Deleted Added
full compact
secondary.c (219721) secondary.c (219818)
1/*-
2 * Copyright (c) 2009-2010 The FreeBSD Foundation
3 * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
4 * All rights reserved.
5 *
6 * This software was developed by Pawel Jakub Dawidek under sponsorship from
7 * the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31#include <sys/cdefs.h>
1/*-
2 * Copyright (c) 2009-2010 The FreeBSD Foundation
3 * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
4 * All rights reserved.
5 *
6 * This software was developed by Pawel Jakub Dawidek under sponsorship from
7 * the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD: head/sbin/hastd/secondary.c 219721 2011-03-17 21:02:14Z trociny $");
32__FBSDID("$FreeBSD: head/sbin/hastd/secondary.c 219818 2011-03-21 08:54:59Z pjd $");
33
34#include <sys/param.h>
35#include <sys/time.h>
36#include <sys/bio.h>
37#include <sys/disk.h>
38#include <sys/stat.h>
39
40#include <err.h>
41#include <errno.h>
42#include <fcntl.h>
43#include <libgeom.h>
44#include <pthread.h>
45#include <signal.h>
46#include <stdint.h>
47#include <stdio.h>
48#include <string.h>
49#include <sysexits.h>
50#include <unistd.h>
51
52#include <activemap.h>
53#include <nv.h>
54#include <pjdlog.h>
55
56#include "control.h"
57#include "event.h"
58#include "hast.h"
59#include "hast_proto.h"
60#include "hastd.h"
61#include "hooks.h"
62#include "metadata.h"
63#include "proto.h"
64#include "subr.h"
65#include "synch.h"
66
67struct hio {
68 uint64_t hio_seq;
69 int hio_error;
70 struct nv *hio_nv;
71 void *hio_data;
72 uint8_t hio_cmd;
73 uint64_t hio_offset;
74 uint64_t hio_length;
75 TAILQ_ENTRY(hio) hio_next;
76};
77
78static struct hast_resource *gres;
79
80/*
81 * Free list holds unused structures. When free list is empty, we have to wait
82 * until some in-progress requests are freed.
83 */
84static TAILQ_HEAD(, hio) hio_free_list;
85static pthread_mutex_t hio_free_list_lock;
86static pthread_cond_t hio_free_list_cond;
87/*
88 * Disk thread (the one that do I/O requests) takes requests from this list.
89 */
90static TAILQ_HEAD(, hio) hio_disk_list;
91static pthread_mutex_t hio_disk_list_lock;
92static pthread_cond_t hio_disk_list_cond;
93/*
94 * There is one recv list for every component, although local components don't
95 * use recv lists as local requests are done synchronously.
96 */
97static TAILQ_HEAD(, hio) hio_send_list;
98static pthread_mutex_t hio_send_list_lock;
99static pthread_cond_t hio_send_list_cond;
100
101/*
102 * Maximum number of outstanding I/O requests.
103 */
104#define HAST_HIO_MAX 256
105
106static void *recv_thread(void *arg);
107static void *disk_thread(void *arg);
108static void *send_thread(void *arg);
109
110#define QUEUE_INSERT(name, hio) do { \
111 bool _wakeup; \
112 \
113 mtx_lock(&hio_##name##_list_lock); \
114 _wakeup = TAILQ_EMPTY(&hio_##name##_list); \
115 TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \
116 mtx_unlock(&hio_##name##_list_lock); \
117 if (_wakeup) \
118 cv_signal(&hio_##name##_list_cond); \
119} while (0)
120#define QUEUE_TAKE(name, hio) do { \
121 mtx_lock(&hio_##name##_list_lock); \
122 while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \
123 cv_wait(&hio_##name##_list_cond, \
124 &hio_##name##_list_lock); \
125 } \
126 TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \
127 mtx_unlock(&hio_##name##_list_lock); \
128} while (0)
129
130static void
131init_environment(void)
132{
133 struct hio *hio;
134 unsigned int ii;
135
136 /*
137 * Initialize lists, their locks and theirs condition variables.
138 */
139 TAILQ_INIT(&hio_free_list);
140 mtx_init(&hio_free_list_lock);
141 cv_init(&hio_free_list_cond);
142 TAILQ_INIT(&hio_disk_list);
143 mtx_init(&hio_disk_list_lock);
144 cv_init(&hio_disk_list_cond);
145 TAILQ_INIT(&hio_send_list);
146 mtx_init(&hio_send_list_lock);
147 cv_init(&hio_send_list_cond);
148
149 /*
150 * Allocate requests pool and initialize requests.
151 */
152 for (ii = 0; ii < HAST_HIO_MAX; ii++) {
153 hio = malloc(sizeof(*hio));
154 if (hio == NULL) {
155 pjdlog_exitx(EX_TEMPFAIL,
156 "Unable to allocate memory (%zu bytes) for hio request.",
157 sizeof(*hio));
158 }
159 hio->hio_error = 0;
160 hio->hio_data = malloc(MAXPHYS);
161 if (hio->hio_data == NULL) {
162 pjdlog_exitx(EX_TEMPFAIL,
163 "Unable to allocate memory (%zu bytes) for gctl_data.",
164 (size_t)MAXPHYS);
165 }
166 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);
167 }
168}
169
170static void
171init_local(struct hast_resource *res)
172{
173
174 if (metadata_read(res, true) < 0)
175 exit(EX_NOINPUT);
176}
177
178static void
179init_remote(struct hast_resource *res, struct nv *nvin)
180{
181 uint64_t resuid;
182 struct nv *nvout;
183 unsigned char *map;
184 size_t mapsize;
185
186 map = NULL;
187 mapsize = 0;
188 nvout = nv_alloc();
189 nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize");
190 nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize");
191 resuid = nv_get_uint64(nvin, "resuid");
192 res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt");
193 res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt");
194 nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt");
195 nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt");
196 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
197 METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize);
198 map = malloc(mapsize);
199 if (map == NULL) {
200 pjdlog_exitx(EX_TEMPFAIL,
201 "Unable to allocate memory (%zu bytes) for activemap.",
202 mapsize);
203 }
204 nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize");
205 /*
206 * When we work as primary and secondary is missing we will increase
207 * localcnt in our metadata. When secondary is connected and synced
208 * we make localcnt be equal to remotecnt, which means nodes are more
209 * or less in sync.
210 * Split-brain condition is when both nodes are not able to communicate
211 * and are both configured as primary nodes. In turn, they can both
212 * make incompatible changes to the data and we have to detect that.
213 * Under split-brain condition we will increase our localcnt on first
214 * write and remote node will increase its localcnt on first write.
215 * When we connect we can see that primary's localcnt is greater than
216 * our remotecnt (primary was modified while we weren't watching) and
217 * our localcnt is greater than primary's remotecnt (we were modified
218 * while primary wasn't watching).
219 * There are many possible combinations which are all gathered below.
220 * Don't pay too much attention to exact numbers, the more important
221 * is to compare them. We compare secondary's local with primary's
222 * remote and secondary's remote with primary's local.
223 * Note that every case where primary's localcnt is smaller than
224 * secondary's remotecnt and where secondary's localcnt is smaller than
225 * primary's remotecnt should be impossible in practise. We will perform
226 * full synchronization then. Those cases are marked with an asterisk.
227 * Regular synchronization means that only extents marked as dirty are
228 * synchronized (regular synchronization).
229 *
230 * SECONDARY METADATA PRIMARY METADATA
231 * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary.
232 * local=3 remote=3 local=2 remote=3* ?! Full sync from primary.
233 * local=3 remote=3 local=2 remote=4* ?! Full sync from primary.
234 * local=3 remote=3 local=3 remote=2 Primary is out-of-date,
235 * regular sync from secondary.
236 * local=3 remote=3 local=3 remote=3 Regular sync just in case.
237 * local=3 remote=3 local=3 remote=4* ?! Full sync from primary.
238 * local=3 remote=3 local=4 remote=2 Split-brain condition.
239 * local=3 remote=3 local=4 remote=3 Secondary out-of-date,
240 * regular sync from primary.
241 * local=3 remote=3 local=4 remote=4* ?! Full sync from primary.
242 */
243 if (res->hr_resuid == 0) {
244 /*
245 * Provider is used for the first time. If primary node done no
246 * writes yet as well (we will find "virgin" argument) then
247 * there is no need to synchronize anything. If primary node
248 * done any writes already we have to synchronize everything.
249 */
250 PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
251 res->hr_resuid = resuid;
252 if (metadata_write(res) < 0)
253 exit(EX_NOINPUT);
254 if (nv_exists(nvin, "virgin")) {
255 free(map);
256 map = NULL;
257 mapsize = 0;
258 } else {
259 memset(map, 0xff, mapsize);
260 }
261 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
262 } else if (
263 /* Is primary is out-of-date? */
264 (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
265 res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
266 /* Node are more or less in sync? */
267 (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
268 res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
269 /* Is secondary is out-of-date? */
270 (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
271 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) {
272 /*
273 * Nodes are more or less in sync or one of the nodes is
274 * out-of-date.
275 * It doesn't matter at this point which one, we just have to
276 * send out local bitmap to the remote node.
277 */
278 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) !=
279 (ssize_t)mapsize) {
280 pjdlog_exit(LOG_ERR, "Unable to read activemap");
281 }
282 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
283 res->hr_secondary_remotecnt == res->hr_primary_localcnt) {
284 /* Primary is out-of-date, sync from secondary. */
285 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
286 } else {
287 /*
288 * Secondary is out-of-date or counts match.
289 * Sync from primary.
290 */
291 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
292 }
293 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
294 res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
295 /*
296 * Not good, we have split-brain condition.
297 */
298 pjdlog_error("Split-brain detected, exiting.");
299 nv_add_string(nvout, "Split-brain condition!", "errmsg");
300 free(map);
301 map = NULL;
302 mapsize = 0;
303 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
304 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ {
305 /*
306 * This should never happen in practise, but we will perform
307 * full synchronization.
308 */
309 PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
310 res->hr_primary_localcnt < res->hr_secondary_remotecnt);
311 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
312 METADATA_SIZE, res->hr_extentsize,
313 res->hr_local_sectorsize);
314 memset(map, 0xff, mapsize);
315 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) {
316 /* In this one of five cases sync from secondary. */
317 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
318 } else {
319 /* For the rest four cases sync from primary. */
320 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
321 }
322 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).",
323 (uintmax_t)res->hr_primary_localcnt,
324 (uintmax_t)res->hr_primary_remotecnt,
325 (uintmax_t)res->hr_secondary_localcnt,
326 (uintmax_t)res->hr_secondary_remotecnt);
327 }
328 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) {
329 pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s",
330 res->hr_remoteaddr);
331 }
332 if (map != NULL)
333 free(map);
334 nv_free(nvout);
335 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
336 res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
337 /* Exit on split-brain. */
338 event_send(res, EVENT_SPLITBRAIN);
339 exit(EX_CONFIG);
340 }
341}
342
343void
344hastd_secondary(struct hast_resource *res, struct nv *nvin)
345{
346 sigset_t mask;
347 pthread_t td;
348 pid_t pid;
349 int error, mode, debuglevel;
350
351 /*
352 * Create communication channel between parent and child.
353 */
33
34#include <sys/param.h>
35#include <sys/time.h>
36#include <sys/bio.h>
37#include <sys/disk.h>
38#include <sys/stat.h>
39
40#include <err.h>
41#include <errno.h>
42#include <fcntl.h>
43#include <libgeom.h>
44#include <pthread.h>
45#include <signal.h>
46#include <stdint.h>
47#include <stdio.h>
48#include <string.h>
49#include <sysexits.h>
50#include <unistd.h>
51
52#include <activemap.h>
53#include <nv.h>
54#include <pjdlog.h>
55
56#include "control.h"
57#include "event.h"
58#include "hast.h"
59#include "hast_proto.h"
60#include "hastd.h"
61#include "hooks.h"
62#include "metadata.h"
63#include "proto.h"
64#include "subr.h"
65#include "synch.h"
66
67struct hio {
68 uint64_t hio_seq;
69 int hio_error;
70 struct nv *hio_nv;
71 void *hio_data;
72 uint8_t hio_cmd;
73 uint64_t hio_offset;
74 uint64_t hio_length;
75 TAILQ_ENTRY(hio) hio_next;
76};
77
78static struct hast_resource *gres;
79
80/*
81 * Free list holds unused structures. When free list is empty, we have to wait
82 * until some in-progress requests are freed.
83 */
84static TAILQ_HEAD(, hio) hio_free_list;
85static pthread_mutex_t hio_free_list_lock;
86static pthread_cond_t hio_free_list_cond;
87/*
88 * Disk thread (the one that do I/O requests) takes requests from this list.
89 */
90static TAILQ_HEAD(, hio) hio_disk_list;
91static pthread_mutex_t hio_disk_list_lock;
92static pthread_cond_t hio_disk_list_cond;
93/*
94 * There is one recv list for every component, although local components don't
95 * use recv lists as local requests are done synchronously.
96 */
97static TAILQ_HEAD(, hio) hio_send_list;
98static pthread_mutex_t hio_send_list_lock;
99static pthread_cond_t hio_send_list_cond;
100
101/*
102 * Maximum number of outstanding I/O requests.
103 */
104#define HAST_HIO_MAX 256
105
106static void *recv_thread(void *arg);
107static void *disk_thread(void *arg);
108static void *send_thread(void *arg);
109
110#define QUEUE_INSERT(name, hio) do { \
111 bool _wakeup; \
112 \
113 mtx_lock(&hio_##name##_list_lock); \
114 _wakeup = TAILQ_EMPTY(&hio_##name##_list); \
115 TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \
116 mtx_unlock(&hio_##name##_list_lock); \
117 if (_wakeup) \
118 cv_signal(&hio_##name##_list_cond); \
119} while (0)
120#define QUEUE_TAKE(name, hio) do { \
121 mtx_lock(&hio_##name##_list_lock); \
122 while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \
123 cv_wait(&hio_##name##_list_cond, \
124 &hio_##name##_list_lock); \
125 } \
126 TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \
127 mtx_unlock(&hio_##name##_list_lock); \
128} while (0)
129
130static void
131init_environment(void)
132{
133 struct hio *hio;
134 unsigned int ii;
135
136 /*
137 * Initialize lists, their locks and theirs condition variables.
138 */
139 TAILQ_INIT(&hio_free_list);
140 mtx_init(&hio_free_list_lock);
141 cv_init(&hio_free_list_cond);
142 TAILQ_INIT(&hio_disk_list);
143 mtx_init(&hio_disk_list_lock);
144 cv_init(&hio_disk_list_cond);
145 TAILQ_INIT(&hio_send_list);
146 mtx_init(&hio_send_list_lock);
147 cv_init(&hio_send_list_cond);
148
149 /*
150 * Allocate requests pool and initialize requests.
151 */
152 for (ii = 0; ii < HAST_HIO_MAX; ii++) {
153 hio = malloc(sizeof(*hio));
154 if (hio == NULL) {
155 pjdlog_exitx(EX_TEMPFAIL,
156 "Unable to allocate memory (%zu bytes) for hio request.",
157 sizeof(*hio));
158 }
159 hio->hio_error = 0;
160 hio->hio_data = malloc(MAXPHYS);
161 if (hio->hio_data == NULL) {
162 pjdlog_exitx(EX_TEMPFAIL,
163 "Unable to allocate memory (%zu bytes) for gctl_data.",
164 (size_t)MAXPHYS);
165 }
166 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);
167 }
168}
169
170static void
171init_local(struct hast_resource *res)
172{
173
174 if (metadata_read(res, true) < 0)
175 exit(EX_NOINPUT);
176}
177
178static void
179init_remote(struct hast_resource *res, struct nv *nvin)
180{
181 uint64_t resuid;
182 struct nv *nvout;
183 unsigned char *map;
184 size_t mapsize;
185
186 map = NULL;
187 mapsize = 0;
188 nvout = nv_alloc();
189 nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize");
190 nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize");
191 resuid = nv_get_uint64(nvin, "resuid");
192 res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt");
193 res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt");
194 nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt");
195 nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt");
196 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
197 METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize);
198 map = malloc(mapsize);
199 if (map == NULL) {
200 pjdlog_exitx(EX_TEMPFAIL,
201 "Unable to allocate memory (%zu bytes) for activemap.",
202 mapsize);
203 }
204 nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize");
205 /*
206 * When we work as primary and secondary is missing we will increase
207 * localcnt in our metadata. When secondary is connected and synced
208 * we make localcnt be equal to remotecnt, which means nodes are more
209 * or less in sync.
210 * Split-brain condition is when both nodes are not able to communicate
211 * and are both configured as primary nodes. In turn, they can both
212 * make incompatible changes to the data and we have to detect that.
213 * Under split-brain condition we will increase our localcnt on first
214 * write and remote node will increase its localcnt on first write.
215 * When we connect we can see that primary's localcnt is greater than
216 * our remotecnt (primary was modified while we weren't watching) and
217 * our localcnt is greater than primary's remotecnt (we were modified
218 * while primary wasn't watching).
219 * There are many possible combinations which are all gathered below.
220 * Don't pay too much attention to exact numbers, the more important
221 * is to compare them. We compare secondary's local with primary's
222 * remote and secondary's remote with primary's local.
223 * Note that every case where primary's localcnt is smaller than
224 * secondary's remotecnt and where secondary's localcnt is smaller than
225 * primary's remotecnt should be impossible in practise. We will perform
226 * full synchronization then. Those cases are marked with an asterisk.
227 * Regular synchronization means that only extents marked as dirty are
228 * synchronized (regular synchronization).
229 *
230 * SECONDARY METADATA PRIMARY METADATA
231 * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary.
232 * local=3 remote=3 local=2 remote=3* ?! Full sync from primary.
233 * local=3 remote=3 local=2 remote=4* ?! Full sync from primary.
234 * local=3 remote=3 local=3 remote=2 Primary is out-of-date,
235 * regular sync from secondary.
236 * local=3 remote=3 local=3 remote=3 Regular sync just in case.
237 * local=3 remote=3 local=3 remote=4* ?! Full sync from primary.
238 * local=3 remote=3 local=4 remote=2 Split-brain condition.
239 * local=3 remote=3 local=4 remote=3 Secondary out-of-date,
240 * regular sync from primary.
241 * local=3 remote=3 local=4 remote=4* ?! Full sync from primary.
242 */
243 if (res->hr_resuid == 0) {
244 /*
245 * Provider is used for the first time. If primary node done no
246 * writes yet as well (we will find "virgin" argument) then
247 * there is no need to synchronize anything. If primary node
248 * done any writes already we have to synchronize everything.
249 */
250 PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
251 res->hr_resuid = resuid;
252 if (metadata_write(res) < 0)
253 exit(EX_NOINPUT);
254 if (nv_exists(nvin, "virgin")) {
255 free(map);
256 map = NULL;
257 mapsize = 0;
258 } else {
259 memset(map, 0xff, mapsize);
260 }
261 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
262 } else if (
263 /* Is primary is out-of-date? */
264 (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
265 res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
266 /* Node are more or less in sync? */
267 (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
268 res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
269 /* Is secondary is out-of-date? */
270 (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
271 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) {
272 /*
273 * Nodes are more or less in sync or one of the nodes is
274 * out-of-date.
275 * It doesn't matter at this point which one, we just have to
276 * send out local bitmap to the remote node.
277 */
278 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) !=
279 (ssize_t)mapsize) {
280 pjdlog_exit(LOG_ERR, "Unable to read activemap");
281 }
282 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
283 res->hr_secondary_remotecnt == res->hr_primary_localcnt) {
284 /* Primary is out-of-date, sync from secondary. */
285 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
286 } else {
287 /*
288 * Secondary is out-of-date or counts match.
289 * Sync from primary.
290 */
291 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
292 }
293 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
294 res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
295 /*
296 * Not good, we have split-brain condition.
297 */
298 pjdlog_error("Split-brain detected, exiting.");
299 nv_add_string(nvout, "Split-brain condition!", "errmsg");
300 free(map);
301 map = NULL;
302 mapsize = 0;
303 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
304 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ {
305 /*
306 * This should never happen in practise, but we will perform
307 * full synchronization.
308 */
309 PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
310 res->hr_primary_localcnt < res->hr_secondary_remotecnt);
311 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
312 METADATA_SIZE, res->hr_extentsize,
313 res->hr_local_sectorsize);
314 memset(map, 0xff, mapsize);
315 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) {
316 /* In this one of five cases sync from secondary. */
317 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
318 } else {
319 /* For the rest four cases sync from primary. */
320 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
321 }
322 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).",
323 (uintmax_t)res->hr_primary_localcnt,
324 (uintmax_t)res->hr_primary_remotecnt,
325 (uintmax_t)res->hr_secondary_localcnt,
326 (uintmax_t)res->hr_secondary_remotecnt);
327 }
328 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) {
329 pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s",
330 res->hr_remoteaddr);
331 }
332 if (map != NULL)
333 free(map);
334 nv_free(nvout);
335 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
336 res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
337 /* Exit on split-brain. */
338 event_send(res, EVENT_SPLITBRAIN);
339 exit(EX_CONFIG);
340 }
341}
342
343void
344hastd_secondary(struct hast_resource *res, struct nv *nvin)
345{
346 sigset_t mask;
347 pthread_t td;
348 pid_t pid;
349 int error, mode, debuglevel;
350
351 /*
352 * Create communication channel between parent and child.
353 */
354 if (proto_client("socketpair://", &res->hr_ctrl) < 0) {
354 if (proto_client(NULL, "socketpair://", &res->hr_ctrl) < 0) {
355 KEEP_ERRNO((void)pidfile_remove(pfh));
356 pjdlog_exit(EX_OSERR,
357 "Unable to create control sockets between parent and child");
358 }
359 /*
360 * Create communication channel between child and parent.
361 */
355 KEEP_ERRNO((void)pidfile_remove(pfh));
356 pjdlog_exit(EX_OSERR,
357 "Unable to create control sockets between parent and child");
358 }
359 /*
360 * Create communication channel between child and parent.
361 */
362 if (proto_client("socketpair://", &res->hr_event) < 0) {
362 if (proto_client(NULL, "socketpair://", &res->hr_event) < 0) {
363 KEEP_ERRNO((void)pidfile_remove(pfh));
364 pjdlog_exit(EX_OSERR,
365 "Unable to create event sockets between child and parent");
366 }
367 /*
368 * Create communication channel for sending connection requests from
369 * parent to child.
370 */
363 KEEP_ERRNO((void)pidfile_remove(pfh));
364 pjdlog_exit(EX_OSERR,
365 "Unable to create event sockets between child and parent");
366 }
367 /*
368 * Create communication channel for sending connection requests from
369 * parent to child.
370 */
371 if (proto_client("socketpair://", &res->hr_conn) < 0) {
371 if (proto_client(NULL, "socketpair://", &res->hr_conn) < 0) {
372 /* TODO: There's no need for this to be fatal error. */
373 KEEP_ERRNO((void)pidfile_remove(pfh));
374 pjdlog_exit(EX_OSERR,
375 "Unable to create connection sockets between parent and child");
376 }
377
378 pid = fork();
379 if (pid < 0) {
380 KEEP_ERRNO((void)pidfile_remove(pfh));
381 pjdlog_exit(EX_OSERR, "Unable to fork");
382 }
383
384 if (pid > 0) {
385 /* This is parent. */
386 proto_close(res->hr_remotein);
387 res->hr_remotein = NULL;
388 proto_close(res->hr_remoteout);
389 res->hr_remoteout = NULL;
390 /* Declare that we are receiver. */
391 proto_recv(res->hr_event, NULL, 0);
392 /* Declare that we are sender. */
393 proto_send(res->hr_ctrl, NULL, 0);
394 proto_send(res->hr_conn, NULL, 0);
395 res->hr_workerpid = pid;
396 return;
397 }
398
399 gres = res;
400 mode = pjdlog_mode_get();
401 debuglevel = pjdlog_debug_get();
402
403 /* Declare that we are sender. */
404 proto_send(res->hr_event, NULL, 0);
405 /* Declare that we are receiver. */
406 proto_recv(res->hr_ctrl, NULL, 0);
407 proto_recv(res->hr_conn, NULL, 0);
408 descriptors_cleanup(res);
409
410 descriptors_assert(res, mode);
411
412 pjdlog_init(mode);
413 pjdlog_debug_set(debuglevel);
414 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
415 setproctitle("%s (secondary)", res->hr_name);
416
417 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
418 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
419
420 /* Error in setting timeout is not critical, but why should it fail? */
421 if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0)
422 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
423 if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0)
424 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
425
426 init_local(res);
427 init_environment();
428
429 if (drop_privs() != 0)
430 exit(EX_CONFIG);
431 pjdlog_info("Privileges successfully dropped.");
432
433 /*
434 * Create the control thread before sending any event to the parent,
435 * as we can deadlock when parent sends control request to worker,
436 * but worker has no control thread started yet, so parent waits.
437 * In the meantime worker sends an event to the parent, but parent
438 * is unable to handle the event, because it waits for control
439 * request response.
440 */
441 error = pthread_create(&td, NULL, ctrl_thread, res);
442 PJDLOG_ASSERT(error == 0);
443
444 init_remote(res, nvin);
445 event_send(res, EVENT_CONNECT);
446
447 error = pthread_create(&td, NULL, recv_thread, res);
448 PJDLOG_ASSERT(error == 0);
449 error = pthread_create(&td, NULL, disk_thread, res);
450 PJDLOG_ASSERT(error == 0);
451 (void)send_thread(res);
452}
453
454static void
455reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...)
456{
457 char msg[1024];
458 va_list ap;
459 int len;
460
461 va_start(ap, fmt);
462 len = vsnprintf(msg, sizeof(msg), fmt, ap);
463 va_end(ap);
464 if ((size_t)len < sizeof(msg)) {
465 switch (hio->hio_cmd) {
466 case HIO_READ:
467 (void)snprintf(msg + len, sizeof(msg) - len,
468 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset,
469 (uintmax_t)hio->hio_length);
470 break;
471 case HIO_DELETE:
472 (void)snprintf(msg + len, sizeof(msg) - len,
473 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset,
474 (uintmax_t)hio->hio_length);
475 break;
476 case HIO_FLUSH:
477 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH.");
478 break;
479 case HIO_WRITE:
480 (void)snprintf(msg + len, sizeof(msg) - len,
481 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset,
482 (uintmax_t)hio->hio_length);
483 break;
484 case HIO_KEEPALIVE:
485 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE.");
486 break;
487 default:
488 (void)snprintf(msg + len, sizeof(msg) - len,
489 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd);
490 break;
491 }
492 }
493 pjdlog_common(loglevel, debuglevel, error, "%s", msg);
494}
495
496static int
497requnpack(struct hast_resource *res, struct hio *hio)
498{
499
500 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd");
501 if (hio->hio_cmd == 0) {
502 pjdlog_error("Header contains no 'cmd' field.");
503 hio->hio_error = EINVAL;
504 goto end;
505 }
506 switch (hio->hio_cmd) {
507 case HIO_KEEPALIVE:
508 break;
509 case HIO_READ:
510 case HIO_WRITE:
511 case HIO_DELETE:
512 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset");
513 if (nv_error(hio->hio_nv) != 0) {
514 pjdlog_error("Header is missing 'offset' field.");
515 hio->hio_error = EINVAL;
516 goto end;
517 }
518 hio->hio_length = nv_get_uint64(hio->hio_nv, "length");
519 if (nv_error(hio->hio_nv) != 0) {
520 pjdlog_error("Header is missing 'length' field.");
521 hio->hio_error = EINVAL;
522 goto end;
523 }
524 if (hio->hio_length == 0) {
525 pjdlog_error("Data length is zero.");
526 hio->hio_error = EINVAL;
527 goto end;
528 }
529 if (hio->hio_length > MAXPHYS) {
530 pjdlog_error("Data length is too large (%ju > %ju).",
531 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);
532 hio->hio_error = EINVAL;
533 goto end;
534 }
535 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) {
536 pjdlog_error("Offset %ju is not multiple of sector size.",
537 (uintmax_t)hio->hio_offset);
538 hio->hio_error = EINVAL;
539 goto end;
540 }
541 if ((hio->hio_length % res->hr_local_sectorsize) != 0) {
542 pjdlog_error("Length %ju is not multiple of sector size.",
543 (uintmax_t)hio->hio_length);
544 hio->hio_error = EINVAL;
545 goto end;
546 }
547 if (hio->hio_offset + hio->hio_length >
548 (uint64_t)res->hr_datasize) {
549 pjdlog_error("Data offset is too large (%ju > %ju).",
550 (uintmax_t)(hio->hio_offset + hio->hio_length),
551 (uintmax_t)res->hr_datasize);
552 hio->hio_error = EINVAL;
553 goto end;
554 }
555 break;
556 default:
557 pjdlog_error("Header contains invalid 'cmd' (%hhu).",
558 hio->hio_cmd);
559 hio->hio_error = EINVAL;
560 goto end;
561 }
562 hio->hio_error = 0;
563end:
564 return (hio->hio_error);
565}
566
567static __dead2 void
568secondary_exit(int exitcode, const char *fmt, ...)
569{
570 va_list ap;
571
572 PJDLOG_ASSERT(exitcode != EX_OK);
573 va_start(ap, fmt);
574 pjdlogv_errno(LOG_ERR, fmt, ap);
575 va_end(ap);
576 event_send(gres, EVENT_DISCONNECT);
577 exit(exitcode);
578}
579
580/*
581 * Thread receives requests from the primary node.
582 */
583static void *
584recv_thread(void *arg)
585{
586 struct hast_resource *res = arg;
587 struct hio *hio;
588
589 for (;;) {
590 pjdlog_debug(2, "recv: Taking free request.");
591 QUEUE_TAKE(free, hio);
592 pjdlog_debug(2, "recv: (%p) Got request.", hio);
593 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) {
594 secondary_exit(EX_TEMPFAIL,
595 "Unable to receive request header");
596 }
597 if (requnpack(res, hio) != 0) {
598 pjdlog_debug(2,
599 "recv: (%p) Moving request to the send queue.",
600 hio);
601 QUEUE_INSERT(send, hio);
602 continue;
603 }
604 reqlog(LOG_DEBUG, 2, -1, hio,
605 "recv: (%p) Got request header: ", hio);
606 if (hio->hio_cmd == HIO_KEEPALIVE) {
607 pjdlog_debug(2,
608 "recv: (%p) Moving request to the free queue.",
609 hio);
610 nv_free(hio->hio_nv);
611 QUEUE_INSERT(free, hio);
612 continue;
613 } else if (hio->hio_cmd == HIO_WRITE) {
614 if (hast_proto_recv_data(res, res->hr_remotein,
615 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) {
616 secondary_exit(EX_TEMPFAIL,
617 "Unable to receive request data");
618 }
619 }
620 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.",
621 hio);
622 QUEUE_INSERT(disk, hio);
623 }
624 /* NOTREACHED */
625 return (NULL);
626}
627
628/*
629 * Thread reads from or writes to local component and also handles DELETE and
630 * FLUSH requests.
631 */
632static void *
633disk_thread(void *arg)
634{
635 struct hast_resource *res = arg;
636 struct hio *hio;
637 ssize_t ret;
638 bool clear_activemap;
639
640 clear_activemap = true;
641
642 for (;;) {
643 pjdlog_debug(2, "disk: Taking request.");
644 QUEUE_TAKE(disk, hio);
645 while (clear_activemap) {
646 unsigned char *map;
647 size_t mapsize;
648
649 /*
650 * When first request is received, it means that primary
651 * already received our activemap, merged it and stored
652 * locally. We can now safely clear our activemap.
653 */
654 mapsize =
655 activemap_calc_ondisk_size(res->hr_local_mediasize -
656 METADATA_SIZE, res->hr_extentsize,
657 res->hr_local_sectorsize);
658 map = calloc(1, mapsize);
659 if (map == NULL) {
660 pjdlog_warning("Unable to allocate memory to clear local activemap.");
661 break;
662 }
663 if (pwrite(res->hr_localfd, map, mapsize,
664 METADATA_SIZE) != (ssize_t)mapsize) {
665 pjdlog_errno(LOG_WARNING,
666 "Unable to store cleared activemap");
667 free(map);
668 break;
669 }
670 free(map);
671 clear_activemap = false;
672 pjdlog_debug(1, "Local activemap cleared.");
673 }
674 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio);
675 /* Handle the actual request. */
676 switch (hio->hio_cmd) {
677 case HIO_READ:
678 ret = pread(res->hr_localfd, hio->hio_data,
679 hio->hio_length,
680 hio->hio_offset + res->hr_localoff);
681 if (ret < 0)
682 hio->hio_error = errno;
683 else if (ret != (int64_t)hio->hio_length)
684 hio->hio_error = EIO;
685 else
686 hio->hio_error = 0;
687 break;
688 case HIO_WRITE:
689 ret = pwrite(res->hr_localfd, hio->hio_data,
690 hio->hio_length,
691 hio->hio_offset + res->hr_localoff);
692 if (ret < 0)
693 hio->hio_error = errno;
694 else if (ret != (int64_t)hio->hio_length)
695 hio->hio_error = EIO;
696 else
697 hio->hio_error = 0;
698 break;
699 case HIO_DELETE:
700 ret = g_delete(res->hr_localfd,
701 hio->hio_offset + res->hr_localoff,
702 hio->hio_length);
703 if (ret < 0)
704 hio->hio_error = errno;
705 else
706 hio->hio_error = 0;
707 break;
708 case HIO_FLUSH:
709 ret = g_flush(res->hr_localfd);
710 if (ret < 0)
711 hio->hio_error = errno;
712 else
713 hio->hio_error = 0;
714 break;
715 }
716 if (hio->hio_error != 0) {
717 reqlog(LOG_ERR, 0, hio->hio_error, hio,
718 "Request failed: ");
719 }
720 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.",
721 hio);
722 QUEUE_INSERT(send, hio);
723 }
724 /* NOTREACHED */
725 return (NULL);
726}
727
728/*
729 * Thread sends requests back to primary node.
730 */
731static void *
732send_thread(void *arg)
733{
734 struct hast_resource *res = arg;
735 struct nv *nvout;
736 struct hio *hio;
737 void *data;
738 size_t length;
739
740 for (;;) {
741 pjdlog_debug(2, "send: Taking request.");
742 QUEUE_TAKE(send, hio);
743 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio);
744 nvout = nv_alloc();
745 /* Copy sequence number. */
746 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq");
747 switch (hio->hio_cmd) {
748 case HIO_READ:
749 if (hio->hio_error == 0) {
750 data = hio->hio_data;
751 length = hio->hio_length;
752 break;
753 }
754 /*
755 * We send no data in case of an error.
756 */
757 /* FALLTHROUGH */
758 case HIO_DELETE:
759 case HIO_FLUSH:
760 case HIO_WRITE:
761 data = NULL;
762 length = 0;
763 break;
764 default:
765 abort();
766 break;
767 }
768 if (hio->hio_error != 0)
769 nv_add_int16(nvout, hio->hio_error, "error");
770 if (hast_proto_send(res, res->hr_remoteout, nvout, data,
771 length) < 0) {
772 secondary_exit(EX_TEMPFAIL, "Unable to send reply.");
773 }
774 nv_free(nvout);
775 pjdlog_debug(2, "send: (%p) Moving request to the free queue.",
776 hio);
777 nv_free(hio->hio_nv);
778 hio->hio_error = 0;
779 QUEUE_INSERT(free, hio);
780 }
781 /* NOTREACHED */
782 return (NULL);
783}
372 /* TODO: There's no need for this to be fatal error. */
373 KEEP_ERRNO((void)pidfile_remove(pfh));
374 pjdlog_exit(EX_OSERR,
375 "Unable to create connection sockets between parent and child");
376 }
377
378 pid = fork();
379 if (pid < 0) {
380 KEEP_ERRNO((void)pidfile_remove(pfh));
381 pjdlog_exit(EX_OSERR, "Unable to fork");
382 }
383
384 if (pid > 0) {
385 /* This is parent. */
386 proto_close(res->hr_remotein);
387 res->hr_remotein = NULL;
388 proto_close(res->hr_remoteout);
389 res->hr_remoteout = NULL;
390 /* Declare that we are receiver. */
391 proto_recv(res->hr_event, NULL, 0);
392 /* Declare that we are sender. */
393 proto_send(res->hr_ctrl, NULL, 0);
394 proto_send(res->hr_conn, NULL, 0);
395 res->hr_workerpid = pid;
396 return;
397 }
398
399 gres = res;
400 mode = pjdlog_mode_get();
401 debuglevel = pjdlog_debug_get();
402
403 /* Declare that we are sender. */
404 proto_send(res->hr_event, NULL, 0);
405 /* Declare that we are receiver. */
406 proto_recv(res->hr_ctrl, NULL, 0);
407 proto_recv(res->hr_conn, NULL, 0);
408 descriptors_cleanup(res);
409
410 descriptors_assert(res, mode);
411
412 pjdlog_init(mode);
413 pjdlog_debug_set(debuglevel);
414 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
415 setproctitle("%s (secondary)", res->hr_name);
416
417 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
418 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
419
420 /* Error in setting timeout is not critical, but why should it fail? */
421 if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0)
422 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
423 if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0)
424 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
425
426 init_local(res);
427 init_environment();
428
429 if (drop_privs() != 0)
430 exit(EX_CONFIG);
431 pjdlog_info("Privileges successfully dropped.");
432
433 /*
434 * Create the control thread before sending any event to the parent,
435 * as we can deadlock when parent sends control request to worker,
436 * but worker has no control thread started yet, so parent waits.
437 * In the meantime worker sends an event to the parent, but parent
438 * is unable to handle the event, because it waits for control
439 * request response.
440 */
441 error = pthread_create(&td, NULL, ctrl_thread, res);
442 PJDLOG_ASSERT(error == 0);
443
444 init_remote(res, nvin);
445 event_send(res, EVENT_CONNECT);
446
447 error = pthread_create(&td, NULL, recv_thread, res);
448 PJDLOG_ASSERT(error == 0);
449 error = pthread_create(&td, NULL, disk_thread, res);
450 PJDLOG_ASSERT(error == 0);
451 (void)send_thread(res);
452}
453
454static void
455reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...)
456{
457 char msg[1024];
458 va_list ap;
459 int len;
460
461 va_start(ap, fmt);
462 len = vsnprintf(msg, sizeof(msg), fmt, ap);
463 va_end(ap);
464 if ((size_t)len < sizeof(msg)) {
465 switch (hio->hio_cmd) {
466 case HIO_READ:
467 (void)snprintf(msg + len, sizeof(msg) - len,
468 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset,
469 (uintmax_t)hio->hio_length);
470 break;
471 case HIO_DELETE:
472 (void)snprintf(msg + len, sizeof(msg) - len,
473 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset,
474 (uintmax_t)hio->hio_length);
475 break;
476 case HIO_FLUSH:
477 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH.");
478 break;
479 case HIO_WRITE:
480 (void)snprintf(msg + len, sizeof(msg) - len,
481 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset,
482 (uintmax_t)hio->hio_length);
483 break;
484 case HIO_KEEPALIVE:
485 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE.");
486 break;
487 default:
488 (void)snprintf(msg + len, sizeof(msg) - len,
489 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd);
490 break;
491 }
492 }
493 pjdlog_common(loglevel, debuglevel, error, "%s", msg);
494}
495
496static int
497requnpack(struct hast_resource *res, struct hio *hio)
498{
499
500 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd");
501 if (hio->hio_cmd == 0) {
502 pjdlog_error("Header contains no 'cmd' field.");
503 hio->hio_error = EINVAL;
504 goto end;
505 }
506 switch (hio->hio_cmd) {
507 case HIO_KEEPALIVE:
508 break;
509 case HIO_READ:
510 case HIO_WRITE:
511 case HIO_DELETE:
512 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset");
513 if (nv_error(hio->hio_nv) != 0) {
514 pjdlog_error("Header is missing 'offset' field.");
515 hio->hio_error = EINVAL;
516 goto end;
517 }
518 hio->hio_length = nv_get_uint64(hio->hio_nv, "length");
519 if (nv_error(hio->hio_nv) != 0) {
520 pjdlog_error("Header is missing 'length' field.");
521 hio->hio_error = EINVAL;
522 goto end;
523 }
524 if (hio->hio_length == 0) {
525 pjdlog_error("Data length is zero.");
526 hio->hio_error = EINVAL;
527 goto end;
528 }
529 if (hio->hio_length > MAXPHYS) {
530 pjdlog_error("Data length is too large (%ju > %ju).",
531 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);
532 hio->hio_error = EINVAL;
533 goto end;
534 }
535 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) {
536 pjdlog_error("Offset %ju is not multiple of sector size.",
537 (uintmax_t)hio->hio_offset);
538 hio->hio_error = EINVAL;
539 goto end;
540 }
541 if ((hio->hio_length % res->hr_local_sectorsize) != 0) {
542 pjdlog_error("Length %ju is not multiple of sector size.",
543 (uintmax_t)hio->hio_length);
544 hio->hio_error = EINVAL;
545 goto end;
546 }
547 if (hio->hio_offset + hio->hio_length >
548 (uint64_t)res->hr_datasize) {
549 pjdlog_error("Data offset is too large (%ju > %ju).",
550 (uintmax_t)(hio->hio_offset + hio->hio_length),
551 (uintmax_t)res->hr_datasize);
552 hio->hio_error = EINVAL;
553 goto end;
554 }
555 break;
556 default:
557 pjdlog_error("Header contains invalid 'cmd' (%hhu).",
558 hio->hio_cmd);
559 hio->hio_error = EINVAL;
560 goto end;
561 }
562 hio->hio_error = 0;
563end:
564 return (hio->hio_error);
565}
566
567static __dead2 void
568secondary_exit(int exitcode, const char *fmt, ...)
569{
570 va_list ap;
571
572 PJDLOG_ASSERT(exitcode != EX_OK);
573 va_start(ap, fmt);
574 pjdlogv_errno(LOG_ERR, fmt, ap);
575 va_end(ap);
576 event_send(gres, EVENT_DISCONNECT);
577 exit(exitcode);
578}
579
580/*
581 * Thread receives requests from the primary node.
582 */
583static void *
584recv_thread(void *arg)
585{
586 struct hast_resource *res = arg;
587 struct hio *hio;
588
589 for (;;) {
590 pjdlog_debug(2, "recv: Taking free request.");
591 QUEUE_TAKE(free, hio);
592 pjdlog_debug(2, "recv: (%p) Got request.", hio);
593 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) {
594 secondary_exit(EX_TEMPFAIL,
595 "Unable to receive request header");
596 }
597 if (requnpack(res, hio) != 0) {
598 pjdlog_debug(2,
599 "recv: (%p) Moving request to the send queue.",
600 hio);
601 QUEUE_INSERT(send, hio);
602 continue;
603 }
604 reqlog(LOG_DEBUG, 2, -1, hio,
605 "recv: (%p) Got request header: ", hio);
606 if (hio->hio_cmd == HIO_KEEPALIVE) {
607 pjdlog_debug(2,
608 "recv: (%p) Moving request to the free queue.",
609 hio);
610 nv_free(hio->hio_nv);
611 QUEUE_INSERT(free, hio);
612 continue;
613 } else if (hio->hio_cmd == HIO_WRITE) {
614 if (hast_proto_recv_data(res, res->hr_remotein,
615 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) {
616 secondary_exit(EX_TEMPFAIL,
617 "Unable to receive request data");
618 }
619 }
620 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.",
621 hio);
622 QUEUE_INSERT(disk, hio);
623 }
624 /* NOTREACHED */
625 return (NULL);
626}
627
628/*
629 * Thread reads from or writes to local component and also handles DELETE and
630 * FLUSH requests.
631 */
632static void *
633disk_thread(void *arg)
634{
635 struct hast_resource *res = arg;
636 struct hio *hio;
637 ssize_t ret;
638 bool clear_activemap;
639
640 clear_activemap = true;
641
642 for (;;) {
643 pjdlog_debug(2, "disk: Taking request.");
644 QUEUE_TAKE(disk, hio);
645 while (clear_activemap) {
646 unsigned char *map;
647 size_t mapsize;
648
649 /*
650 * When first request is received, it means that primary
651 * already received our activemap, merged it and stored
652 * locally. We can now safely clear our activemap.
653 */
654 mapsize =
655 activemap_calc_ondisk_size(res->hr_local_mediasize -
656 METADATA_SIZE, res->hr_extentsize,
657 res->hr_local_sectorsize);
658 map = calloc(1, mapsize);
659 if (map == NULL) {
660 pjdlog_warning("Unable to allocate memory to clear local activemap.");
661 break;
662 }
663 if (pwrite(res->hr_localfd, map, mapsize,
664 METADATA_SIZE) != (ssize_t)mapsize) {
665 pjdlog_errno(LOG_WARNING,
666 "Unable to store cleared activemap");
667 free(map);
668 break;
669 }
670 free(map);
671 clear_activemap = false;
672 pjdlog_debug(1, "Local activemap cleared.");
673 }
674 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio);
675 /* Handle the actual request. */
676 switch (hio->hio_cmd) {
677 case HIO_READ:
678 ret = pread(res->hr_localfd, hio->hio_data,
679 hio->hio_length,
680 hio->hio_offset + res->hr_localoff);
681 if (ret < 0)
682 hio->hio_error = errno;
683 else if (ret != (int64_t)hio->hio_length)
684 hio->hio_error = EIO;
685 else
686 hio->hio_error = 0;
687 break;
688 case HIO_WRITE:
689 ret = pwrite(res->hr_localfd, hio->hio_data,
690 hio->hio_length,
691 hio->hio_offset + res->hr_localoff);
692 if (ret < 0)
693 hio->hio_error = errno;
694 else if (ret != (int64_t)hio->hio_length)
695 hio->hio_error = EIO;
696 else
697 hio->hio_error = 0;
698 break;
699 case HIO_DELETE:
700 ret = g_delete(res->hr_localfd,
701 hio->hio_offset + res->hr_localoff,
702 hio->hio_length);
703 if (ret < 0)
704 hio->hio_error = errno;
705 else
706 hio->hio_error = 0;
707 break;
708 case HIO_FLUSH:
709 ret = g_flush(res->hr_localfd);
710 if (ret < 0)
711 hio->hio_error = errno;
712 else
713 hio->hio_error = 0;
714 break;
715 }
716 if (hio->hio_error != 0) {
717 reqlog(LOG_ERR, 0, hio->hio_error, hio,
718 "Request failed: ");
719 }
720 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.",
721 hio);
722 QUEUE_INSERT(send, hio);
723 }
724 /* NOTREACHED */
725 return (NULL);
726}
727
728/*
729 * Thread sends requests back to primary node.
730 */
731static void *
732send_thread(void *arg)
733{
734 struct hast_resource *res = arg;
735 struct nv *nvout;
736 struct hio *hio;
737 void *data;
738 size_t length;
739
740 for (;;) {
741 pjdlog_debug(2, "send: Taking request.");
742 QUEUE_TAKE(send, hio);
743 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio);
744 nvout = nv_alloc();
745 /* Copy sequence number. */
746 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq");
747 switch (hio->hio_cmd) {
748 case HIO_READ:
749 if (hio->hio_error == 0) {
750 data = hio->hio_data;
751 length = hio->hio_length;
752 break;
753 }
754 /*
755 * We send no data in case of an error.
756 */
757 /* FALLTHROUGH */
758 case HIO_DELETE:
759 case HIO_FLUSH:
760 case HIO_WRITE:
761 data = NULL;
762 length = 0;
763 break;
764 default:
765 abort();
766 break;
767 }
768 if (hio->hio_error != 0)
769 nv_add_int16(nvout, hio->hio_error, "error");
770 if (hast_proto_send(res, res->hr_remoteout, nvout, data,
771 length) < 0) {
772 secondary_exit(EX_TEMPFAIL, "Unable to send reply.");
773 }
774 nv_free(nvout);
775 pjdlog_debug(2, "send: (%p) Moving request to the free queue.",
776 hio);
777 nv_free(hio->hio_nv);
778 hio->hio_error = 0;
779 QUEUE_INSERT(free, hio);
780 }
781 /* NOTREACHED */
782 return (NULL);
783}