
4726	}
4727
4728	err = -EMSGSIZE;
4729	dh = genlmsg_put(skb, portid: 0, seq, family: &drbd_genl_family, flags: 0, cmd: DRBD_PEER_DEVICE_STATE);
4730	if (!dh)
4731	goto nla_put_failure;
4732	dh->minor = -1U;
4733	dh->ret_code = NO_ERROR;
4734	if (nla_put_drbd_cfg_context(skb, resource, connection: peer_device->connection, device: peer_device->device) ||
4735	nla_put_notification_header(msg: skb, type) ||
4736	((type & ~NOTIFY_FLAGS) != NOTIFY_DESTROY &&
4737	peer_device_info_to_skb(skb, s: peer_device_info, exclude_sensitive: true)))
4738	goto nla_put_failure;
4739	peer_device_to_statistics(s: &peer_device_statistics, peer_device);
4740	peer_device_statistics_to_skb(skb, s: &peer_device_statistics, exclude_sensitive: !capable(CAP_SYS_ADMIN));
4741	genlmsg_end(skb, hdr: dh);
4742	if (multicast) {
4743	err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4744	/* skb has been consumed or freed in netlink_broadcast() */
4745	if (err && err != -ESRCH)
4746	goto failed;
4747	}
4748	return 0;
4749
4750	nla_put_failure:
4751	nlmsg_free(skb);
4752	failed:
4753	drbd_err(peer_device, "Error %d while broadcasting event. Event seq:%u\n",
4754	err, seq);
4755	return err;
4756	}
4757
4758	void notify_helper(enum drbd_notification_type type,
4759	struct drbd_device *device, struct drbd_connection *connection,
4760	const char *name, int status)
4761	{
4762	struct drbd_resource *resource = device ? device->resource : connection->resource;
4763	struct drbd_helper_info helper_info;
4764	unsigned int seq = atomic_inc_return(v: &notify_genl_seq);
4765	struct sk_buff *skb = NULL;
4766	struct drbd_genlmsghdr *dh;
4767	int err;
4768
4769	strscpy(helper_info.helper_name, name, sizeof(helper_info.helper_name));
4770	helper_info.helper_name_len = min(strlen(name), sizeof(helper_info.helper_name));
4771	helper_info.helper_status = status;
4772
4773	skb = genlmsg_new(NLMSG_GOODSIZE, GFP_NOIO);
4774	err = -ENOMEM;
4775	if (!skb)
4776	goto fail;
4777
4778	err = -EMSGSIZE;
4779	dh = genlmsg_put(skb, portid: 0, seq, family: &drbd_genl_family, flags: 0, cmd: DRBD_HELPER);
4780	if (!dh)
4781	goto fail;
4782	dh->minor = device ? device->minor : -1;
4783	dh->ret_code = NO_ERROR;
4784	mutex_lock(&notification_mutex);
4785	if (nla_put_drbd_cfg_context(skb, resource, connection, device) ||
4786	nla_put_notification_header(msg: skb, type) ||
4787	drbd_helper_info_to_skb(skb, s: &helper_info, exclude_sensitive: true))
4788	goto unlock_fail;
4789	genlmsg_end(skb, hdr: dh);
4790	err = drbd_genl_multicast_events(skb, GFP_NOWAIT);
4791	skb = NULL;
4792	/* skb has been consumed or freed in netlink_broadcast() */
4793	if (err && err != -ESRCH)
4794	goto unlock_fail;
4795	mutex_unlock(lock: &notification_mutex);
4796	return;
4797
4798	unlock_fail:
4799	mutex_unlock(lock: &notification_mutex);
4800	fail:
4801	nlmsg_free(skb);
4802	drbd_err(resource, "Error %d while broadcasting event. Event seq:%u\n",
4803	err, seq);
4804	}
4805
4806	static int notify_initial_state_done(struct sk_buff *skb, unsigned int seq)
4807	{
4808	struct drbd_genlmsghdr *dh;
4809	int err;
4810
4811	err = -EMSGSIZE;
4812	dh = genlmsg_put(skb, portid: 0, seq, family: &drbd_genl_family, flags: 0, cmd: DRBD_INITIAL_STATE_DONE);
4813	if (!dh)
4814	goto nla_put_failure;
4815	dh->minor = -1U;
4816	dh->ret_code = NO_ERROR;
4817	if (nla_put_notification_header(msg: skb, type: NOTIFY_EXISTS))
4818	goto nla_put_failure;
4819	genlmsg_end(skb, hdr: dh);
4820	return 0;
4821
4822	nla_put_failure:
4823	nlmsg_free(skb);
4824	pr_err("Error %d sending event. Event seq:%u\n", err, seq);
4825	return err;
4826	}
4827
4828	static void free_state_changes(struct list_head *list)
4829	{
4830	while (!list_empty(head: list)) {
4831	struct drbd_state_change *state_change =
4832	list_first_entry(list, struct drbd_state_change, list);
4833	list_del(entry: &state_change->list);
4834	forget_state_change(state_change);
4835	}
4836	}
4837
4838	static unsigned int notifications_for_state_change(struct drbd_state_change *state_change)
4839	{
4840	return 1 +
4841	state_change->n_connections +
4842	state_change->n_devices +
4843	state_change->n_devices * state_change->n_connections;
4844	}
4845
4846	static int get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4847	{
4848	struct drbd_state_change *state_change = (struct drbd_state_change *)cb->args[0];
4849	unsigned int seq = cb->args[2];
4850	unsigned int n;
4851	enum drbd_notification_type flags = 0;
4852	int err = 0;
4853
4854	/* There is no need for taking notification_mutex here: it doesn't
4855	matter if the initial state events mix with later state chage
4856	events; we can always tell the events apart by the NOTIFY_EXISTS
4857	flag. */
4858
4859	cb->args[5]--;
4860	if (cb->args[5] == 1) {
4861	err = notify_initial_state_done(skb, seq);
4862	goto out;
4863	}
4864	n = cb->args[4]++;
4865	if (cb->args[4] < cb->args[3])
4866	flags |= NOTIFY_CONTINUES;
4867	if (n < 1) {
4868	err = notify_resource_state_change(skb, seq, state_change->resource,
4869	type: NOTIFY_EXISTS | flags);
4870	goto next;
4871	}
4872	n--;
4873	if (n < state_change->n_connections) {
4874	err = notify_connection_state_change(skb, seq, &state_change->connections[n],
4875	type: NOTIFY_EXISTS | flags);
4876	goto next;
4877	}
4878	n -= state_change->n_connections;
4879	if (n < state_change->n_devices) {
4880	err = notify_device_state_change(skb, seq, &state_change->devices[n],
4881	type: NOTIFY_EXISTS | flags);
4882	goto next;
4883	}
4884	n -= state_change->n_devices;
4885	if (n < state_change->n_devices * state_change->n_connections) {
4886	err = notify_peer_device_state_change(skb, seq, &state_change->peer_devices[n],
4887	type: NOTIFY_EXISTS | flags);
4888	goto next;
4889	}
4890
4891	next:
4892	if (cb->args[4] == cb->args[3]) {
4893	struct drbd_state_change *next_state_change =
4894	list_entry(state_change->list.next,
4895	struct drbd_state_change, list);
4896	cb->args[0] = (long)next_state_change;
4897	cb->args[3] = notifications_for_state_change(state_change: next_state_change);
4898	cb->args[4] = 0;
4899	}
4900	out:
4901	if (err)
4902	return err;
4903	else
4904	return skb->len;
4905	}
4906
4907	int drbd_adm_get_initial_state(struct sk_buff *skb, struct netlink_callback *cb)
4908	{
4909	struct drbd_resource *resource;
4910	LIST_HEAD(head);
4911
4912	if (cb->args[5] >= 1) {
4913	if (cb->args[5] > 1)
4914	return get_initial_state(skb, cb);
4915	if (cb->args[0]) {
4916	struct drbd_state_change *state_change =
4917	(struct drbd_state_change *)cb->args[0];
4918
4919	/* connect list to head */
4920	list_add(new: &head, head: &state_change->list);
4921	free_state_changes(list: &head);
4922	}
4923	return 0;
4924	}
4925
4926	cb->args[5] = 2; /* number of iterations */
4927	mutex_lock(&resources_mutex);
4928	for_each_resource(resource, &drbd_resources) {
4929	struct drbd_state_change *state_change;
4930
4931	state_change = remember_old_state(resource, GFP_KERNEL);
4932	if (!state_change) {
4933	if (!list_empty(head: &head))
4934	free_state_changes(list: &head);
4935	mutex_unlock(lock: &resources_mutex);
4936	return -ENOMEM;
4937	}
4938	copy_old_to_new_state_change(state_change);
4939	list_add_tail(new: &state_change->list, head: &head);
4940	cb->args[5] += notifications_for_state_change(state_change);
4941	}
4942	mutex_unlock(lock: &resources_mutex);
4943
4944	if (!list_empty(head: &head)) {
4945	struct drbd_state_change *state_change =
4946	list_entry(head.next, struct drbd_state_change, list);
4947	cb->args[0] = (long)state_change;
4948	cb->args[3] = notifications_for_state_change(state_change);
4949	list_del(entry: &head); /* detach list from head */
4950	}
4951
4952	cb->args[2] = cb->nlh->nlmsg_seq;
4953	return get_initial_state(skb, cb);
4954	}
4955