Merge branch 'for-3.17/drivers' of git://git.kernel.dk/linux-block
[sfrench/cifs-2.6.git] / drivers / block / drbd / drbd_nl.c
1 /*
2    drbd_nl.c
3
4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
5
6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
9
10    drbd is free software; you can redistribute it and/or modify
11    it under the terms of the GNU General Public License as published by
12    the Free Software Foundation; either version 2, or (at your option)
13    any later version.
14
15    drbd is distributed in the hope that it will be useful,
16    but WITHOUT ANY WARRANTY; without even the implied warranty of
17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18    GNU General Public License for more details.
19
20    You should have received a copy of the GNU General Public License
21    along with drbd; see the file COPYING.  If not, write to
22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
23
24  */
25
26 #define pr_fmt(fmt)     KBUILD_MODNAME ": " fmt
27
28 #include <linux/module.h>
29 #include <linux/drbd.h>
30 #include <linux/in.h>
31 #include <linux/fs.h>
32 #include <linux/file.h>
33 #include <linux/slab.h>
34 #include <linux/blkpg.h>
35 #include <linux/cpumask.h>
36 #include "drbd_int.h"
37 #include "drbd_protocol.h"
38 #include "drbd_req.h"
39 #include <asm/unaligned.h>
40 #include <linux/drbd_limits.h>
41 #include <linux/kthread.h>
42
43 #include <net/genetlink.h>
44
45 /* .doit */
46 // int drbd_adm_create_resource(struct sk_buff *skb, struct genl_info *info);
47 // int drbd_adm_delete_resource(struct sk_buff *skb, struct genl_info *info);
48
49 int drbd_adm_new_minor(struct sk_buff *skb, struct genl_info *info);
50 int drbd_adm_del_minor(struct sk_buff *skb, struct genl_info *info);
51
52 int drbd_adm_new_resource(struct sk_buff *skb, struct genl_info *info);
53 int drbd_adm_del_resource(struct sk_buff *skb, struct genl_info *info);
54 int drbd_adm_down(struct sk_buff *skb, struct genl_info *info);
55
56 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info);
57 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info);
58 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info);
59 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info);
60 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info);
61 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info);
62 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info);
63 int drbd_adm_start_ov(struct sk_buff *skb, struct genl_info *info);
64 int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info);
65 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info);
66 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info);
67 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info);
68 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info);
69 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info);
70 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info);
71 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info);
72 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info);
73 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info);
74 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info);
75 int drbd_adm_get_timeout_type(struct sk_buff *skb, struct genl_info *info);
76 /* .dumpit */
77 int drbd_adm_get_status_all(struct sk_buff *skb, struct netlink_callback *cb);
78
79 #include <linux/drbd_genl_api.h>
80 #include "drbd_nla.h"
81 #include <linux/genl_magic_func.h>
82
83 /* used blkdev_get_by_path, to claim our meta data device(s) */
84 static char *drbd_m_holder = "Hands off! this is DRBD's meta data device.";
85
86 static void drbd_adm_send_reply(struct sk_buff *skb, struct genl_info *info)
87 {
88         genlmsg_end(skb, genlmsg_data(nlmsg_data(nlmsg_hdr(skb))));
89         if (genlmsg_reply(skb, info))
90                 pr_err("error sending genl reply\n");
91 }
92
93 /* Used on a fresh "drbd_adm_prepare"d reply_skb, this cannot fail: The only
94  * reason it could fail was no space in skb, and there are 4k available. */
95 int drbd_msg_put_info(struct sk_buff *skb, const char *info)
96 {
97         struct nlattr *nla;
98         int err = -EMSGSIZE;
99
100         if (!info || !info[0])
101                 return 0;
102
103         nla = nla_nest_start(skb, DRBD_NLA_CFG_REPLY);
104         if (!nla)
105                 return err;
106
107         err = nla_put_string(skb, T_info_text, info);
108         if (err) {
109                 nla_nest_cancel(skb, nla);
110                 return err;
111         } else
112                 nla_nest_end(skb, nla);
113         return 0;
114 }
115
116 /* This would be a good candidate for a "pre_doit" hook,
117  * and per-family private info->pointers.
118  * But we need to stay compatible with older kernels.
119  * If it returns successfully, adm_ctx members are valid.
120  *
121  * At this point, we still rely on the global genl_lock().
122  * If we want to avoid that, and allow "genl_family.parallel_ops", we may need
123  * to add additional synchronization against object destruction/modification.
124  */
125 #define DRBD_ADM_NEED_MINOR     1
126 #define DRBD_ADM_NEED_RESOURCE  2
127 #define DRBD_ADM_NEED_CONNECTION 4
128 static int drbd_adm_prepare(struct drbd_config_context *adm_ctx,
129         struct sk_buff *skb, struct genl_info *info, unsigned flags)
130 {
131         struct drbd_genlmsghdr *d_in = info->userhdr;
132         const u8 cmd = info->genlhdr->cmd;
133         int err;
134
135         memset(adm_ctx, 0, sizeof(*adm_ctx));
136
137         /* genl_rcv_msg only checks for CAP_NET_ADMIN on "GENL_ADMIN_PERM" :( */
138         if (cmd != DRBD_ADM_GET_STATUS && !capable(CAP_NET_ADMIN))
139                return -EPERM;
140
141         adm_ctx->reply_skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
142         if (!adm_ctx->reply_skb) {
143                 err = -ENOMEM;
144                 goto fail;
145         }
146
147         adm_ctx->reply_dh = genlmsg_put_reply(adm_ctx->reply_skb,
148                                         info, &drbd_genl_family, 0, cmd);
149         /* put of a few bytes into a fresh skb of >= 4k will always succeed.
150          * but anyways */
151         if (!adm_ctx->reply_dh) {
152                 err = -ENOMEM;
153                 goto fail;
154         }
155
156         adm_ctx->reply_dh->minor = d_in->minor;
157         adm_ctx->reply_dh->ret_code = NO_ERROR;
158
159         adm_ctx->volume = VOLUME_UNSPECIFIED;
160         if (info->attrs[DRBD_NLA_CFG_CONTEXT]) {
161                 struct nlattr *nla;
162                 /* parse and validate only */
163                 err = drbd_cfg_context_from_attrs(NULL, info);
164                 if (err)
165                         goto fail;
166
167                 /* It was present, and valid,
168                  * copy it over to the reply skb. */
169                 err = nla_put_nohdr(adm_ctx->reply_skb,
170                                 info->attrs[DRBD_NLA_CFG_CONTEXT]->nla_len,
171                                 info->attrs[DRBD_NLA_CFG_CONTEXT]);
172                 if (err)
173                         goto fail;
174
175                 /* and assign stuff to the adm_ctx */
176                 nla = nested_attr_tb[__nla_type(T_ctx_volume)];
177                 if (nla)
178                         adm_ctx->volume = nla_get_u32(nla);
179                 nla = nested_attr_tb[__nla_type(T_ctx_resource_name)];
180                 if (nla)
181                         adm_ctx->resource_name = nla_data(nla);
182                 adm_ctx->my_addr = nested_attr_tb[__nla_type(T_ctx_my_addr)];
183                 adm_ctx->peer_addr = nested_attr_tb[__nla_type(T_ctx_peer_addr)];
184                 if ((adm_ctx->my_addr &&
185                      nla_len(adm_ctx->my_addr) > sizeof(adm_ctx->connection->my_addr)) ||
186                     (adm_ctx->peer_addr &&
187                      nla_len(adm_ctx->peer_addr) > sizeof(adm_ctx->connection->peer_addr))) {
188                         err = -EINVAL;
189                         goto fail;
190                 }
191         }
192
193         adm_ctx->minor = d_in->minor;
194         adm_ctx->device = minor_to_device(d_in->minor);
195
196         /* We are protected by the global genl_lock().
197          * But we may explicitly drop it/retake it in drbd_adm_set_role(),
198          * so make sure this object stays around. */
199         if (adm_ctx->device)
200                 kref_get(&adm_ctx->device->kref);
201
202         if (adm_ctx->resource_name) {
203                 adm_ctx->resource = drbd_find_resource(adm_ctx->resource_name);
204         }
205
206         if (!adm_ctx->device && (flags & DRBD_ADM_NEED_MINOR)) {
207                 drbd_msg_put_info(adm_ctx->reply_skb, "unknown minor");
208                 return ERR_MINOR_INVALID;
209         }
210         if (!adm_ctx->resource && (flags & DRBD_ADM_NEED_RESOURCE)) {
211                 drbd_msg_put_info(adm_ctx->reply_skb, "unknown resource");
212                 if (adm_ctx->resource_name)
213                         return ERR_RES_NOT_KNOWN;
214                 return ERR_INVALID_REQUEST;
215         }
216
217         if (flags & DRBD_ADM_NEED_CONNECTION) {
218                 if (adm_ctx->resource) {
219                         drbd_msg_put_info(adm_ctx->reply_skb, "no resource name expected");
220                         return ERR_INVALID_REQUEST;
221                 }
222                 if (adm_ctx->device) {
223                         drbd_msg_put_info(adm_ctx->reply_skb, "no minor number expected");
224                         return ERR_INVALID_REQUEST;
225                 }
226                 if (adm_ctx->my_addr && adm_ctx->peer_addr)
227                         adm_ctx->connection = conn_get_by_addrs(nla_data(adm_ctx->my_addr),
228                                                           nla_len(adm_ctx->my_addr),
229                                                           nla_data(adm_ctx->peer_addr),
230                                                           nla_len(adm_ctx->peer_addr));
231                 if (!adm_ctx->connection) {
232                         drbd_msg_put_info(adm_ctx->reply_skb, "unknown connection");
233                         return ERR_INVALID_REQUEST;
234                 }
235         }
236
237         /* some more paranoia, if the request was over-determined */
238         if (adm_ctx->device && adm_ctx->resource &&
239             adm_ctx->device->resource != adm_ctx->resource) {
240                 pr_warning("request: minor=%u, resource=%s; but that minor belongs to resource %s\n",
241                                 adm_ctx->minor, adm_ctx->resource->name,
242                                 adm_ctx->device->resource->name);
243                 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists in different resource");
244                 return ERR_INVALID_REQUEST;
245         }
246         if (adm_ctx->device &&
247             adm_ctx->volume != VOLUME_UNSPECIFIED &&
248             adm_ctx->volume != adm_ctx->device->vnr) {
249                 pr_warning("request: minor=%u, volume=%u; but that minor is volume %u in %s\n",
250                                 adm_ctx->minor, adm_ctx->volume,
251                                 adm_ctx->device->vnr,
252                                 adm_ctx->device->resource->name);
253                 drbd_msg_put_info(adm_ctx->reply_skb, "minor exists as different volume");
254                 return ERR_INVALID_REQUEST;
255         }
256
257         /* still, provide adm_ctx->resource always, if possible. */
258         if (!adm_ctx->resource) {
259                 adm_ctx->resource = adm_ctx->device ? adm_ctx->device->resource
260                         : adm_ctx->connection ? adm_ctx->connection->resource : NULL;
261                 if (adm_ctx->resource)
262                         kref_get(&adm_ctx->resource->kref);
263         }
264
265         return NO_ERROR;
266
267 fail:
268         nlmsg_free(adm_ctx->reply_skb);
269         adm_ctx->reply_skb = NULL;
270         return err;
271 }
272
273 static int drbd_adm_finish(struct drbd_config_context *adm_ctx,
274         struct genl_info *info, int retcode)
275 {
276         if (adm_ctx->device) {
277                 kref_put(&adm_ctx->device->kref, drbd_destroy_device);
278                 adm_ctx->device = NULL;
279         }
280         if (adm_ctx->connection) {
281                 kref_put(&adm_ctx->connection->kref, &drbd_destroy_connection);
282                 adm_ctx->connection = NULL;
283         }
284         if (adm_ctx->resource) {
285                 kref_put(&adm_ctx->resource->kref, drbd_destroy_resource);
286                 adm_ctx->resource = NULL;
287         }
288
289         if (!adm_ctx->reply_skb)
290                 return -ENOMEM;
291
292         adm_ctx->reply_dh->ret_code = retcode;
293         drbd_adm_send_reply(adm_ctx->reply_skb, info);
294         return 0;
295 }
296
297 static void setup_khelper_env(struct drbd_connection *connection, char **envp)
298 {
299         char *afs;
300
301         /* FIXME: A future version will not allow this case. */
302         if (connection->my_addr_len == 0 || connection->peer_addr_len == 0)
303                 return;
304
305         switch (((struct sockaddr *)&connection->peer_addr)->sa_family) {
306         case AF_INET6:
307                 afs = "ipv6";
308                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI6",
309                          &((struct sockaddr_in6 *)&connection->peer_addr)->sin6_addr);
310                 break;
311         case AF_INET:
312                 afs = "ipv4";
313                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
314                          &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
315                 break;
316         default:
317                 afs = "ssocks";
318                 snprintf(envp[4], 60, "DRBD_PEER_ADDRESS=%pI4",
319                          &((struct sockaddr_in *)&connection->peer_addr)->sin_addr);
320         }
321         snprintf(envp[3], 20, "DRBD_PEER_AF=%s", afs);
322 }
323
324 int drbd_khelper(struct drbd_device *device, char *cmd)
325 {
326         char *envp[] = { "HOME=/",
327                         "TERM=linux",
328                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
329                          (char[20]) { }, /* address family */
330                          (char[60]) { }, /* address */
331                         NULL };
332         char mb[12];
333         char *argv[] = {usermode_helper, cmd, mb, NULL };
334         struct drbd_connection *connection = first_peer_device(device)->connection;
335         struct sib_info sib;
336         int ret;
337
338         if (current == connection->worker.task)
339                 set_bit(CALLBACK_PENDING, &connection->flags);
340
341         snprintf(mb, 12, "minor-%d", device_to_minor(device));
342         setup_khelper_env(connection, envp);
343
344         /* The helper may take some time.
345          * write out any unsynced meta data changes now */
346         drbd_md_sync(device);
347
348         drbd_info(device, "helper command: %s %s %s\n", usermode_helper, cmd, mb);
349         sib.sib_reason = SIB_HELPER_PRE;
350         sib.helper_name = cmd;
351         drbd_bcast_event(device, &sib);
352         ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
353         if (ret)
354                 drbd_warn(device, "helper command: %s %s %s exit code %u (0x%x)\n",
355                                 usermode_helper, cmd, mb,
356                                 (ret >> 8) & 0xff, ret);
357         else
358                 drbd_info(device, "helper command: %s %s %s exit code %u (0x%x)\n",
359                                 usermode_helper, cmd, mb,
360                                 (ret >> 8) & 0xff, ret);
361         sib.sib_reason = SIB_HELPER_POST;
362         sib.helper_exit_code = ret;
363         drbd_bcast_event(device, &sib);
364
365         if (current == connection->worker.task)
366                 clear_bit(CALLBACK_PENDING, &connection->flags);
367
368         if (ret < 0) /* Ignore any ERRNOs we got. */
369                 ret = 0;
370
371         return ret;
372 }
373
374 static int conn_khelper(struct drbd_connection *connection, char *cmd)
375 {
376         char *envp[] = { "HOME=/",
377                         "TERM=linux",
378                         "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
379                          (char[20]) { }, /* address family */
380                          (char[60]) { }, /* address */
381                         NULL };
382         char *resource_name = connection->resource->name;
383         char *argv[] = {usermode_helper, cmd, resource_name, NULL };
384         int ret;
385
386         setup_khelper_env(connection, envp);
387         conn_md_sync(connection);
388
389         drbd_info(connection, "helper command: %s %s %s\n", usermode_helper, cmd, resource_name);
390         /* TODO: conn_bcast_event() ?? */
391
392         ret = call_usermodehelper(usermode_helper, argv, envp, UMH_WAIT_PROC);
393         if (ret)
394                 drbd_warn(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
395                           usermode_helper, cmd, resource_name,
396                           (ret >> 8) & 0xff, ret);
397         else
398                 drbd_info(connection, "helper command: %s %s %s exit code %u (0x%x)\n",
399                           usermode_helper, cmd, resource_name,
400                           (ret >> 8) & 0xff, ret);
401         /* TODO: conn_bcast_event() ?? */
402
403         if (ret < 0) /* Ignore any ERRNOs we got. */
404                 ret = 0;
405
406         return ret;
407 }
408
409 static enum drbd_fencing_p highest_fencing_policy(struct drbd_connection *connection)
410 {
411         enum drbd_fencing_p fp = FP_NOT_AVAIL;
412         struct drbd_peer_device *peer_device;
413         int vnr;
414
415         rcu_read_lock();
416         idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
417                 struct drbd_device *device = peer_device->device;
418                 if (get_ldev_if_state(device, D_CONSISTENT)) {
419                         struct disk_conf *disk_conf =
420                                 rcu_dereference(peer_device->device->ldev->disk_conf);
421                         fp = max_t(enum drbd_fencing_p, fp, disk_conf->fencing);
422                         put_ldev(device);
423                 }
424         }
425         rcu_read_unlock();
426
427         if (fp == FP_NOT_AVAIL) {
428                 /* IO Suspending works on the whole resource.
429                    Do it only for one device. */
430                 vnr = 0;
431                 peer_device = idr_get_next(&connection->peer_devices, &vnr);
432                 drbd_change_state(peer_device->device, CS_VERBOSE | CS_HARD, NS(susp_fen, 0));
433         }
434
435         return fp;
436 }
437
438 bool conn_try_outdate_peer(struct drbd_connection *connection)
439 {
440         unsigned int connect_cnt;
441         union drbd_state mask = { };
442         union drbd_state val = { };
443         enum drbd_fencing_p fp;
444         char *ex_to_string;
445         int r;
446
447         spin_lock_irq(&connection->resource->req_lock);
448         if (connection->cstate >= C_WF_REPORT_PARAMS) {
449                 drbd_err(connection, "Expected cstate < C_WF_REPORT_PARAMS\n");
450                 spin_unlock_irq(&connection->resource->req_lock);
451                 return false;
452         }
453
454         connect_cnt = connection->connect_cnt;
455         spin_unlock_irq(&connection->resource->req_lock);
456
457         fp = highest_fencing_policy(connection);
458         switch (fp) {
459         case FP_NOT_AVAIL:
460                 drbd_warn(connection, "Not fencing peer, I'm not even Consistent myself.\n");
461                 goto out;
462         case FP_DONT_CARE:
463                 return true;
464         default: ;
465         }
466
467         r = conn_khelper(connection, "fence-peer");
468
469         switch ((r>>8) & 0xff) {
470         case 3: /* peer is inconsistent */
471                 ex_to_string = "peer is inconsistent or worse";
472                 mask.pdsk = D_MASK;
473                 val.pdsk = D_INCONSISTENT;
474                 break;
475         case 4: /* peer got outdated, or was already outdated */
476                 ex_to_string = "peer was fenced";
477                 mask.pdsk = D_MASK;
478                 val.pdsk = D_OUTDATED;
479                 break;
480         case 5: /* peer was down */
481                 if (conn_highest_disk(connection) == D_UP_TO_DATE) {
482                         /* we will(have) create(d) a new UUID anyways... */
483                         ex_to_string = "peer is unreachable, assumed to be dead";
484                         mask.pdsk = D_MASK;
485                         val.pdsk = D_OUTDATED;
486                 } else {
487                         ex_to_string = "peer unreachable, doing nothing since disk != UpToDate";
488                 }
489                 break;
490         case 6: /* Peer is primary, voluntarily outdate myself.
491                  * This is useful when an unconnected R_SECONDARY is asked to
492                  * become R_PRIMARY, but finds the other peer being active. */
493                 ex_to_string = "peer is active";
494                 drbd_warn(connection, "Peer is primary, outdating myself.\n");
495                 mask.disk = D_MASK;
496                 val.disk = D_OUTDATED;
497                 break;
498         case 7:
499                 if (fp != FP_STONITH)
500                         drbd_err(connection, "fence-peer() = 7 && fencing != Stonith !!!\n");
501                 ex_to_string = "peer was stonithed";
502                 mask.pdsk = D_MASK;
503                 val.pdsk = D_OUTDATED;
504                 break;
505         default:
506                 /* The script is broken ... */
507                 drbd_err(connection, "fence-peer helper broken, returned %d\n", (r>>8)&0xff);
508                 return false; /* Eventually leave IO frozen */
509         }
510
511         drbd_info(connection, "fence-peer helper returned %d (%s)\n",
512                   (r>>8) & 0xff, ex_to_string);
513
514  out:
515
516         /* Not using
517            conn_request_state(connection, mask, val, CS_VERBOSE);
518            here, because we might were able to re-establish the connection in the
519            meantime. */
520         spin_lock_irq(&connection->resource->req_lock);
521         if (connection->cstate < C_WF_REPORT_PARAMS && !test_bit(STATE_SENT, &connection->flags)) {
522                 if (connection->connect_cnt != connect_cnt)
523                         /* In case the connection was established and droped
524                            while the fence-peer handler was running, ignore it */
525                         drbd_info(connection, "Ignoring fence-peer exit code\n");
526                 else
527                         _conn_request_state(connection, mask, val, CS_VERBOSE);
528         }
529         spin_unlock_irq(&connection->resource->req_lock);
530
531         return conn_highest_pdsk(connection) <= D_OUTDATED;
532 }
533
534 static int _try_outdate_peer_async(void *data)
535 {
536         struct drbd_connection *connection = (struct drbd_connection *)data;
537
538         conn_try_outdate_peer(connection);
539
540         kref_put(&connection->kref, drbd_destroy_connection);
541         return 0;
542 }
543
544 void conn_try_outdate_peer_async(struct drbd_connection *connection)
545 {
546         struct task_struct *opa;
547
548         kref_get(&connection->kref);
549         /* We may just have force_sig()'ed this thread
550          * to get it out of some blocking network function.
551          * Clear signals; otherwise kthread_run(), which internally uses
552          * wait_on_completion_killable(), will mistake our pending signal
553          * for a new fatal signal and fail. */
554         flush_signals(current);
555         opa = kthread_run(_try_outdate_peer_async, connection, "drbd_async_h");
556         if (IS_ERR(opa)) {
557                 drbd_err(connection, "out of mem, failed to invoke fence-peer helper\n");
558                 kref_put(&connection->kref, drbd_destroy_connection);
559         }
560 }
561
562 enum drbd_state_rv
563 drbd_set_role(struct drbd_device *const device, enum drbd_role new_role, int force)
564 {
565         struct drbd_peer_device *const peer_device = first_peer_device(device);
566         struct drbd_connection *const connection = peer_device ? peer_device->connection : NULL;
567         const int max_tries = 4;
568         enum drbd_state_rv rv = SS_UNKNOWN_ERROR;
569         struct net_conf *nc;
570         int try = 0;
571         int forced = 0;
572         union drbd_state mask, val;
573
574         if (new_role == R_PRIMARY) {
575                 struct drbd_connection *connection;
576
577                 /* Detect dead peers as soon as possible.  */
578
579                 rcu_read_lock();
580                 for_each_connection(connection, device->resource)
581                         request_ping(connection);
582                 rcu_read_unlock();
583         }
584
585         mutex_lock(device->state_mutex);
586
587         mask.i = 0; mask.role = R_MASK;
588         val.i  = 0; val.role  = new_role;
589
590         while (try++ < max_tries) {
591                 rv = _drbd_request_state(device, mask, val, CS_WAIT_COMPLETE);
592
593                 /* in case we first succeeded to outdate,
594                  * but now suddenly could establish a connection */
595                 if (rv == SS_CW_FAILED_BY_PEER && mask.pdsk != 0) {
596                         val.pdsk = 0;
597                         mask.pdsk = 0;
598                         continue;
599                 }
600
601                 if (rv == SS_NO_UP_TO_DATE_DISK && force &&
602                     (device->state.disk < D_UP_TO_DATE &&
603                      device->state.disk >= D_INCONSISTENT)) {
604                         mask.disk = D_MASK;
605                         val.disk  = D_UP_TO_DATE;
606                         forced = 1;
607                         continue;
608                 }
609
610                 if (rv == SS_NO_UP_TO_DATE_DISK &&
611                     device->state.disk == D_CONSISTENT && mask.pdsk == 0) {
612                         D_ASSERT(device, device->state.pdsk == D_UNKNOWN);
613
614                         if (conn_try_outdate_peer(connection)) {
615                                 val.disk = D_UP_TO_DATE;
616                                 mask.disk = D_MASK;
617                         }
618                         continue;
619                 }
620
621                 if (rv == SS_NOTHING_TO_DO)
622                         goto out;
623                 if (rv == SS_PRIMARY_NOP && mask.pdsk == 0) {
624                         if (!conn_try_outdate_peer(connection) && force) {
625                                 drbd_warn(device, "Forced into split brain situation!\n");
626                                 mask.pdsk = D_MASK;
627                                 val.pdsk  = D_OUTDATED;
628
629                         }
630                         continue;
631                 }
632                 if (rv == SS_TWO_PRIMARIES) {
633                         /* Maybe the peer is detected as dead very soon...
634                            retry at most once more in this case. */
635                         int timeo;
636                         rcu_read_lock();
637                         nc = rcu_dereference(connection->net_conf);
638                         timeo = nc ? (nc->ping_timeo + 1) * HZ / 10 : 1;
639                         rcu_read_unlock();
640                         schedule_timeout_interruptible(timeo);
641                         if (try < max_tries)
642                                 try = max_tries - 1;
643                         continue;
644                 }
645                 if (rv < SS_SUCCESS) {
646                         rv = _drbd_request_state(device, mask, val,
647                                                 CS_VERBOSE + CS_WAIT_COMPLETE);
648                         if (rv < SS_SUCCESS)
649                                 goto out;
650                 }
651                 break;
652         }
653
654         if (rv < SS_SUCCESS)
655                 goto out;
656
657         if (forced)
658                 drbd_warn(device, "Forced to consider local data as UpToDate!\n");
659
660         /* Wait until nothing is on the fly :) */
661         wait_event(device->misc_wait, atomic_read(&device->ap_pending_cnt) == 0);
662
663         /* FIXME also wait for all pending P_BARRIER_ACK? */
664
665         if (new_role == R_SECONDARY) {
666                 if (get_ldev(device)) {
667                         device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
668                         put_ldev(device);
669                 }
670         } else {
671                 mutex_lock(&device->resource->conf_update);
672                 nc = connection->net_conf;
673                 if (nc)
674                         nc->discard_my_data = 0; /* without copy; single bit op is atomic */
675                 mutex_unlock(&device->resource->conf_update);
676
677                 if (get_ldev(device)) {
678                         if (((device->state.conn < C_CONNECTED ||
679                                device->state.pdsk <= D_FAILED)
680                               && device->ldev->md.uuid[UI_BITMAP] == 0) || forced)
681                                 drbd_uuid_new_current(device);
682
683                         device->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
684                         put_ldev(device);
685                 }
686         }
687
688         /* writeout of activity log covered areas of the bitmap
689          * to stable storage done in after state change already */
690
691         if (device->state.conn >= C_WF_REPORT_PARAMS) {
692                 /* if this was forced, we should consider sync */
693                 if (forced)
694                         drbd_send_uuids(peer_device);
695                 drbd_send_current_state(peer_device);
696         }
697
698         drbd_md_sync(device);
699         set_disk_ro(device->vdisk, new_role == R_SECONDARY);
700         kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
701 out:
702         mutex_unlock(device->state_mutex);
703         return rv;
704 }
705
706 static const char *from_attrs_err_to_txt(int err)
707 {
708         return  err == -ENOMSG ? "required attribute missing" :
709                 err == -EOPNOTSUPP ? "unknown mandatory attribute" :
710                 err == -EEXIST ? "can not change invariant setting" :
711                 "invalid attribute value";
712 }
713
714 int drbd_adm_set_role(struct sk_buff *skb, struct genl_info *info)
715 {
716         struct drbd_config_context adm_ctx;
717         struct set_role_parms parms;
718         int err;
719         enum drbd_ret_code retcode;
720
721         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
722         if (!adm_ctx.reply_skb)
723                 return retcode;
724         if (retcode != NO_ERROR)
725                 goto out;
726
727         memset(&parms, 0, sizeof(parms));
728         if (info->attrs[DRBD_NLA_SET_ROLE_PARMS]) {
729                 err = set_role_parms_from_attrs(&parms, info);
730                 if (err) {
731                         retcode = ERR_MANDATORY_TAG;
732                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
733                         goto out;
734                 }
735         }
736         genl_unlock();
737         mutex_lock(&adm_ctx.resource->adm_mutex);
738
739         if (info->genlhdr->cmd == DRBD_ADM_PRIMARY)
740                 retcode = drbd_set_role(adm_ctx.device, R_PRIMARY, parms.assume_uptodate);
741         else
742                 retcode = drbd_set_role(adm_ctx.device, R_SECONDARY, 0);
743
744         mutex_unlock(&adm_ctx.resource->adm_mutex);
745         genl_lock();
746 out:
747         drbd_adm_finish(&adm_ctx, info, retcode);
748         return 0;
749 }
750
751 /* Initializes the md.*_offset members, so we are able to find
752  * the on disk meta data.
753  *
754  * We currently have two possible layouts:
755  * external:
756  *   |----------- md_size_sect ------------------|
757  *   [ 4k superblock ][ activity log ][  Bitmap  ]
758  *   | al_offset == 8 |
759  *   | bm_offset = al_offset + X      |
760  *  ==> bitmap sectors = md_size_sect - bm_offset
761  *
762  * internal:
763  *            |----------- md_size_sect ------------------|
764  * [data.....][  Bitmap  ][ activity log ][ 4k superblock ]
765  *                        | al_offset < 0 |
766  *            | bm_offset = al_offset - Y |
767  *  ==> bitmap sectors = Y = al_offset - bm_offset
768  *
769  *  Activity log size used to be fixed 32kB,
770  *  but is about to become configurable.
771  */
772 static void drbd_md_set_sector_offsets(struct drbd_device *device,
773                                        struct drbd_backing_dev *bdev)
774 {
775         sector_t md_size_sect = 0;
776         unsigned int al_size_sect = bdev->md.al_size_4k * 8;
777
778         bdev->md.md_offset = drbd_md_ss(bdev);
779
780         switch (bdev->md.meta_dev_idx) {
781         default:
782                 /* v07 style fixed size indexed meta data */
783                 bdev->md.md_size_sect = MD_128MB_SECT;
784                 bdev->md.al_offset = MD_4kB_SECT;
785                 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
786                 break;
787         case DRBD_MD_INDEX_FLEX_EXT:
788                 /* just occupy the full device; unit: sectors */
789                 bdev->md.md_size_sect = drbd_get_capacity(bdev->md_bdev);
790                 bdev->md.al_offset = MD_4kB_SECT;
791                 bdev->md.bm_offset = MD_4kB_SECT + al_size_sect;
792                 break;
793         case DRBD_MD_INDEX_INTERNAL:
794         case DRBD_MD_INDEX_FLEX_INT:
795                 /* al size is still fixed */
796                 bdev->md.al_offset = -al_size_sect;
797                 /* we need (slightly less than) ~ this much bitmap sectors: */
798                 md_size_sect = drbd_get_capacity(bdev->backing_bdev);
799                 md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
800                 md_size_sect = BM_SECT_TO_EXT(md_size_sect);
801                 md_size_sect = ALIGN(md_size_sect, 8);
802
803                 /* plus the "drbd meta data super block",
804                  * and the activity log; */
805                 md_size_sect += MD_4kB_SECT + al_size_sect;
806
807                 bdev->md.md_size_sect = md_size_sect;
808                 /* bitmap offset is adjusted by 'super' block size */
809                 bdev->md.bm_offset   = -md_size_sect + MD_4kB_SECT;
810                 break;
811         }
812 }
813
814 /* input size is expected to be in KB */
815 char *ppsize(char *buf, unsigned long long size)
816 {
817         /* Needs 9 bytes at max including trailing NUL:
818          * -1ULL ==> "16384 EB" */
819         static char units[] = { 'K', 'M', 'G', 'T', 'P', 'E' };
820         int base = 0;
821         while (size >= 10000 && base < sizeof(units)-1) {
822                 /* shift + round */
823                 size = (size >> 10) + !!(size & (1<<9));
824                 base++;
825         }
826         sprintf(buf, "%u %cB", (unsigned)size, units[base]);
827
828         return buf;
829 }
830
831 /* there is still a theoretical deadlock when called from receiver
832  * on an D_INCONSISTENT R_PRIMARY:
833  *  remote READ does inc_ap_bio, receiver would need to receive answer
834  *  packet from remote to dec_ap_bio again.
835  *  receiver receive_sizes(), comes here,
836  *  waits for ap_bio_cnt == 0. -> deadlock.
837  * but this cannot happen, actually, because:
838  *  R_PRIMARY D_INCONSISTENT, and peer's disk is unreachable
839  *  (not connected, or bad/no disk on peer):
840  *  see drbd_fail_request_early, ap_bio_cnt is zero.
841  *  R_PRIMARY D_INCONSISTENT, and C_SYNC_TARGET:
842  *  peer may not initiate a resize.
843  */
844 /* Note these are not to be confused with
845  * drbd_adm_suspend_io/drbd_adm_resume_io,
846  * which are (sub) state changes triggered by admin (drbdsetup),
847  * and can be long lived.
848  * This changes an device->flag, is triggered by drbd internals,
849  * and should be short-lived. */
850 void drbd_suspend_io(struct drbd_device *device)
851 {
852         set_bit(SUSPEND_IO, &device->flags);
853         if (drbd_suspended(device))
854                 return;
855         wait_event(device->misc_wait, !atomic_read(&device->ap_bio_cnt));
856 }
857
858 void drbd_resume_io(struct drbd_device *device)
859 {
860         clear_bit(SUSPEND_IO, &device->flags);
861         wake_up(&device->misc_wait);
862 }
863
864 /**
865  * drbd_determine_dev_size() -  Sets the right device size obeying all constraints
866  * @device:     DRBD device.
867  *
868  * Returns 0 on success, negative return values indicate errors.
869  * You should call drbd_md_sync() after calling this function.
870  */
871 enum determine_dev_size
872 drbd_determine_dev_size(struct drbd_device *device, enum dds_flags flags, struct resize_parms *rs) __must_hold(local)
873 {
874         sector_t prev_first_sect, prev_size; /* previous meta location */
875         sector_t la_size_sect, u_size;
876         struct drbd_md *md = &device->ldev->md;
877         u32 prev_al_stripe_size_4k;
878         u32 prev_al_stripes;
879         sector_t size;
880         char ppb[10];
881         void *buffer;
882
883         int md_moved, la_size_changed;
884         enum determine_dev_size rv = DS_UNCHANGED;
885
886         /* race:
887          * application request passes inc_ap_bio,
888          * but then cannot get an AL-reference.
889          * this function later may wait on ap_bio_cnt == 0. -> deadlock.
890          *
891          * to avoid that:
892          * Suspend IO right here.
893          * still lock the act_log to not trigger ASSERTs there.
894          */
895         drbd_suspend_io(device);
896         buffer = drbd_md_get_buffer(device, __func__); /* Lock meta-data IO */
897         if (!buffer) {
898                 drbd_resume_io(device);
899                 return DS_ERROR;
900         }
901
902         /* no wait necessary anymore, actually we could assert that */
903         wait_event(device->al_wait, lc_try_lock(device->act_log));
904
905         prev_first_sect = drbd_md_first_sector(device->ldev);
906         prev_size = device->ldev->md.md_size_sect;
907         la_size_sect = device->ldev->md.la_size_sect;
908
909         if (rs) {
910                 /* rs is non NULL if we should change the AL layout only */
911
912                 prev_al_stripes = md->al_stripes;
913                 prev_al_stripe_size_4k = md->al_stripe_size_4k;
914
915                 md->al_stripes = rs->al_stripes;
916                 md->al_stripe_size_4k = rs->al_stripe_size / 4;
917                 md->al_size_4k = (u64)rs->al_stripes * rs->al_stripe_size / 4;
918         }
919
920         drbd_md_set_sector_offsets(device, device->ldev);
921
922         rcu_read_lock();
923         u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
924         rcu_read_unlock();
925         size = drbd_new_dev_size(device, device->ldev, u_size, flags & DDSF_FORCED);
926
927         if (size < la_size_sect) {
928                 if (rs && u_size == 0) {
929                         /* Remove "rs &&" later. This check should always be active, but
930                            right now the receiver expects the permissive behavior */
931                         drbd_warn(device, "Implicit shrink not allowed. "
932                                  "Use --size=%llus for explicit shrink.\n",
933                                  (unsigned long long)size);
934                         rv = DS_ERROR_SHRINK;
935                 }
936                 if (u_size > size)
937                         rv = DS_ERROR_SPACE_MD;
938                 if (rv != DS_UNCHANGED)
939                         goto err_out;
940         }
941
942         if (drbd_get_capacity(device->this_bdev) != size ||
943             drbd_bm_capacity(device) != size) {
944                 int err;
945                 err = drbd_bm_resize(device, size, !(flags & DDSF_NO_RESYNC));
946                 if (unlikely(err)) {
947                         /* currently there is only one error: ENOMEM! */
948                         size = drbd_bm_capacity(device)>>1;
949                         if (size == 0) {
950                                 drbd_err(device, "OUT OF MEMORY! "
951                                     "Could not allocate bitmap!\n");
952                         } else {
953                                 drbd_err(device, "BM resizing failed. "
954                                     "Leaving size unchanged at size = %lu KB\n",
955                                     (unsigned long)size);
956                         }
957                         rv = DS_ERROR;
958                 }
959                 /* racy, see comments above. */
960                 drbd_set_my_capacity(device, size);
961                 device->ldev->md.la_size_sect = size;
962                 drbd_info(device, "size = %s (%llu KB)\n", ppsize(ppb, size>>1),
963                      (unsigned long long)size>>1);
964         }
965         if (rv <= DS_ERROR)
966                 goto err_out;
967
968         la_size_changed = (la_size_sect != device->ldev->md.la_size_sect);
969
970         md_moved = prev_first_sect != drbd_md_first_sector(device->ldev)
971                 || prev_size       != device->ldev->md.md_size_sect;
972
973         if (la_size_changed || md_moved || rs) {
974                 u32 prev_flags;
975
976                 /* We do some synchronous IO below, which may take some time.
977                  * Clear the timer, to avoid scary "timer expired!" messages,
978                  * "Superblock" is written out at least twice below, anyways. */
979                 del_timer(&device->md_sync_timer);
980                 drbd_al_shrink(device); /* All extents inactive. */
981
982                 prev_flags = md->flags;
983                 md->flags &= ~MDF_PRIMARY_IND;
984                 drbd_md_write(device, buffer);
985
986                 drbd_info(device, "Writing the whole bitmap, %s\n",
987                          la_size_changed && md_moved ? "size changed and md moved" :
988                          la_size_changed ? "size changed" : "md moved");
989                 /* next line implicitly does drbd_suspend_io()+drbd_resume_io() */
990                 drbd_bitmap_io(device, md_moved ? &drbd_bm_write_all : &drbd_bm_write,
991                                "size changed", BM_LOCKED_MASK);
992                 drbd_initialize_al(device, buffer);
993
994                 md->flags = prev_flags;
995                 drbd_md_write(device, buffer);
996
997                 if (rs)
998                         drbd_info(device, "Changed AL layout to al-stripes = %d, al-stripe-size-kB = %d\n",
999                                   md->al_stripes, md->al_stripe_size_4k * 4);
1000         }
1001
1002         if (size > la_size_sect)
1003                 rv = la_size_sect ? DS_GREW : DS_GREW_FROM_ZERO;
1004         if (size < la_size_sect)
1005                 rv = DS_SHRUNK;
1006
1007         if (0) {
1008         err_out:
1009                 if (rs) {
1010                         md->al_stripes = prev_al_stripes;
1011                         md->al_stripe_size_4k = prev_al_stripe_size_4k;
1012                         md->al_size_4k = (u64)prev_al_stripes * prev_al_stripe_size_4k;
1013
1014                         drbd_md_set_sector_offsets(device, device->ldev);
1015                 }
1016         }
1017         lc_unlock(device->act_log);
1018         wake_up(&device->al_wait);
1019         drbd_md_put_buffer(device);
1020         drbd_resume_io(device);
1021
1022         return rv;
1023 }
1024
1025 sector_t
1026 drbd_new_dev_size(struct drbd_device *device, struct drbd_backing_dev *bdev,
1027                   sector_t u_size, int assume_peer_has_space)
1028 {
1029         sector_t p_size = device->p_size;   /* partner's disk size. */
1030         sector_t la_size_sect = bdev->md.la_size_sect; /* last agreed size. */
1031         sector_t m_size; /* my size */
1032         sector_t size = 0;
1033
1034         m_size = drbd_get_max_capacity(bdev);
1035
1036         if (device->state.conn < C_CONNECTED && assume_peer_has_space) {
1037                 drbd_warn(device, "Resize while not connected was forced by the user!\n");
1038                 p_size = m_size;
1039         }
1040
1041         if (p_size && m_size) {
1042                 size = min_t(sector_t, p_size, m_size);
1043         } else {
1044                 if (la_size_sect) {
1045                         size = la_size_sect;
1046                         if (m_size && m_size < size)
1047                                 size = m_size;
1048                         if (p_size && p_size < size)
1049                                 size = p_size;
1050                 } else {
1051                         if (m_size)
1052                                 size = m_size;
1053                         if (p_size)
1054                                 size = p_size;
1055                 }
1056         }
1057
1058         if (size == 0)
1059                 drbd_err(device, "Both nodes diskless!\n");
1060
1061         if (u_size) {
1062                 if (u_size > size)
1063                         drbd_err(device, "Requested disk size is too big (%lu > %lu)\n",
1064                             (unsigned long)u_size>>1, (unsigned long)size>>1);
1065                 else
1066                         size = u_size;
1067         }
1068
1069         return size;
1070 }
1071
1072 /**
1073  * drbd_check_al_size() - Ensures that the AL is of the right size
1074  * @device:     DRBD device.
1075  *
1076  * Returns -EBUSY if current al lru is still used, -ENOMEM when allocation
1077  * failed, and 0 on success. You should call drbd_md_sync() after you called
1078  * this function.
1079  */
1080 static int drbd_check_al_size(struct drbd_device *device, struct disk_conf *dc)
1081 {
1082         struct lru_cache *n, *t;
1083         struct lc_element *e;
1084         unsigned int in_use;
1085         int i;
1086
1087         if (device->act_log &&
1088             device->act_log->nr_elements == dc->al_extents)
1089                 return 0;
1090
1091         in_use = 0;
1092         t = device->act_log;
1093         n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
1094                 dc->al_extents, sizeof(struct lc_element), 0);
1095
1096         if (n == NULL) {
1097                 drbd_err(device, "Cannot allocate act_log lru!\n");
1098                 return -ENOMEM;
1099         }
1100         spin_lock_irq(&device->al_lock);
1101         if (t) {
1102                 for (i = 0; i < t->nr_elements; i++) {
1103                         e = lc_element_by_index(t, i);
1104                         if (e->refcnt)
1105                                 drbd_err(device, "refcnt(%d)==%d\n",
1106                                     e->lc_number, e->refcnt);
1107                         in_use += e->refcnt;
1108                 }
1109         }
1110         if (!in_use)
1111                 device->act_log = n;
1112         spin_unlock_irq(&device->al_lock);
1113         if (in_use) {
1114                 drbd_err(device, "Activity log still in use!\n");
1115                 lc_destroy(n);
1116                 return -EBUSY;
1117         } else {
1118                 if (t)
1119                         lc_destroy(t);
1120         }
1121         drbd_md_mark_dirty(device); /* we changed device->act_log->nr_elemens */
1122         return 0;
1123 }
1124
1125 static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backing_dev *bdev,
1126                                    unsigned int max_bio_size)
1127 {
1128         struct request_queue * const q = device->rq_queue;
1129         unsigned int max_hw_sectors = max_bio_size >> 9;
1130         unsigned int max_segments = 0;
1131         struct request_queue *b = NULL;
1132
1133         if (bdev) {
1134                 b = bdev->backing_bdev->bd_disk->queue;
1135
1136                 max_hw_sectors = min(queue_max_hw_sectors(b), max_bio_size >> 9);
1137                 rcu_read_lock();
1138                 max_segments = rcu_dereference(device->ldev->disk_conf)->max_bio_bvecs;
1139                 rcu_read_unlock();
1140
1141                 blk_set_stacking_limits(&q->limits);
1142                 blk_queue_max_write_same_sectors(q, 0);
1143         }
1144
1145         blk_queue_logical_block_size(q, 512);
1146         blk_queue_max_hw_sectors(q, max_hw_sectors);
1147         /* This is the workaround for "bio would need to, but cannot, be split" */
1148         blk_queue_max_segments(q, max_segments ? max_segments : BLK_MAX_SEGMENTS);
1149         blk_queue_segment_boundary(q, PAGE_CACHE_SIZE-1);
1150
1151         if (b) {
1152                 struct drbd_connection *connection = first_peer_device(device)->connection;
1153
1154                 if (blk_queue_discard(b) &&
1155                     (connection->cstate < C_CONNECTED || connection->agreed_features & FF_TRIM)) {
1156                         /* For now, don't allow more than one activity log extent worth of data
1157                          * to be discarded in one go. We may need to rework drbd_al_begin_io()
1158                          * to allow for even larger discard ranges */
1159                         q->limits.max_discard_sectors = DRBD_MAX_DISCARD_SECTORS;
1160
1161                         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1162                         /* REALLY? Is stacking secdiscard "legal"? */
1163                         if (blk_queue_secdiscard(b))
1164                                 queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, q);
1165                 } else {
1166                         q->limits.max_discard_sectors = 0;
1167                         queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1168                         queue_flag_clear_unlocked(QUEUE_FLAG_SECDISCARD, q);
1169                 }
1170
1171                 blk_queue_stack_limits(q, b);
1172
1173                 if (q->backing_dev_info.ra_pages != b->backing_dev_info.ra_pages) {
1174                         drbd_info(device, "Adjusting my ra_pages to backing device's (%lu -> %lu)\n",
1175                                  q->backing_dev_info.ra_pages,
1176                                  b->backing_dev_info.ra_pages);
1177                         q->backing_dev_info.ra_pages = b->backing_dev_info.ra_pages;
1178                 }
1179         }
1180 }
1181
1182 void drbd_reconsider_max_bio_size(struct drbd_device *device, struct drbd_backing_dev *bdev)
1183 {
1184         unsigned int now, new, local, peer;
1185
1186         now = queue_max_hw_sectors(device->rq_queue) << 9;
1187         local = device->local_max_bio_size; /* Eventually last known value, from volatile memory */
1188         peer = device->peer_max_bio_size; /* Eventually last known value, from meta data */
1189
1190         if (bdev) {
1191                 local = queue_max_hw_sectors(bdev->backing_bdev->bd_disk->queue) << 9;
1192                 device->local_max_bio_size = local;
1193         }
1194         local = min(local, DRBD_MAX_BIO_SIZE);
1195
1196         /* We may ignore peer limits if the peer is modern enough.
1197            Because new from 8.3.8 onwards the peer can use multiple
1198            BIOs for a single peer_request */
1199         if (device->state.conn >= C_WF_REPORT_PARAMS) {
1200                 if (first_peer_device(device)->connection->agreed_pro_version < 94)
1201                         peer = min(device->peer_max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1202                         /* Correct old drbd (up to 8.3.7) if it believes it can do more than 32KiB */
1203                 else if (first_peer_device(device)->connection->agreed_pro_version == 94)
1204                         peer = DRBD_MAX_SIZE_H80_PACKET;
1205                 else if (first_peer_device(device)->connection->agreed_pro_version < 100)
1206                         peer = DRBD_MAX_BIO_SIZE_P95;  /* drbd 8.3.8 onwards, before 8.4.0 */
1207                 else
1208                         peer = DRBD_MAX_BIO_SIZE;
1209
1210                 /* We may later detach and re-attach on a disconnected Primary.
1211                  * Avoid this setting to jump back in that case.
1212                  * We want to store what we know the peer DRBD can handle,
1213                  * not what the peer IO backend can handle. */
1214                 if (peer > device->peer_max_bio_size)
1215                         device->peer_max_bio_size = peer;
1216         }
1217         new = min(local, peer);
1218
1219         if (device->state.role == R_PRIMARY && new < now)
1220                 drbd_err(device, "ASSERT FAILED new < now; (%u < %u)\n", new, now);
1221
1222         if (new != now)
1223                 drbd_info(device, "max BIO size = %u\n", new);
1224
1225         drbd_setup_queue_param(device, bdev, new);
1226 }
1227
1228 /* Starts the worker thread */
1229 static void conn_reconfig_start(struct drbd_connection *connection)
1230 {
1231         drbd_thread_start(&connection->worker);
1232         drbd_flush_workqueue(&connection->sender_work);
1233 }
1234
1235 /* if still unconfigured, stops worker again. */
1236 static void conn_reconfig_done(struct drbd_connection *connection)
1237 {
1238         bool stop_threads;
1239         spin_lock_irq(&connection->resource->req_lock);
1240         stop_threads = conn_all_vols_unconf(connection) &&
1241                 connection->cstate == C_STANDALONE;
1242         spin_unlock_irq(&connection->resource->req_lock);
1243         if (stop_threads) {
1244                 /* asender is implicitly stopped by receiver
1245                  * in conn_disconnect() */
1246                 drbd_thread_stop(&connection->receiver);
1247                 drbd_thread_stop(&connection->worker);
1248         }
1249 }
1250
1251 /* Make sure IO is suspended before calling this function(). */
1252 static void drbd_suspend_al(struct drbd_device *device)
1253 {
1254         int s = 0;
1255
1256         if (!lc_try_lock(device->act_log)) {
1257                 drbd_warn(device, "Failed to lock al in drbd_suspend_al()\n");
1258                 return;
1259         }
1260
1261         drbd_al_shrink(device);
1262         spin_lock_irq(&device->resource->req_lock);
1263         if (device->state.conn < C_CONNECTED)
1264                 s = !test_and_set_bit(AL_SUSPENDED, &device->flags);
1265         spin_unlock_irq(&device->resource->req_lock);
1266         lc_unlock(device->act_log);
1267
1268         if (s)
1269                 drbd_info(device, "Suspended AL updates\n");
1270 }
1271
1272
1273 static bool should_set_defaults(struct genl_info *info)
1274 {
1275         unsigned flags = ((struct drbd_genlmsghdr*)info->userhdr)->flags;
1276         return 0 != (flags & DRBD_GENL_F_SET_DEFAULTS);
1277 }
1278
1279 static unsigned int drbd_al_extents_max(struct drbd_backing_dev *bdev)
1280 {
1281         /* This is limited by 16 bit "slot" numbers,
1282          * and by available on-disk context storage.
1283          *
1284          * Also (u16)~0 is special (denotes a "free" extent).
1285          *
1286          * One transaction occupies one 4kB on-disk block,
1287          * we have n such blocks in the on disk ring buffer,
1288          * the "current" transaction may fail (n-1),
1289          * and there is 919 slot numbers context information per transaction.
1290          *
1291          * 72 transaction blocks amounts to more than 2**16 context slots,
1292          * so cap there first.
1293          */
1294         const unsigned int max_al_nr = DRBD_AL_EXTENTS_MAX;
1295         const unsigned int sufficient_on_disk =
1296                 (max_al_nr + AL_CONTEXT_PER_TRANSACTION -1)
1297                 /AL_CONTEXT_PER_TRANSACTION;
1298
1299         unsigned int al_size_4k = bdev->md.al_size_4k;
1300
1301         if (al_size_4k > sufficient_on_disk)
1302                 return max_al_nr;
1303
1304         return (al_size_4k - 1) * AL_CONTEXT_PER_TRANSACTION;
1305 }
1306
1307 static bool write_ordering_changed(struct disk_conf *a, struct disk_conf *b)
1308 {
1309         return  a->disk_barrier != b->disk_barrier ||
1310                 a->disk_flushes != b->disk_flushes ||
1311                 a->disk_drain != b->disk_drain;
1312 }
1313
1314 int drbd_adm_disk_opts(struct sk_buff *skb, struct genl_info *info)
1315 {
1316         struct drbd_config_context adm_ctx;
1317         enum drbd_ret_code retcode;
1318         struct drbd_device *device;
1319         struct disk_conf *new_disk_conf, *old_disk_conf;
1320         struct fifo_buffer *old_plan = NULL, *new_plan = NULL;
1321         int err, fifo_size;
1322
1323         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1324         if (!adm_ctx.reply_skb)
1325                 return retcode;
1326         if (retcode != NO_ERROR)
1327                 goto finish;
1328
1329         device = adm_ctx.device;
1330         mutex_lock(&adm_ctx.resource->adm_mutex);
1331
1332         /* we also need a disk
1333          * to change the options on */
1334         if (!get_ldev(device)) {
1335                 retcode = ERR_NO_DISK;
1336                 goto out;
1337         }
1338
1339         new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
1340         if (!new_disk_conf) {
1341                 retcode = ERR_NOMEM;
1342                 goto fail;
1343         }
1344
1345         mutex_lock(&device->resource->conf_update);
1346         old_disk_conf = device->ldev->disk_conf;
1347         *new_disk_conf = *old_disk_conf;
1348         if (should_set_defaults(info))
1349                 set_disk_conf_defaults(new_disk_conf);
1350
1351         err = disk_conf_from_attrs_for_change(new_disk_conf, info);
1352         if (err && err != -ENOMSG) {
1353                 retcode = ERR_MANDATORY_TAG;
1354                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1355                 goto fail_unlock;
1356         }
1357
1358         if (!expect(new_disk_conf->resync_rate >= 1))
1359                 new_disk_conf->resync_rate = 1;
1360
1361         if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1362                 new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1363         if (new_disk_conf->al_extents > drbd_al_extents_max(device->ldev))
1364                 new_disk_conf->al_extents = drbd_al_extents_max(device->ldev);
1365
1366         if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1367                 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1368
1369         fifo_size = (new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ;
1370         if (fifo_size != device->rs_plan_s->size) {
1371                 new_plan = fifo_alloc(fifo_size);
1372                 if (!new_plan) {
1373                         drbd_err(device, "kmalloc of fifo_buffer failed");
1374                         retcode = ERR_NOMEM;
1375                         goto fail_unlock;
1376                 }
1377         }
1378
1379         drbd_suspend_io(device);
1380         wait_event(device->al_wait, lc_try_lock(device->act_log));
1381         drbd_al_shrink(device);
1382         err = drbd_check_al_size(device, new_disk_conf);
1383         lc_unlock(device->act_log);
1384         wake_up(&device->al_wait);
1385         drbd_resume_io(device);
1386
1387         if (err) {
1388                 retcode = ERR_NOMEM;
1389                 goto fail_unlock;
1390         }
1391
1392         write_lock_irq(&global_state_lock);
1393         retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1394         if (retcode == NO_ERROR) {
1395                 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
1396                 drbd_resync_after_changed(device);
1397         }
1398         write_unlock_irq(&global_state_lock);
1399
1400         if (retcode != NO_ERROR)
1401                 goto fail_unlock;
1402
1403         if (new_plan) {
1404                 old_plan = device->rs_plan_s;
1405                 rcu_assign_pointer(device->rs_plan_s, new_plan);
1406         }
1407
1408         mutex_unlock(&device->resource->conf_update);
1409
1410         if (new_disk_conf->al_updates)
1411                 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1412         else
1413                 device->ldev->md.flags |= MDF_AL_DISABLED;
1414
1415         if (new_disk_conf->md_flushes)
1416                 clear_bit(MD_NO_FUA, &device->flags);
1417         else
1418                 set_bit(MD_NO_FUA, &device->flags);
1419
1420         if (write_ordering_changed(old_disk_conf, new_disk_conf))
1421                 drbd_bump_write_ordering(device->resource, NULL, WO_bdev_flush);
1422
1423         drbd_md_sync(device);
1424
1425         if (device->state.conn >= C_CONNECTED) {
1426                 struct drbd_peer_device *peer_device;
1427
1428                 for_each_peer_device(peer_device, device)
1429                         drbd_send_sync_param(peer_device);
1430         }
1431
1432         synchronize_rcu();
1433         kfree(old_disk_conf);
1434         kfree(old_plan);
1435         mod_timer(&device->request_timer, jiffies + HZ);
1436         goto success;
1437
1438 fail_unlock:
1439         mutex_unlock(&device->resource->conf_update);
1440  fail:
1441         kfree(new_disk_conf);
1442         kfree(new_plan);
1443 success:
1444         put_ldev(device);
1445  out:
1446         mutex_unlock(&adm_ctx.resource->adm_mutex);
1447  finish:
1448         drbd_adm_finish(&adm_ctx, info, retcode);
1449         return 0;
1450 }
1451
1452 int drbd_adm_attach(struct sk_buff *skb, struct genl_info *info)
1453 {
1454         struct drbd_config_context adm_ctx;
1455         struct drbd_device *device;
1456         struct drbd_peer_device *peer_device;
1457         struct drbd_connection *connection;
1458         int err;
1459         enum drbd_ret_code retcode;
1460         enum determine_dev_size dd;
1461         sector_t max_possible_sectors;
1462         sector_t min_md_device_sectors;
1463         struct drbd_backing_dev *nbc = NULL; /* new_backing_conf */
1464         struct disk_conf *new_disk_conf = NULL;
1465         struct block_device *bdev;
1466         struct lru_cache *resync_lru = NULL;
1467         struct fifo_buffer *new_plan = NULL;
1468         union drbd_state ns, os;
1469         enum drbd_state_rv rv;
1470         struct net_conf *nc;
1471
1472         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1473         if (!adm_ctx.reply_skb)
1474                 return retcode;
1475         if (retcode != NO_ERROR)
1476                 goto finish;
1477
1478         device = adm_ctx.device;
1479         mutex_lock(&adm_ctx.resource->adm_mutex);
1480         peer_device = first_peer_device(device);
1481         connection = peer_device ? peer_device->connection : NULL;
1482         conn_reconfig_start(connection);
1483
1484         /* if you want to reconfigure, please tear down first */
1485         if (device->state.disk > D_DISKLESS) {
1486                 retcode = ERR_DISK_CONFIGURED;
1487                 goto fail;
1488         }
1489         /* It may just now have detached because of IO error.  Make sure
1490          * drbd_ldev_destroy is done already, we may end up here very fast,
1491          * e.g. if someone calls attach from the on-io-error handler,
1492          * to realize a "hot spare" feature (not that I'd recommend that) */
1493         wait_event(device->misc_wait, !test_bit(GOING_DISKLESS, &device->flags));
1494
1495         /* make sure there is no leftover from previous force-detach attempts */
1496         clear_bit(FORCE_DETACH, &device->flags);
1497         clear_bit(WAS_IO_ERROR, &device->flags);
1498         clear_bit(WAS_READ_ERROR, &device->flags);
1499
1500         /* and no leftover from previously aborted resync or verify, either */
1501         device->rs_total = 0;
1502         device->rs_failed = 0;
1503         atomic_set(&device->rs_pending_cnt, 0);
1504
1505         /* allocation not in the IO path, drbdsetup context */
1506         nbc = kzalloc(sizeof(struct drbd_backing_dev), GFP_KERNEL);
1507         if (!nbc) {
1508                 retcode = ERR_NOMEM;
1509                 goto fail;
1510         }
1511         spin_lock_init(&nbc->md.uuid_lock);
1512
1513         new_disk_conf = kzalloc(sizeof(struct disk_conf), GFP_KERNEL);
1514         if (!new_disk_conf) {
1515                 retcode = ERR_NOMEM;
1516                 goto fail;
1517         }
1518         nbc->disk_conf = new_disk_conf;
1519
1520         set_disk_conf_defaults(new_disk_conf);
1521         err = disk_conf_from_attrs(new_disk_conf, info);
1522         if (err) {
1523                 retcode = ERR_MANDATORY_TAG;
1524                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1525                 goto fail;
1526         }
1527
1528         if (new_disk_conf->c_plan_ahead > DRBD_C_PLAN_AHEAD_MAX)
1529                 new_disk_conf->c_plan_ahead = DRBD_C_PLAN_AHEAD_MAX;
1530
1531         new_plan = fifo_alloc((new_disk_conf->c_plan_ahead * 10 * SLEEP_TIME) / HZ);
1532         if (!new_plan) {
1533                 retcode = ERR_NOMEM;
1534                 goto fail;
1535         }
1536
1537         if (new_disk_conf->meta_dev_idx < DRBD_MD_INDEX_FLEX_INT) {
1538                 retcode = ERR_MD_IDX_INVALID;
1539                 goto fail;
1540         }
1541
1542         write_lock_irq(&global_state_lock);
1543         retcode = drbd_resync_after_valid(device, new_disk_conf->resync_after);
1544         write_unlock_irq(&global_state_lock);
1545         if (retcode != NO_ERROR)
1546                 goto fail;
1547
1548         rcu_read_lock();
1549         nc = rcu_dereference(connection->net_conf);
1550         if (nc) {
1551                 if (new_disk_conf->fencing == FP_STONITH && nc->wire_protocol == DRBD_PROT_A) {
1552                         rcu_read_unlock();
1553                         retcode = ERR_STONITH_AND_PROT_A;
1554                         goto fail;
1555                 }
1556         }
1557         rcu_read_unlock();
1558
1559         bdev = blkdev_get_by_path(new_disk_conf->backing_dev,
1560                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL, device);
1561         if (IS_ERR(bdev)) {
1562                 drbd_err(device, "open(\"%s\") failed with %ld\n", new_disk_conf->backing_dev,
1563                         PTR_ERR(bdev));
1564                 retcode = ERR_OPEN_DISK;
1565                 goto fail;
1566         }
1567         nbc->backing_bdev = bdev;
1568
1569         /*
1570          * meta_dev_idx >= 0: external fixed size, possibly multiple
1571          * drbd sharing one meta device.  TODO in that case, paranoia
1572          * check that [md_bdev, meta_dev_idx] is not yet used by some
1573          * other drbd minor!  (if you use drbd.conf + drbdadm, that
1574          * should check it for you already; but if you don't, or
1575          * someone fooled it, we need to double check here)
1576          */
1577         bdev = blkdev_get_by_path(new_disk_conf->meta_dev,
1578                                   FMODE_READ | FMODE_WRITE | FMODE_EXCL,
1579                                   (new_disk_conf->meta_dev_idx < 0) ?
1580                                   (void *)device : (void *)drbd_m_holder);
1581         if (IS_ERR(bdev)) {
1582                 drbd_err(device, "open(\"%s\") failed with %ld\n", new_disk_conf->meta_dev,
1583                         PTR_ERR(bdev));
1584                 retcode = ERR_OPEN_MD_DISK;
1585                 goto fail;
1586         }
1587         nbc->md_bdev = bdev;
1588
1589         if ((nbc->backing_bdev == nbc->md_bdev) !=
1590             (new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_INTERNAL ||
1591              new_disk_conf->meta_dev_idx == DRBD_MD_INDEX_FLEX_INT)) {
1592                 retcode = ERR_MD_IDX_INVALID;
1593                 goto fail;
1594         }
1595
1596         resync_lru = lc_create("resync", drbd_bm_ext_cache,
1597                         1, 61, sizeof(struct bm_extent),
1598                         offsetof(struct bm_extent, lce));
1599         if (!resync_lru) {
1600                 retcode = ERR_NOMEM;
1601                 goto fail;
1602         }
1603
1604         /* Read our meta data super block early.
1605          * This also sets other on-disk offsets. */
1606         retcode = drbd_md_read(device, nbc);
1607         if (retcode != NO_ERROR)
1608                 goto fail;
1609
1610         if (new_disk_conf->al_extents < DRBD_AL_EXTENTS_MIN)
1611                 new_disk_conf->al_extents = DRBD_AL_EXTENTS_MIN;
1612         if (new_disk_conf->al_extents > drbd_al_extents_max(nbc))
1613                 new_disk_conf->al_extents = drbd_al_extents_max(nbc);
1614
1615         if (drbd_get_max_capacity(nbc) < new_disk_conf->disk_size) {
1616                 drbd_err(device, "max capacity %llu smaller than disk size %llu\n",
1617                         (unsigned long long) drbd_get_max_capacity(nbc),
1618                         (unsigned long long) new_disk_conf->disk_size);
1619                 retcode = ERR_DISK_TOO_SMALL;
1620                 goto fail;
1621         }
1622
1623         if (new_disk_conf->meta_dev_idx < 0) {
1624                 max_possible_sectors = DRBD_MAX_SECTORS_FLEX;
1625                 /* at least one MB, otherwise it does not make sense */
1626                 min_md_device_sectors = (2<<10);
1627         } else {
1628                 max_possible_sectors = DRBD_MAX_SECTORS;
1629                 min_md_device_sectors = MD_128MB_SECT * (new_disk_conf->meta_dev_idx + 1);
1630         }
1631
1632         if (drbd_get_capacity(nbc->md_bdev) < min_md_device_sectors) {
1633                 retcode = ERR_MD_DISK_TOO_SMALL;
1634                 drbd_warn(device, "refusing attach: md-device too small, "
1635                      "at least %llu sectors needed for this meta-disk type\n",
1636                      (unsigned long long) min_md_device_sectors);
1637                 goto fail;
1638         }
1639
1640         /* Make sure the new disk is big enough
1641          * (we may currently be R_PRIMARY with no local disk...) */
1642         if (drbd_get_max_capacity(nbc) <
1643             drbd_get_capacity(device->this_bdev)) {
1644                 retcode = ERR_DISK_TOO_SMALL;
1645                 goto fail;
1646         }
1647
1648         nbc->known_size = drbd_get_capacity(nbc->backing_bdev);
1649
1650         if (nbc->known_size > max_possible_sectors) {
1651                 drbd_warn(device, "==> truncating very big lower level device "
1652                         "to currently maximum possible %llu sectors <==\n",
1653                         (unsigned long long) max_possible_sectors);
1654                 if (new_disk_conf->meta_dev_idx >= 0)
1655                         drbd_warn(device, "==>> using internal or flexible "
1656                                       "meta data may help <<==\n");
1657         }
1658
1659         drbd_suspend_io(device);
1660         /* also wait for the last barrier ack. */
1661         /* FIXME see also https://daiquiri.linbit/cgi-bin/bugzilla/show_bug.cgi?id=171
1662          * We need a way to either ignore barrier acks for barriers sent before a device
1663          * was attached, or a way to wait for all pending barrier acks to come in.
1664          * As barriers are counted per resource,
1665          * we'd need to suspend io on all devices of a resource.
1666          */
1667         wait_event(device->misc_wait, !atomic_read(&device->ap_pending_cnt) || drbd_suspended(device));
1668         /* and for any other previously queued work */
1669         drbd_flush_workqueue(&connection->sender_work);
1670
1671         rv = _drbd_request_state(device, NS(disk, D_ATTACHING), CS_VERBOSE);
1672         retcode = rv;  /* FIXME: Type mismatch. */
1673         drbd_resume_io(device);
1674         if (rv < SS_SUCCESS)
1675                 goto fail;
1676
1677         if (!get_ldev_if_state(device, D_ATTACHING))
1678                 goto force_diskless;
1679
1680         if (!device->bitmap) {
1681                 if (drbd_bm_init(device)) {
1682                         retcode = ERR_NOMEM;
1683                         goto force_diskless_dec;
1684                 }
1685         }
1686
1687         if (device->state.conn < C_CONNECTED &&
1688             device->state.role == R_PRIMARY && device->ed_uuid &&
1689             (device->ed_uuid & ~((u64)1)) != (nbc->md.uuid[UI_CURRENT] & ~((u64)1))) {
1690                 drbd_err(device, "Can only attach to data with current UUID=%016llX\n",
1691                     (unsigned long long)device->ed_uuid);
1692                 retcode = ERR_DATA_NOT_CURRENT;
1693                 goto force_diskless_dec;
1694         }
1695
1696         /* Since we are diskless, fix the activity log first... */
1697         if (drbd_check_al_size(device, new_disk_conf)) {
1698                 retcode = ERR_NOMEM;
1699                 goto force_diskless_dec;
1700         }
1701
1702         /* Prevent shrinking of consistent devices ! */
1703         if (drbd_md_test_flag(nbc, MDF_CONSISTENT) &&
1704             drbd_new_dev_size(device, nbc, nbc->disk_conf->disk_size, 0) < nbc->md.la_size_sect) {
1705                 drbd_warn(device, "refusing to truncate a consistent device\n");
1706                 retcode = ERR_DISK_TOO_SMALL;
1707                 goto force_diskless_dec;
1708         }
1709
1710         /* Reset the "barriers don't work" bits here, then force meta data to
1711          * be written, to ensure we determine if barriers are supported. */
1712         if (new_disk_conf->md_flushes)
1713                 clear_bit(MD_NO_FUA, &device->flags);
1714         else
1715                 set_bit(MD_NO_FUA, &device->flags);
1716
1717         /* Point of no return reached.
1718          * Devices and memory are no longer released by error cleanup below.
1719          * now device takes over responsibility, and the state engine should
1720          * clean it up somewhere.  */
1721         D_ASSERT(device, device->ldev == NULL);
1722         device->ldev = nbc;
1723         device->resync = resync_lru;
1724         device->rs_plan_s = new_plan;
1725         nbc = NULL;
1726         resync_lru = NULL;
1727         new_disk_conf = NULL;
1728         new_plan = NULL;
1729
1730         drbd_bump_write_ordering(device->resource, device->ldev, WO_bdev_flush);
1731
1732         if (drbd_md_test_flag(device->ldev, MDF_CRASHED_PRIMARY))
1733                 set_bit(CRASHED_PRIMARY, &device->flags);
1734         else
1735                 clear_bit(CRASHED_PRIMARY, &device->flags);
1736
1737         if (drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1738             !(device->state.role == R_PRIMARY && device->resource->susp_nod))
1739                 set_bit(CRASHED_PRIMARY, &device->flags);
1740
1741         device->send_cnt = 0;
1742         device->recv_cnt = 0;
1743         device->read_cnt = 0;
1744         device->writ_cnt = 0;
1745
1746         drbd_reconsider_max_bio_size(device, device->ldev);
1747
1748         /* If I am currently not R_PRIMARY,
1749          * but meta data primary indicator is set,
1750          * I just now recover from a hard crash,
1751          * and have been R_PRIMARY before that crash.
1752          *
1753          * Now, if I had no connection before that crash
1754          * (have been degraded R_PRIMARY), chances are that
1755          * I won't find my peer now either.
1756          *
1757          * In that case, and _only_ in that case,
1758          * we use the degr-wfc-timeout instead of the default,
1759          * so we can automatically recover from a crash of a
1760          * degraded but active "cluster" after a certain timeout.
1761          */
1762         clear_bit(USE_DEGR_WFC_T, &device->flags);
1763         if (device->state.role != R_PRIMARY &&
1764              drbd_md_test_flag(device->ldev, MDF_PRIMARY_IND) &&
1765             !drbd_md_test_flag(device->ldev, MDF_CONNECTED_IND))
1766                 set_bit(USE_DEGR_WFC_T, &device->flags);
1767
1768         dd = drbd_determine_dev_size(device, 0, NULL);
1769         if (dd <= DS_ERROR) {
1770                 retcode = ERR_NOMEM_BITMAP;
1771                 goto force_diskless_dec;
1772         } else if (dd == DS_GREW)
1773                 set_bit(RESYNC_AFTER_NEG, &device->flags);
1774
1775         if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC) ||
1776             (test_bit(CRASHED_PRIMARY, &device->flags) &&
1777              drbd_md_test_flag(device->ldev, MDF_AL_DISABLED))) {
1778                 drbd_info(device, "Assuming that all blocks are out of sync "
1779                      "(aka FullSync)\n");
1780                 if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
1781                         "set_n_write from attaching", BM_LOCKED_MASK)) {
1782                         retcode = ERR_IO_MD_DISK;
1783                         goto force_diskless_dec;
1784                 }
1785         } else {
1786                 if (drbd_bitmap_io(device, &drbd_bm_read,
1787                         "read from attaching", BM_LOCKED_MASK)) {
1788                         retcode = ERR_IO_MD_DISK;
1789                         goto force_diskless_dec;
1790                 }
1791         }
1792
1793         if (_drbd_bm_total_weight(device) == drbd_bm_bits(device))
1794                 drbd_suspend_al(device); /* IO is still suspended here... */
1795
1796         spin_lock_irq(&device->resource->req_lock);
1797         os = drbd_read_state(device);
1798         ns = os;
1799         /* If MDF_CONSISTENT is not set go into inconsistent state,
1800            otherwise investigate MDF_WasUpToDate...
1801            If MDF_WAS_UP_TO_DATE is not set go into D_OUTDATED disk state,
1802            otherwise into D_CONSISTENT state.
1803         */
1804         if (drbd_md_test_flag(device->ldev, MDF_CONSISTENT)) {
1805                 if (drbd_md_test_flag(device->ldev, MDF_WAS_UP_TO_DATE))
1806                         ns.disk = D_CONSISTENT;
1807                 else
1808                         ns.disk = D_OUTDATED;
1809         } else {
1810                 ns.disk = D_INCONSISTENT;
1811         }
1812
1813         if (drbd_md_test_flag(device->ldev, MDF_PEER_OUT_DATED))
1814                 ns.pdsk = D_OUTDATED;
1815
1816         rcu_read_lock();
1817         if (ns.disk == D_CONSISTENT &&
1818             (ns.pdsk == D_OUTDATED || rcu_dereference(device->ldev->disk_conf)->fencing == FP_DONT_CARE))
1819                 ns.disk = D_UP_TO_DATE;
1820
1821         /* All tests on MDF_PRIMARY_IND, MDF_CONNECTED_IND,
1822            MDF_CONSISTENT and MDF_WAS_UP_TO_DATE must happen before
1823            this point, because drbd_request_state() modifies these
1824            flags. */
1825
1826         if (rcu_dereference(device->ldev->disk_conf)->al_updates)
1827                 device->ldev->md.flags &= ~MDF_AL_DISABLED;
1828         else
1829                 device->ldev->md.flags |= MDF_AL_DISABLED;
1830
1831         rcu_read_unlock();
1832
1833         /* In case we are C_CONNECTED postpone any decision on the new disk
1834            state after the negotiation phase. */
1835         if (device->state.conn == C_CONNECTED) {
1836                 device->new_state_tmp.i = ns.i;
1837                 ns.i = os.i;
1838                 ns.disk = D_NEGOTIATING;
1839
1840                 /* We expect to receive up-to-date UUIDs soon.
1841                    To avoid a race in receive_state, free p_uuid while
1842                    holding req_lock. I.e. atomic with the state change */
1843                 kfree(device->p_uuid);
1844                 device->p_uuid = NULL;
1845         }
1846
1847         rv = _drbd_set_state(device, ns, CS_VERBOSE, NULL);
1848         spin_unlock_irq(&device->resource->req_lock);
1849
1850         if (rv < SS_SUCCESS)
1851                 goto force_diskless_dec;
1852
1853         mod_timer(&device->request_timer, jiffies + HZ);
1854
1855         if (device->state.role == R_PRIMARY)
1856                 device->ldev->md.uuid[UI_CURRENT] |=  (u64)1;
1857         else
1858                 device->ldev->md.uuid[UI_CURRENT] &= ~(u64)1;
1859
1860         drbd_md_mark_dirty(device);
1861         drbd_md_sync(device);
1862
1863         kobject_uevent(&disk_to_dev(device->vdisk)->kobj, KOBJ_CHANGE);
1864         put_ldev(device);
1865         conn_reconfig_done(connection);
1866         mutex_unlock(&adm_ctx.resource->adm_mutex);
1867         drbd_adm_finish(&adm_ctx, info, retcode);
1868         return 0;
1869
1870  force_diskless_dec:
1871         put_ldev(device);
1872  force_diskless:
1873         drbd_force_state(device, NS(disk, D_DISKLESS));
1874         drbd_md_sync(device);
1875  fail:
1876         conn_reconfig_done(connection);
1877         if (nbc) {
1878                 if (nbc->backing_bdev)
1879                         blkdev_put(nbc->backing_bdev,
1880                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1881                 if (nbc->md_bdev)
1882                         blkdev_put(nbc->md_bdev,
1883                                    FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1884                 kfree(nbc);
1885         }
1886         kfree(new_disk_conf);
1887         lc_destroy(resync_lru);
1888         kfree(new_plan);
1889         mutex_unlock(&adm_ctx.resource->adm_mutex);
1890  finish:
1891         drbd_adm_finish(&adm_ctx, info, retcode);
1892         return 0;
1893 }
1894
1895 static int adm_detach(struct drbd_device *device, int force)
1896 {
1897         enum drbd_state_rv retcode;
1898         int ret;
1899
1900         if (force) {
1901                 set_bit(FORCE_DETACH, &device->flags);
1902                 drbd_force_state(device, NS(disk, D_FAILED));
1903                 retcode = SS_SUCCESS;
1904                 goto out;
1905         }
1906
1907         drbd_suspend_io(device); /* so no-one is stuck in drbd_al_begin_io */
1908         drbd_md_get_buffer(device, __func__); /* make sure there is no in-flight meta-data IO */
1909         retcode = drbd_request_state(device, NS(disk, D_FAILED));
1910         drbd_md_put_buffer(device);
1911         /* D_FAILED will transition to DISKLESS. */
1912         ret = wait_event_interruptible(device->misc_wait,
1913                         device->state.disk != D_FAILED);
1914         drbd_resume_io(device);
1915         if ((int)retcode == (int)SS_IS_DISKLESS)
1916                 retcode = SS_NOTHING_TO_DO;
1917         if (ret)
1918                 retcode = ERR_INTR;
1919 out:
1920         return retcode;
1921 }
1922
1923 /* Detaching the disk is a process in multiple stages.  First we need to lock
1924  * out application IO, in-flight IO, IO stuck in drbd_al_begin_io.
1925  * Then we transition to D_DISKLESS, and wait for put_ldev() to return all
1926  * internal references as well.
1927  * Only then we have finally detached. */
1928 int drbd_adm_detach(struct sk_buff *skb, struct genl_info *info)
1929 {
1930         struct drbd_config_context adm_ctx;
1931         enum drbd_ret_code retcode;
1932         struct detach_parms parms = { };
1933         int err;
1934
1935         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
1936         if (!adm_ctx.reply_skb)
1937                 return retcode;
1938         if (retcode != NO_ERROR)
1939                 goto out;
1940
1941         if (info->attrs[DRBD_NLA_DETACH_PARMS]) {
1942                 err = detach_parms_from_attrs(&parms, info);
1943                 if (err) {
1944                         retcode = ERR_MANDATORY_TAG;
1945                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
1946                         goto out;
1947                 }
1948         }
1949
1950         mutex_lock(&adm_ctx.resource->adm_mutex);
1951         retcode = adm_detach(adm_ctx.device, parms.force_detach);
1952         mutex_unlock(&adm_ctx.resource->adm_mutex);
1953 out:
1954         drbd_adm_finish(&adm_ctx, info, retcode);
1955         return 0;
1956 }
1957
1958 static bool conn_resync_running(struct drbd_connection *connection)
1959 {
1960         struct drbd_peer_device *peer_device;
1961         bool rv = false;
1962         int vnr;
1963
1964         rcu_read_lock();
1965         idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
1966                 struct drbd_device *device = peer_device->device;
1967                 if (device->state.conn == C_SYNC_SOURCE ||
1968                     device->state.conn == C_SYNC_TARGET ||
1969                     device->state.conn == C_PAUSED_SYNC_S ||
1970                     device->state.conn == C_PAUSED_SYNC_T) {
1971                         rv = true;
1972                         break;
1973                 }
1974         }
1975         rcu_read_unlock();
1976
1977         return rv;
1978 }
1979
1980 static bool conn_ov_running(struct drbd_connection *connection)
1981 {
1982         struct drbd_peer_device *peer_device;
1983         bool rv = false;
1984         int vnr;
1985
1986         rcu_read_lock();
1987         idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
1988                 struct drbd_device *device = peer_device->device;
1989                 if (device->state.conn == C_VERIFY_S ||
1990                     device->state.conn == C_VERIFY_T) {
1991                         rv = true;
1992                         break;
1993                 }
1994         }
1995         rcu_read_unlock();
1996
1997         return rv;
1998 }
1999
2000 static enum drbd_ret_code
2001 _check_net_options(struct drbd_connection *connection, struct net_conf *old_net_conf, struct net_conf *new_net_conf)
2002 {
2003         struct drbd_peer_device *peer_device;
2004         int i;
2005
2006         if (old_net_conf && connection->cstate == C_WF_REPORT_PARAMS && connection->agreed_pro_version < 100) {
2007                 if (new_net_conf->wire_protocol != old_net_conf->wire_protocol)
2008                         return ERR_NEED_APV_100;
2009
2010                 if (new_net_conf->two_primaries != old_net_conf->two_primaries)
2011                         return ERR_NEED_APV_100;
2012
2013                 if (strcmp(new_net_conf->integrity_alg, old_net_conf->integrity_alg))
2014                         return ERR_NEED_APV_100;
2015         }
2016
2017         if (!new_net_conf->two_primaries &&
2018             conn_highest_role(connection) == R_PRIMARY &&
2019             conn_highest_peer(connection) == R_PRIMARY)
2020                 return ERR_NEED_ALLOW_TWO_PRI;
2021
2022         if (new_net_conf->two_primaries &&
2023             (new_net_conf->wire_protocol != DRBD_PROT_C))
2024                 return ERR_NOT_PROTO_C;
2025
2026         idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2027                 struct drbd_device *device = peer_device->device;
2028                 if (get_ldev(device)) {
2029                         enum drbd_fencing_p fp = rcu_dereference(device->ldev->disk_conf)->fencing;
2030                         put_ldev(device);
2031                         if (new_net_conf->wire_protocol == DRBD_PROT_A && fp == FP_STONITH)
2032                                 return ERR_STONITH_AND_PROT_A;
2033                 }
2034                 if (device->state.role == R_PRIMARY && new_net_conf->discard_my_data)
2035                         return ERR_DISCARD_IMPOSSIBLE;
2036         }
2037
2038         if (new_net_conf->on_congestion != OC_BLOCK && new_net_conf->wire_protocol != DRBD_PROT_A)
2039                 return ERR_CONG_NOT_PROTO_A;
2040
2041         return NO_ERROR;
2042 }
2043
2044 static enum drbd_ret_code
2045 check_net_options(struct drbd_connection *connection, struct net_conf *new_net_conf)
2046 {
2047         static enum drbd_ret_code rv;
2048         struct drbd_peer_device *peer_device;
2049         int i;
2050
2051         rcu_read_lock();
2052         rv = _check_net_options(connection, rcu_dereference(connection->net_conf), new_net_conf);
2053         rcu_read_unlock();
2054
2055         /* connection->volumes protected by genl_lock() here */
2056         idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2057                 struct drbd_device *device = peer_device->device;
2058                 if (!device->bitmap) {
2059                         if (drbd_bm_init(device))
2060                                 return ERR_NOMEM;
2061                 }
2062         }
2063
2064         return rv;
2065 }
2066
2067 struct crypto {
2068         struct crypto_hash *verify_tfm;
2069         struct crypto_hash *csums_tfm;
2070         struct crypto_hash *cram_hmac_tfm;
2071         struct crypto_hash *integrity_tfm;
2072 };
2073
2074 static int
2075 alloc_hash(struct crypto_hash **tfm, char *tfm_name, int err_alg)
2076 {
2077         if (!tfm_name[0])
2078                 return NO_ERROR;
2079
2080         *tfm = crypto_alloc_hash(tfm_name, 0, CRYPTO_ALG_ASYNC);
2081         if (IS_ERR(*tfm)) {
2082                 *tfm = NULL;
2083                 return err_alg;
2084         }
2085
2086         return NO_ERROR;
2087 }
2088
2089 static enum drbd_ret_code
2090 alloc_crypto(struct crypto *crypto, struct net_conf *new_net_conf)
2091 {
2092         char hmac_name[CRYPTO_MAX_ALG_NAME];
2093         enum drbd_ret_code rv;
2094
2095         rv = alloc_hash(&crypto->csums_tfm, new_net_conf->csums_alg,
2096                        ERR_CSUMS_ALG);
2097         if (rv != NO_ERROR)
2098                 return rv;
2099         rv = alloc_hash(&crypto->verify_tfm, new_net_conf->verify_alg,
2100                        ERR_VERIFY_ALG);
2101         if (rv != NO_ERROR)
2102                 return rv;
2103         rv = alloc_hash(&crypto->integrity_tfm, new_net_conf->integrity_alg,
2104                        ERR_INTEGRITY_ALG);
2105         if (rv != NO_ERROR)
2106                 return rv;
2107         if (new_net_conf->cram_hmac_alg[0] != 0) {
2108                 snprintf(hmac_name, CRYPTO_MAX_ALG_NAME, "hmac(%s)",
2109                          new_net_conf->cram_hmac_alg);
2110
2111                 rv = alloc_hash(&crypto->cram_hmac_tfm, hmac_name,
2112                                ERR_AUTH_ALG);
2113         }
2114
2115         return rv;
2116 }
2117
2118 static void free_crypto(struct crypto *crypto)
2119 {
2120         crypto_free_hash(crypto->cram_hmac_tfm);
2121         crypto_free_hash(crypto->integrity_tfm);
2122         crypto_free_hash(crypto->csums_tfm);
2123         crypto_free_hash(crypto->verify_tfm);
2124 }
2125
2126 int drbd_adm_net_opts(struct sk_buff *skb, struct genl_info *info)
2127 {
2128         struct drbd_config_context adm_ctx;
2129         enum drbd_ret_code retcode;
2130         struct drbd_connection *connection;
2131         struct net_conf *old_net_conf, *new_net_conf = NULL;
2132         int err;
2133         int ovr; /* online verify running */
2134         int rsr; /* re-sync running */
2135         struct crypto crypto = { };
2136
2137         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2138         if (!adm_ctx.reply_skb)
2139                 return retcode;
2140         if (retcode != NO_ERROR)
2141                 goto finish;
2142
2143         connection = adm_ctx.connection;
2144         mutex_lock(&adm_ctx.resource->adm_mutex);
2145
2146         new_net_conf = kzalloc(sizeof(struct net_conf), GFP_KERNEL);
2147         if (!new_net_conf) {
2148                 retcode = ERR_NOMEM;
2149                 goto out;
2150         }
2151
2152         conn_reconfig_start(connection);
2153
2154         mutex_lock(&connection->data.mutex);
2155         mutex_lock(&connection->resource->conf_update);
2156         old_net_conf = connection->net_conf;
2157
2158         if (!old_net_conf) {
2159                 drbd_msg_put_info(adm_ctx.reply_skb, "net conf missing, try connect");
2160                 retcode = ERR_INVALID_REQUEST;
2161                 goto fail;
2162         }
2163
2164         *new_net_conf = *old_net_conf;
2165         if (should_set_defaults(info))
2166                 set_net_conf_defaults(new_net_conf);
2167
2168         err = net_conf_from_attrs_for_change(new_net_conf, info);
2169         if (err && err != -ENOMSG) {
2170                 retcode = ERR_MANDATORY_TAG;
2171                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2172                 goto fail;
2173         }
2174
2175         retcode = check_net_options(connection, new_net_conf);
2176         if (retcode != NO_ERROR)
2177                 goto fail;
2178
2179         /* re-sync running */
2180         rsr = conn_resync_running(connection);
2181         if (rsr && strcmp(new_net_conf->csums_alg, old_net_conf->csums_alg)) {
2182                 retcode = ERR_CSUMS_RESYNC_RUNNING;
2183                 goto fail;
2184         }
2185
2186         /* online verify running */
2187         ovr = conn_ov_running(connection);
2188         if (ovr && strcmp(new_net_conf->verify_alg, old_net_conf->verify_alg)) {
2189                 retcode = ERR_VERIFY_RUNNING;
2190                 goto fail;
2191         }
2192
2193         retcode = alloc_crypto(&crypto, new_net_conf);
2194         if (retcode != NO_ERROR)
2195                 goto fail;
2196
2197         rcu_assign_pointer(connection->net_conf, new_net_conf);
2198
2199         if (!rsr) {
2200                 crypto_free_hash(connection->csums_tfm);
2201                 connection->csums_tfm = crypto.csums_tfm;
2202                 crypto.csums_tfm = NULL;
2203         }
2204         if (!ovr) {
2205                 crypto_free_hash(connection->verify_tfm);
2206                 connection->verify_tfm = crypto.verify_tfm;
2207                 crypto.verify_tfm = NULL;
2208         }
2209
2210         crypto_free_hash(connection->integrity_tfm);
2211         connection->integrity_tfm = crypto.integrity_tfm;
2212         if (connection->cstate >= C_WF_REPORT_PARAMS && connection->agreed_pro_version >= 100)
2213                 /* Do this without trying to take connection->data.mutex again.  */
2214                 __drbd_send_protocol(connection, P_PROTOCOL_UPDATE);
2215
2216         crypto_free_hash(connection->cram_hmac_tfm);
2217         connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2218
2219         mutex_unlock(&connection->resource->conf_update);
2220         mutex_unlock(&connection->data.mutex);
2221         synchronize_rcu();
2222         kfree(old_net_conf);
2223
2224         if (connection->cstate >= C_WF_REPORT_PARAMS) {
2225                 struct drbd_peer_device *peer_device;
2226                 int vnr;
2227
2228                 idr_for_each_entry(&connection->peer_devices, peer_device, vnr)
2229                         drbd_send_sync_param(peer_device);
2230         }
2231
2232         goto done;
2233
2234  fail:
2235         mutex_unlock(&connection->resource->conf_update);
2236         mutex_unlock(&connection->data.mutex);
2237         free_crypto(&crypto);
2238         kfree(new_net_conf);
2239  done:
2240         conn_reconfig_done(connection);
2241  out:
2242         mutex_unlock(&adm_ctx.resource->adm_mutex);
2243  finish:
2244         drbd_adm_finish(&adm_ctx, info, retcode);
2245         return 0;
2246 }
2247
2248 int drbd_adm_connect(struct sk_buff *skb, struct genl_info *info)
2249 {
2250         struct drbd_config_context adm_ctx;
2251         struct drbd_peer_device *peer_device;
2252         struct net_conf *old_net_conf, *new_net_conf = NULL;
2253         struct crypto crypto = { };
2254         struct drbd_resource *resource;
2255         struct drbd_connection *connection;
2256         enum drbd_ret_code retcode;
2257         int i;
2258         int err;
2259
2260         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2261
2262         if (!adm_ctx.reply_skb)
2263                 return retcode;
2264         if (retcode != NO_ERROR)
2265                 goto out;
2266         if (!(adm_ctx.my_addr && adm_ctx.peer_addr)) {
2267                 drbd_msg_put_info(adm_ctx.reply_skb, "connection endpoint(s) missing");
2268                 retcode = ERR_INVALID_REQUEST;
2269                 goto out;
2270         }
2271
2272         /* No need for _rcu here. All reconfiguration is
2273          * strictly serialized on genl_lock(). We are protected against
2274          * concurrent reconfiguration/addition/deletion */
2275         for_each_resource(resource, &drbd_resources) {
2276                 for_each_connection(connection, resource) {
2277                         if (nla_len(adm_ctx.my_addr) == connection->my_addr_len &&
2278                             !memcmp(nla_data(adm_ctx.my_addr), &connection->my_addr,
2279                                     connection->my_addr_len)) {
2280                                 retcode = ERR_LOCAL_ADDR;
2281                                 goto out;
2282                         }
2283
2284                         if (nla_len(adm_ctx.peer_addr) == connection->peer_addr_len &&
2285                             !memcmp(nla_data(adm_ctx.peer_addr), &connection->peer_addr,
2286                                     connection->peer_addr_len)) {
2287                                 retcode = ERR_PEER_ADDR;
2288                                 goto out;
2289                         }
2290                 }
2291         }
2292
2293         mutex_lock(&adm_ctx.resource->adm_mutex);
2294         connection = first_connection(adm_ctx.resource);
2295         conn_reconfig_start(connection);
2296
2297         if (connection->cstate > C_STANDALONE) {
2298                 retcode = ERR_NET_CONFIGURED;
2299                 goto fail;
2300         }
2301
2302         /* allocation not in the IO path, drbdsetup / netlink process context */
2303         new_net_conf = kzalloc(sizeof(*new_net_conf), GFP_KERNEL);
2304         if (!new_net_conf) {
2305                 retcode = ERR_NOMEM;
2306                 goto fail;
2307         }
2308
2309         set_net_conf_defaults(new_net_conf);
2310
2311         err = net_conf_from_attrs(new_net_conf, info);
2312         if (err && err != -ENOMSG) {
2313                 retcode = ERR_MANDATORY_TAG;
2314                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2315                 goto fail;
2316         }
2317
2318         retcode = check_net_options(connection, new_net_conf);
2319         if (retcode != NO_ERROR)
2320                 goto fail;
2321
2322         retcode = alloc_crypto(&crypto, new_net_conf);
2323         if (retcode != NO_ERROR)
2324                 goto fail;
2325
2326         ((char *)new_net_conf->shared_secret)[SHARED_SECRET_MAX-1] = 0;
2327
2328         drbd_flush_workqueue(&connection->sender_work);
2329
2330         mutex_lock(&adm_ctx.resource->conf_update);
2331         old_net_conf = connection->net_conf;
2332         if (old_net_conf) {
2333                 retcode = ERR_NET_CONFIGURED;
2334                 mutex_unlock(&adm_ctx.resource->conf_update);
2335                 goto fail;
2336         }
2337         rcu_assign_pointer(connection->net_conf, new_net_conf);
2338
2339         conn_free_crypto(connection);
2340         connection->cram_hmac_tfm = crypto.cram_hmac_tfm;
2341         connection->integrity_tfm = crypto.integrity_tfm;
2342         connection->csums_tfm = crypto.csums_tfm;
2343         connection->verify_tfm = crypto.verify_tfm;
2344
2345         connection->my_addr_len = nla_len(adm_ctx.my_addr);
2346         memcpy(&connection->my_addr, nla_data(adm_ctx.my_addr), connection->my_addr_len);
2347         connection->peer_addr_len = nla_len(adm_ctx.peer_addr);
2348         memcpy(&connection->peer_addr, nla_data(adm_ctx.peer_addr), connection->peer_addr_len);
2349
2350         mutex_unlock(&adm_ctx.resource->conf_update);
2351
2352         rcu_read_lock();
2353         idr_for_each_entry(&connection->peer_devices, peer_device, i) {
2354                 struct drbd_device *device = peer_device->device;
2355                 device->send_cnt = 0;
2356                 device->recv_cnt = 0;
2357         }
2358         rcu_read_unlock();
2359
2360         retcode = conn_request_state(connection, NS(conn, C_UNCONNECTED), CS_VERBOSE);
2361
2362         conn_reconfig_done(connection);
2363         mutex_unlock(&adm_ctx.resource->adm_mutex);
2364         drbd_adm_finish(&adm_ctx, info, retcode);
2365         return 0;
2366
2367 fail:
2368         free_crypto(&crypto);
2369         kfree(new_net_conf);
2370
2371         conn_reconfig_done(connection);
2372         mutex_unlock(&adm_ctx.resource->adm_mutex);
2373 out:
2374         drbd_adm_finish(&adm_ctx, info, retcode);
2375         return 0;
2376 }
2377
2378 static enum drbd_state_rv conn_try_disconnect(struct drbd_connection *connection, bool force)
2379 {
2380         enum drbd_state_rv rv;
2381
2382         rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2383                         force ? CS_HARD : 0);
2384
2385         switch (rv) {
2386         case SS_NOTHING_TO_DO:
2387                 break;
2388         case SS_ALREADY_STANDALONE:
2389                 return SS_SUCCESS;
2390         case SS_PRIMARY_NOP:
2391                 /* Our state checking code wants to see the peer outdated. */
2392                 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING, pdsk, D_OUTDATED), 0);
2393
2394                 if (rv == SS_OUTDATE_WO_CONN) /* lost connection before graceful disconnect succeeded */
2395                         rv = conn_request_state(connection, NS(conn, C_DISCONNECTING), CS_VERBOSE);
2396
2397                 break;
2398         case SS_CW_FAILED_BY_PEER:
2399                 /* The peer probably wants to see us outdated. */
2400                 rv = conn_request_state(connection, NS2(conn, C_DISCONNECTING,
2401                                                         disk, D_OUTDATED), 0);
2402                 if (rv == SS_IS_DISKLESS || rv == SS_LOWER_THAN_OUTDATED) {
2403                         rv = conn_request_state(connection, NS(conn, C_DISCONNECTING),
2404                                         CS_HARD);
2405                 }
2406                 break;
2407         default:;
2408                 /* no special handling necessary */
2409         }
2410
2411         if (rv >= SS_SUCCESS) {
2412                 enum drbd_state_rv rv2;
2413                 /* No one else can reconfigure the network while I am here.
2414                  * The state handling only uses drbd_thread_stop_nowait(),
2415                  * we want to really wait here until the receiver is no more.
2416                  */
2417                 drbd_thread_stop(&connection->receiver);
2418
2419                 /* Race breaker.  This additional state change request may be
2420                  * necessary, if this was a forced disconnect during a receiver
2421                  * restart.  We may have "killed" the receiver thread just
2422                  * after drbd_receiver() returned.  Typically, we should be
2423                  * C_STANDALONE already, now, and this becomes a no-op.
2424                  */
2425                 rv2 = conn_request_state(connection, NS(conn, C_STANDALONE),
2426                                 CS_VERBOSE | CS_HARD);
2427                 if (rv2 < SS_SUCCESS)
2428                         drbd_err(connection,
2429                                 "unexpected rv2=%d in conn_try_disconnect()\n",
2430                                 rv2);
2431         }
2432         return rv;
2433 }
2434
2435 int drbd_adm_disconnect(struct sk_buff *skb, struct genl_info *info)
2436 {
2437         struct drbd_config_context adm_ctx;
2438         struct disconnect_parms parms;
2439         struct drbd_connection *connection;
2440         enum drbd_state_rv rv;
2441         enum drbd_ret_code retcode;
2442         int err;
2443
2444         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_CONNECTION);
2445         if (!adm_ctx.reply_skb)
2446                 return retcode;
2447         if (retcode != NO_ERROR)
2448                 goto fail;
2449
2450         connection = adm_ctx.connection;
2451         memset(&parms, 0, sizeof(parms));
2452         if (info->attrs[DRBD_NLA_DISCONNECT_PARMS]) {
2453                 err = disconnect_parms_from_attrs(&parms, info);
2454                 if (err) {
2455                         retcode = ERR_MANDATORY_TAG;
2456                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2457                         goto fail;
2458                 }
2459         }
2460
2461         mutex_lock(&adm_ctx.resource->adm_mutex);
2462         rv = conn_try_disconnect(connection, parms.force_disconnect);
2463         if (rv < SS_SUCCESS)
2464                 retcode = rv;  /* FIXME: Type mismatch. */
2465         else
2466                 retcode = NO_ERROR;
2467         mutex_unlock(&adm_ctx.resource->adm_mutex);
2468  fail:
2469         drbd_adm_finish(&adm_ctx, info, retcode);
2470         return 0;
2471 }
2472
2473 void resync_after_online_grow(struct drbd_device *device)
2474 {
2475         int iass; /* I am sync source */
2476
2477         drbd_info(device, "Resync of new storage after online grow\n");
2478         if (device->state.role != device->state.peer)
2479                 iass = (device->state.role == R_PRIMARY);
2480         else
2481                 iass = test_bit(RESOLVE_CONFLICTS, &first_peer_device(device)->connection->flags);
2482
2483         if (iass)
2484                 drbd_start_resync(device, C_SYNC_SOURCE);
2485         else
2486                 _drbd_request_state(device, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE + CS_SERIALIZE);
2487 }
2488
2489 int drbd_adm_resize(struct sk_buff *skb, struct genl_info *info)
2490 {
2491         struct drbd_config_context adm_ctx;
2492         struct disk_conf *old_disk_conf, *new_disk_conf = NULL;
2493         struct resize_parms rs;
2494         struct drbd_device *device;
2495         enum drbd_ret_code retcode;
2496         enum determine_dev_size dd;
2497         bool change_al_layout = false;
2498         enum dds_flags ddsf;
2499         sector_t u_size;
2500         int err;
2501
2502         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2503         if (!adm_ctx.reply_skb)
2504                 return retcode;
2505         if (retcode != NO_ERROR)
2506                 goto finish;
2507
2508         mutex_lock(&adm_ctx.resource->adm_mutex);
2509         device = adm_ctx.device;
2510         if (!get_ldev(device)) {
2511                 retcode = ERR_NO_DISK;
2512                 goto fail;
2513         }
2514
2515         memset(&rs, 0, sizeof(struct resize_parms));
2516         rs.al_stripes = device->ldev->md.al_stripes;
2517         rs.al_stripe_size = device->ldev->md.al_stripe_size_4k * 4;
2518         if (info->attrs[DRBD_NLA_RESIZE_PARMS]) {
2519                 err = resize_parms_from_attrs(&rs, info);
2520                 if (err) {
2521                         retcode = ERR_MANDATORY_TAG;
2522                         drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2523                         goto fail_ldev;
2524                 }
2525         }
2526
2527         if (device->state.conn > C_CONNECTED) {
2528                 retcode = ERR_RESIZE_RESYNC;
2529                 goto fail_ldev;
2530         }
2531
2532         if (device->state.role == R_SECONDARY &&
2533             device->state.peer == R_SECONDARY) {
2534                 retcode = ERR_NO_PRIMARY;
2535                 goto fail_ldev;
2536         }
2537
2538         if (rs.no_resync && first_peer_device(device)->connection->agreed_pro_version < 93) {
2539                 retcode = ERR_NEED_APV_93;
2540                 goto fail_ldev;
2541         }
2542
2543         rcu_read_lock();
2544         u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
2545         rcu_read_unlock();
2546         if (u_size != (sector_t)rs.resize_size) {
2547                 new_disk_conf = kmalloc(sizeof(struct disk_conf), GFP_KERNEL);
2548                 if (!new_disk_conf) {
2549                         retcode = ERR_NOMEM;
2550                         goto fail_ldev;
2551                 }
2552         }
2553
2554         if (device->ldev->md.al_stripes != rs.al_stripes ||
2555             device->ldev->md.al_stripe_size_4k != rs.al_stripe_size / 4) {
2556                 u32 al_size_k = rs.al_stripes * rs.al_stripe_size;
2557
2558                 if (al_size_k > (16 * 1024 * 1024)) {
2559                         retcode = ERR_MD_LAYOUT_TOO_BIG;
2560                         goto fail_ldev;
2561                 }
2562
2563                 if (al_size_k < MD_32kB_SECT/2) {
2564                         retcode = ERR_MD_LAYOUT_TOO_SMALL;
2565                         goto fail_ldev;
2566                 }
2567
2568                 if (device->state.conn != C_CONNECTED && !rs.resize_force) {
2569                         retcode = ERR_MD_LAYOUT_CONNECTED;
2570                         goto fail_ldev;
2571                 }
2572
2573                 change_al_layout = true;
2574         }
2575
2576         if (device->ldev->known_size != drbd_get_capacity(device->ldev->backing_bdev))
2577                 device->ldev->known_size = drbd_get_capacity(device->ldev->backing_bdev);
2578
2579         if (new_disk_conf) {
2580                 mutex_lock(&device->resource->conf_update);
2581                 old_disk_conf = device->ldev->disk_conf;
2582                 *new_disk_conf = *old_disk_conf;
2583                 new_disk_conf->disk_size = (sector_t)rs.resize_size;
2584                 rcu_assign_pointer(device->ldev->disk_conf, new_disk_conf);
2585                 mutex_unlock(&device->resource->conf_update);
2586                 synchronize_rcu();
2587                 kfree(old_disk_conf);
2588         }
2589
2590         ddsf = (rs.resize_force ? DDSF_FORCED : 0) | (rs.no_resync ? DDSF_NO_RESYNC : 0);
2591         dd = drbd_determine_dev_size(device, ddsf, change_al_layout ? &rs : NULL);
2592         drbd_md_sync(device);
2593         put_ldev(device);
2594         if (dd == DS_ERROR) {
2595                 retcode = ERR_NOMEM_BITMAP;
2596                 goto fail;
2597         } else if (dd == DS_ERROR_SPACE_MD) {
2598                 retcode = ERR_MD_LAYOUT_NO_FIT;
2599                 goto fail;
2600         } else if (dd == DS_ERROR_SHRINK) {
2601                 retcode = ERR_IMPLICIT_SHRINK;
2602                 goto fail;
2603         }
2604
2605         if (device->state.conn == C_CONNECTED) {
2606                 if (dd == DS_GREW)
2607                         set_bit(RESIZE_PENDING, &device->flags);
2608
2609                 drbd_send_uuids(first_peer_device(device));
2610                 drbd_send_sizes(first_peer_device(device), 1, ddsf);
2611         }
2612
2613  fail:
2614         mutex_unlock(&adm_ctx.resource->adm_mutex);
2615  finish:
2616         drbd_adm_finish(&adm_ctx, info, retcode);
2617         return 0;
2618
2619  fail_ldev:
2620         put_ldev(device);
2621         goto fail;
2622 }
2623
2624 int drbd_adm_resource_opts(struct sk_buff *skb, struct genl_info *info)
2625 {
2626         struct drbd_config_context adm_ctx;
2627         enum drbd_ret_code retcode;
2628         struct res_opts res_opts;
2629         int err;
2630
2631         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_RESOURCE);
2632         if (!adm_ctx.reply_skb)
2633                 return retcode;
2634         if (retcode != NO_ERROR)
2635                 goto fail;
2636
2637         res_opts = adm_ctx.resource->res_opts;
2638         if (should_set_defaults(info))
2639                 set_res_opts_defaults(&res_opts);
2640
2641         err = res_opts_from_attrs(&res_opts, info);
2642         if (err && err != -ENOMSG) {
2643                 retcode = ERR_MANDATORY_TAG;
2644                 drbd_msg_put_info(adm_ctx.reply_skb, from_attrs_err_to_txt(err));
2645                 goto fail;
2646         }
2647
2648         mutex_lock(&adm_ctx.resource->adm_mutex);
2649         err = set_resource_options(adm_ctx.resource, &res_opts);
2650         if (err) {
2651                 retcode = ERR_INVALID_REQUEST;
2652                 if (err == -ENOMEM)
2653                         retcode = ERR_NOMEM;
2654         }
2655         mutex_unlock(&adm_ctx.resource->adm_mutex);
2656
2657 fail:
2658         drbd_adm_finish(&adm_ctx, info, retcode);
2659         return 0;
2660 }
2661
2662 int drbd_adm_invalidate(struct sk_buff *skb, struct genl_info *info)
2663 {
2664         struct drbd_config_context adm_ctx;
2665         struct drbd_device *device;
2666         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2667
2668         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2669         if (!adm_ctx.reply_skb)
2670                 return retcode;
2671         if (retcode != NO_ERROR)
2672                 goto out;
2673
2674         device = adm_ctx.device;
2675         if (!get_ldev(device)) {
2676                 retcode = ERR_NO_DISK;
2677                 goto out;
2678         }
2679
2680         mutex_lock(&adm_ctx.resource->adm_mutex);
2681
2682         /* If there is still bitmap IO pending, probably because of a previous
2683          * resync just being finished, wait for it before requesting a new resync.
2684          * Also wait for it's after_state_ch(). */
2685         drbd_suspend_io(device);
2686         wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2687         drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2688
2689         /* If we happen to be C_STANDALONE R_SECONDARY, just change to
2690          * D_INCONSISTENT, and set all bits in the bitmap.  Otherwise,
2691          * try to start a resync handshake as sync target for full sync.
2692          */
2693         if (device->state.conn == C_STANDALONE && device->state.role == R_SECONDARY) {
2694                 retcode = drbd_request_state(device, NS(disk, D_INCONSISTENT));
2695                 if (retcode >= SS_SUCCESS) {
2696                         if (drbd_bitmap_io(device, &drbd_bmio_set_n_write,
2697                                 "set_n_write from invalidate", BM_LOCKED_MASK))
2698                                 retcode = ERR_IO_MD_DISK;
2699                 }
2700         } else
2701                 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_T));
2702         drbd_resume_io(device);
2703         mutex_unlock(&adm_ctx.resource->adm_mutex);
2704         put_ldev(device);
2705 out:
2706         drbd_adm_finish(&adm_ctx, info, retcode);
2707         return 0;
2708 }
2709
2710 static int drbd_adm_simple_request_state(struct sk_buff *skb, struct genl_info *info,
2711                 union drbd_state mask, union drbd_state val)
2712 {
2713         struct drbd_config_context adm_ctx;
2714         enum drbd_ret_code retcode;
2715
2716         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2717         if (!adm_ctx.reply_skb)
2718                 return retcode;
2719         if (retcode != NO_ERROR)
2720                 goto out;
2721
2722         mutex_lock(&adm_ctx.resource->adm_mutex);
2723         retcode = drbd_request_state(adm_ctx.device, mask, val);
2724         mutex_unlock(&adm_ctx.resource->adm_mutex);
2725 out:
2726         drbd_adm_finish(&adm_ctx, info, retcode);
2727         return 0;
2728 }
2729
2730 static int drbd_bmio_set_susp_al(struct drbd_device *device) __must_hold(local)
2731 {
2732         int rv;
2733
2734         rv = drbd_bmio_set_n_write(device);
2735         drbd_suspend_al(device);
2736         return rv;
2737 }
2738
2739 int drbd_adm_invalidate_peer(struct sk_buff *skb, struct genl_info *info)
2740 {
2741         struct drbd_config_context adm_ctx;
2742         int retcode; /* drbd_ret_code, drbd_state_rv */
2743         struct drbd_device *device;
2744
2745         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2746         if (!adm_ctx.reply_skb)
2747                 return retcode;
2748         if (retcode != NO_ERROR)
2749                 goto out;
2750
2751         device = adm_ctx.device;
2752         if (!get_ldev(device)) {
2753                 retcode = ERR_NO_DISK;
2754                 goto out;
2755         }
2756
2757         mutex_lock(&adm_ctx.resource->adm_mutex);
2758
2759         /* If there is still bitmap IO pending, probably because of a previous
2760          * resync just being finished, wait for it before requesting a new resync.
2761          * Also wait for it's after_state_ch(). */
2762         drbd_suspend_io(device);
2763         wait_event(device->misc_wait, !test_bit(BITMAP_IO, &device->flags));
2764         drbd_flush_workqueue(&first_peer_device(device)->connection->sender_work);
2765
2766         /* If we happen to be C_STANDALONE R_PRIMARY, just set all bits
2767          * in the bitmap.  Otherwise, try to start a resync handshake
2768          * as sync source for full sync.
2769          */
2770         if (device->state.conn == C_STANDALONE && device->state.role == R_PRIMARY) {
2771                 /* The peer will get a resync upon connect anyways. Just make that
2772                    into a full resync. */
2773                 retcode = drbd_request_state(device, NS(pdsk, D_INCONSISTENT));
2774                 if (retcode >= SS_SUCCESS) {
2775                         if (drbd_bitmap_io(device, &drbd_bmio_set_susp_al,
2776                                 "set_n_write from invalidate_peer",
2777                                 BM_LOCKED_SET_ALLOWED))
2778                                 retcode = ERR_IO_MD_DISK;
2779                 }
2780         } else
2781                 retcode = drbd_request_state(device, NS(conn, C_STARTING_SYNC_S));
2782         drbd_resume_io(device);
2783         mutex_unlock(&adm_ctx.resource->adm_mutex);
2784         put_ldev(device);
2785 out:
2786         drbd_adm_finish(&adm_ctx, info, retcode);
2787         return 0;
2788 }
2789
2790 int drbd_adm_pause_sync(struct sk_buff *skb, struct genl_info *info)
2791 {
2792         struct drbd_config_context adm_ctx;
2793         enum drbd_ret_code retcode;
2794
2795         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2796         if (!adm_ctx.reply_skb)
2797                 return retcode;
2798         if (retcode != NO_ERROR)
2799                 goto out;
2800
2801         mutex_lock(&adm_ctx.resource->adm_mutex);
2802         if (drbd_request_state(adm_ctx.device, NS(user_isp, 1)) == SS_NOTHING_TO_DO)
2803                 retcode = ERR_PAUSE_IS_SET;
2804         mutex_unlock(&adm_ctx.resource->adm_mutex);
2805 out:
2806         drbd_adm_finish(&adm_ctx, info, retcode);
2807         return 0;
2808 }
2809
2810 int drbd_adm_resume_sync(struct sk_buff *skb, struct genl_info *info)
2811 {
2812         struct drbd_config_context adm_ctx;
2813         union drbd_dev_state s;
2814         enum drbd_ret_code retcode;
2815
2816         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2817         if (!adm_ctx.reply_skb)
2818                 return retcode;
2819         if (retcode != NO_ERROR)
2820                 goto out;
2821
2822         mutex_lock(&adm_ctx.resource->adm_mutex);
2823         if (drbd_request_state(adm_ctx.device, NS(user_isp, 0)) == SS_NOTHING_TO_DO) {
2824                 s = adm_ctx.device->state;
2825                 if (s.conn == C_PAUSED_SYNC_S || s.conn == C_PAUSED_SYNC_T) {
2826                         retcode = s.aftr_isp ? ERR_PIC_AFTER_DEP :
2827                                   s.peer_isp ? ERR_PIC_PEER_DEP : ERR_PAUSE_IS_CLEAR;
2828                 } else {
2829                         retcode = ERR_PAUSE_IS_CLEAR;
2830                 }
2831         }
2832         mutex_unlock(&adm_ctx.resource->adm_mutex);
2833 out:
2834         drbd_adm_finish(&adm_ctx, info, retcode);
2835         return 0;
2836 }
2837
2838 int drbd_adm_suspend_io(struct sk_buff *skb, struct genl_info *info)
2839 {
2840         return drbd_adm_simple_request_state(skb, info, NS(susp, 1));
2841 }
2842
2843 int drbd_adm_resume_io(struct sk_buff *skb, struct genl_info *info)
2844 {
2845         struct drbd_config_context adm_ctx;
2846         struct drbd_device *device;
2847         int retcode; /* enum drbd_ret_code rsp. enum drbd_state_rv */
2848
2849         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
2850         if (!adm_ctx.reply_skb)
2851                 return retcode;
2852         if (retcode != NO_ERROR)
2853                 goto out;
2854
2855         mutex_lock(&adm_ctx.resource->adm_mutex);
2856         device = adm_ctx.device;
2857         if (test_bit(NEW_CUR_UUID, &device->flags)) {
2858                 drbd_uuid_new_current(device);
2859                 clear_bit(NEW_CUR_UUID, &device->flags);
2860         }
2861         drbd_suspend_io(device);
2862         retcode = drbd_request_state(device, NS3(susp, 0, susp_nod, 0, susp_fen, 0));
2863         if (retcode == SS_SUCCESS) {
2864                 if (device->state.conn < C_CONNECTED)
2865                         tl_clear(first_peer_device(device)->connection);
2866                 if (device->state.disk == D_DISKLESS || device->state.disk == D_FAILED)
2867                         tl_restart(first_peer_device(device)->connection, FAIL_FROZEN_DISK_IO);
2868         }
2869         drbd_resume_io(device);
2870         mutex_unlock(&adm_ctx.resource->adm_mutex);
2871 out:
2872         drbd_adm_finish(&adm_ctx, info, retcode);
2873         return 0;
2874 }
2875
2876 int drbd_adm_outdate(struct sk_buff *skb, struct genl_info *info)
2877 {
2878         return drbd_adm_simple_request_state(skb, info, NS(disk, D_OUTDATED));
2879 }
2880
2881 static int nla_put_drbd_cfg_context(struct sk_buff *skb,
2882                                     struct drbd_resource *resource,
2883                                     struct drbd_connection *connection,
2884                                     struct drbd_device *device)
2885 {
2886         struct nlattr *nla;
2887         nla = nla_nest_start(skb, DRBD_NLA_CFG_CONTEXT);
2888         if (!nla)
2889                 goto nla_put_failure;
2890         if (device &&
2891             nla_put_u32(skb, T_ctx_volume, device->vnr))
2892                 goto nla_put_failure;
2893         if (nla_put_string(skb, T_ctx_resource_name, resource->name))
2894                 goto nla_put_failure;
2895         if (connection) {
2896                 if (connection->my_addr_len &&
2897                     nla_put(skb, T_ctx_my_addr, connection->my_addr_len, &connection->my_addr))
2898                         goto nla_put_failure;
2899                 if (connection->peer_addr_len &&
2900                     nla_put(skb, T_ctx_peer_addr, connection->peer_addr_len, &connection->peer_addr))
2901                         goto nla_put_failure;
2902         }
2903         nla_nest_end(skb, nla);
2904         return 0;
2905
2906 nla_put_failure:
2907         if (nla)
2908                 nla_nest_cancel(skb, nla);
2909         return -EMSGSIZE;
2910 }
2911
2912 /*
2913  * Return the connection of @resource if @resource has exactly one connection.
2914  */
2915 static struct drbd_connection *the_only_connection(struct drbd_resource *resource)
2916 {
2917         struct list_head *connections = &resource->connections;
2918
2919         if (list_empty(connections) || connections->next->next != connections)
2920                 return NULL;
2921         return list_first_entry(&resource->connections, struct drbd_connection, connections);
2922 }
2923
2924 static int nla_put_status_info(struct sk_buff *skb, struct drbd_device *device,
2925                 const struct sib_info *sib)
2926 {
2927         struct drbd_resource *resource = device->resource;
2928         struct state_info *si = NULL; /* for sizeof(si->member); */
2929         struct nlattr *nla;
2930         int got_ldev;
2931         int err = 0;
2932         int exclude_sensitive;
2933
2934         /* If sib != NULL, this is drbd_bcast_event, which anyone can listen
2935          * to.  So we better exclude_sensitive information.
2936          *
2937          * If sib == NULL, this is drbd_adm_get_status, executed synchronously
2938          * in the context of the requesting user process. Exclude sensitive
2939          * information, unless current has superuser.
2940          *
2941          * NOTE: for drbd_adm_get_status_all(), this is a netlink dump, and
2942          * relies on the current implementation of netlink_dump(), which
2943          * executes the dump callback successively from netlink_recvmsg(),
2944          * always in the context of the receiving process */
2945         exclude_sensitive = sib || !capable(CAP_SYS_ADMIN);
2946
2947         got_ldev = get_ldev(device);
2948
2949         /* We need to add connection name and volume number information still.
2950          * Minor number is in drbd_genlmsghdr. */
2951         if (nla_put_drbd_cfg_context(skb, resource, the_only_connection(resource), device))
2952                 goto nla_put_failure;
2953
2954         if (res_opts_to_skb(skb, &device->resource->res_opts, exclude_sensitive))
2955                 goto nla_put_failure;
2956
2957         rcu_read_lock();
2958         if (got_ldev) {
2959                 struct disk_conf *disk_conf;
2960
2961                 disk_conf = rcu_dereference(device->ldev->disk_conf);
2962                 err = disk_conf_to_skb(skb, disk_conf, exclude_sensitive);
2963         }
2964         if (!err) {
2965                 struct net_conf *nc;
2966
2967                 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
2968                 if (nc)
2969                         err = net_conf_to_skb(skb, nc, exclude_sensitive);
2970         }
2971         rcu_read_unlock();
2972         if (err)
2973                 goto nla_put_failure;
2974
2975         nla = nla_nest_start(skb, DRBD_NLA_STATE_INFO);
2976         if (!nla)
2977                 goto nla_put_failure;
2978         if (nla_put_u32(skb, T_sib_reason, sib ? sib->sib_reason : SIB_GET_STATUS_REPLY) ||
2979             nla_put_u32(skb, T_current_state, device->state.i) ||
2980             nla_put_u64(skb, T_ed_uuid, device->ed_uuid) ||
2981             nla_put_u64(skb, T_capacity, drbd_get_capacity(device->this_bdev)) ||
2982             nla_put_u64(skb, T_send_cnt, device->send_cnt) ||
2983             nla_put_u64(skb, T_recv_cnt, device->recv_cnt) ||
2984             nla_put_u64(skb, T_read_cnt, device->read_cnt) ||
2985             nla_put_u64(skb, T_writ_cnt, device->writ_cnt) ||
2986             nla_put_u64(skb, T_al_writ_cnt, device->al_writ_cnt) ||
2987             nla_put_u64(skb, T_bm_writ_cnt, device->bm_writ_cnt) ||
2988             nla_put_u32(skb, T_ap_bio_cnt, atomic_read(&device->ap_bio_cnt)) ||
2989             nla_put_u32(skb, T_ap_pending_cnt, atomic_read(&device->ap_pending_cnt)) ||
2990             nla_put_u32(skb, T_rs_pending_cnt, atomic_read(&device->rs_pending_cnt)))
2991                 goto nla_put_failure;
2992
2993         if (got_ldev) {
2994                 int err;
2995
2996                 spin_lock_irq(&device->ldev->md.uuid_lock);
2997                 err = nla_put(skb, T_uuids, sizeof(si->uuids), device->ldev->md.uuid);
2998                 spin_unlock_irq(&device->ldev->md.uuid_lock);
2999
3000                 if (err)
3001                         goto nla_put_failure;
3002
3003                 if (nla_put_u32(skb, T_disk_flags, device->ldev->md.flags) ||
3004                     nla_put_u64(skb, T_bits_total, drbd_bm_bits(device)) ||
3005                     nla_put_u64(skb, T_bits_oos, drbd_bm_total_weight(device)))
3006                         goto nla_put_failure;
3007                 if (C_SYNC_SOURCE <= device->state.conn &&
3008                     C_PAUSED_SYNC_T >= device->state.conn) {
3009                         if (nla_put_u64(skb, T_bits_rs_total, device->rs_total) ||
3010                             nla_put_u64(skb, T_bits_rs_failed, device->rs_failed))
3011                                 goto nla_put_failure;
3012                 }
3013         }
3014
3015         if (sib) {
3016                 switch(sib->sib_reason) {
3017                 case SIB_SYNC_PROGRESS:
3018                 case SIB_GET_STATUS_REPLY:
3019                         break;
3020                 case SIB_STATE_CHANGE:
3021                         if (nla_put_u32(skb, T_prev_state, sib->os.i) ||
3022                             nla_put_u32(skb, T_new_state, sib->ns.i))
3023                                 goto nla_put_failure;
3024                         break;
3025                 case SIB_HELPER_POST:
3026                         if (nla_put_u32(skb, T_helper_exit_code,
3027                                         sib->helper_exit_code))
3028                                 goto nla_put_failure;
3029                         /* fall through */
3030                 case SIB_HELPER_PRE:
3031                         if (nla_put_string(skb, T_helper, sib->helper_name))
3032                                 goto nla_put_failure;
3033                         break;
3034                 }
3035         }
3036         nla_nest_end(skb, nla);
3037
3038         if (0)
3039 nla_put_failure:
3040                 err = -EMSGSIZE;
3041         if (got_ldev)
3042                 put_ldev(device);
3043         return err;
3044 }
3045
3046 int drbd_adm_get_status(struct sk_buff *skb, struct genl_info *info)
3047 {
3048         struct drbd_config_context adm_ctx;
3049         enum drbd_ret_code retcode;
3050         int err;
3051
3052         retcode = drbd_adm_prepare(&adm_ctx, skb, info, DRBD_ADM_NEED_MINOR);
3053         if (!adm_ctx.reply_skb)
3054                 return retcode;
3055         if (retcode != NO_ERROR)
3056                 goto out;
3057
3058         err = nla_put_status_info(adm_ctx.reply_skb, adm_ctx.device, NULL);
3059         if (err) {
3060                 nlmsg_free(adm_ctx.reply_skb);
3061                 return err;
3062         }
3063 out:
3064         drbd_adm_finish(&adm_ctx, info, retcode);
3065         return 0;
3066 }
3067
3068 static int get_one_status(struct sk_buff *skb, struct netlink_callback *cb)
3069 {
3070         struct drbd_device *device;
3071         struct drbd_genlmsghdr *dh;
3072         struct drbd_resource *pos = (struct drbd_resource *)cb->args[0];
3073         struct drbd_resource *resource = NULL;
3074         struct drbd_resource *tmp;
3075         unsigned volume = cb->args[1];
3076
3077         /* Open coded, deferred, iteration:
3078          * for_each_resource_safe(resource, tmp, &drbd_resources) {
3079          *      connection = "first connection of resource or undefined";
3080          *      idr_for_each_entry(&resource->devices, device, i) {
3081          *        ...
3082          *      }
3083          * }
3084          * where resource is cb->args[0];
3085          * and i is cb->args[1];
3086          *
3087          * cb->args[2] indicates if we shall loop over all resources,
3088          * or just dump all volumes of a single resource.
3089          *
3090          * This may miss entries inserted after this dump started,
3091          * or entries deleted before they are reached.
3092          *
3093          * We need to make sure the device won't disappear while
3094          * we are looking at it, and revalidate our iterators
3095          * on each iteration.
3096          */
3097
3098         /* synchronize with conn_create()/drbd_destroy_connection() */
3099         rcu_read_lock();
3100         /* revalidate iterator position */
3101         for_each_resource_rcu(tmp, &drbd_resources) {
3102                 if (pos == NULL) {
3103                         /* first iteration */
3104                         pos = tmp;
3105                         resource = pos;
3106                         break;
3107                 }
3108                 if (tmp == pos) {
3109                         resource = pos;
3110                         break;
3111                 }
3112         }
3113         if (resource) {
3114 next_resource:
3115                 device = idr_get_next(&resource->devices, &volume);
3116                 if (!device) {
3117                         /* No more volumes to dump on this resource.
3118                          * Advance resource iterator. */
3119                         pos = list_entry_rcu(resource->resources.next,
3120                                              struct drbd_resource, resources);