Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 1 | /* |
| 2 | * QEMU Block driver for RADOS (Ceph) |
| 3 | * |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 4 | * Copyright (C) 2010-2011 Christian Brunner <chb@muc.de>, |
| 5 | * Josh Durgin <josh.durgin@dreamhost.com> |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 6 | * |
| 7 | * This work is licensed under the terms of the GNU GPL, version 2. See |
| 8 | * the COPYING file in the top-level directory. |
| 9 | * |
Paolo Bonzini | 6b620ca | 2012-01-13 17:44:23 +0100 | [diff] [blame] | 10 | * Contributions after 2012-01-13 are licensed under the terms of the |
| 11 | * GNU GPL, version 2 or (at your option) any later version. |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 12 | */ |
| 13 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 14 | #include <inttypes.h> |
| 15 | |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 16 | #include "qemu-common.h" |
Paolo Bonzini | 1de7afc | 2012-12-17 18:20:00 +0100 | [diff] [blame] | 17 | #include "qemu/error-report.h" |
Paolo Bonzini | 737e150 | 2012-12-17 18:19:44 +0100 | [diff] [blame] | 18 | #include "block/block_int.h" |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 19 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 20 | #include <rbd/librbd.h> |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 21 | |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 22 | /* |
| 23 | * When specifying the image filename use: |
| 24 | * |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 25 | * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]] |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 26 | * |
Sage Weil | 9e1fbcd | 2011-09-15 14:11:10 -0700 | [diff] [blame] | 27 | * poolname must be the name of an existing rados pool. |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 28 | * |
Sage Weil | 9e1fbcd | 2011-09-15 14:11:10 -0700 | [diff] [blame] | 29 | * devicename is the name of the rbd image. |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 30 | * |
Sage Weil | 9e1fbcd | 2011-09-15 14:11:10 -0700 | [diff] [blame] | 31 | * Each option given is used to configure rados, and may be any valid |
| 32 | * Ceph option, "id", or "conf". |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 33 | * |
Sage Weil | 9e1fbcd | 2011-09-15 14:11:10 -0700 | [diff] [blame] | 34 | * The "id" option indicates what user we should authenticate as to |
| 35 | * the Ceph cluster. If it is excluded we will use the Ceph default |
| 36 | * (normally 'admin'). |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 37 | * |
Sage Weil | 9e1fbcd | 2011-09-15 14:11:10 -0700 | [diff] [blame] | 38 | * The "conf" option specifies a Ceph configuration file to read. If |
| 39 | * it is not specified, we will read from the default Ceph locations |
| 40 | * (e.g., /etc/ceph/ceph.conf). To avoid reading _any_ configuration |
| 41 | * file, specify conf=/dev/null. |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 42 | * |
Sage Weil | 9e1fbcd | 2011-09-15 14:11:10 -0700 | [diff] [blame] | 43 | * Configuration values containing :, @, or = can be escaped with a |
| 44 | * leading "\". |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 45 | */ |
| 46 | |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 47 | /* rbd_aio_discard added in 0.1.2 */ |
| 48 | #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2) |
| 49 | #define LIBRBD_SUPPORTS_DISCARD |
| 50 | #else |
| 51 | #undef LIBRBD_SUPPORTS_DISCARD |
| 52 | #endif |
| 53 | |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 54 | #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER) |
| 55 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 56 | #define RBD_MAX_CONF_NAME_SIZE 128 |
| 57 | #define RBD_MAX_CONF_VAL_SIZE 512 |
| 58 | #define RBD_MAX_CONF_SIZE 1024 |
| 59 | #define RBD_MAX_POOL_NAME_SIZE 128 |
| 60 | #define RBD_MAX_SNAP_NAME_SIZE 128 |
| 61 | #define RBD_MAX_SNAPS 100 |
| 62 | |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 63 | typedef enum { |
| 64 | RBD_AIO_READ, |
| 65 | RBD_AIO_WRITE, |
| 66 | RBD_AIO_DISCARD |
| 67 | } RBDAIOCmd; |
| 68 | |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 69 | typedef struct RBDAIOCB { |
| 70 | BlockDriverAIOCB common; |
| 71 | QEMUBH *bh; |
Stefan Priebe | 08448d5 | 2012-11-20 13:44:55 +0100 | [diff] [blame] | 72 | int64_t ret; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 73 | QEMUIOVector *qiov; |
| 74 | char *bounce; |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 75 | RBDAIOCmd cmd; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 76 | int64_t sector_num; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 77 | int error; |
| 78 | struct BDRVRBDState *s; |
| 79 | int cancelled; |
Stefan Priebe | 473c7f0 | 2012-11-30 09:55:46 +0100 | [diff] [blame] | 80 | int status; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 81 | } RBDAIOCB; |
| 82 | |
| 83 | typedef struct RADOSCB { |
| 84 | int rcbid; |
| 85 | RBDAIOCB *acb; |
| 86 | struct BDRVRBDState *s; |
| 87 | int done; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 88 | int64_t size; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 89 | char *buf; |
Stefan Priebe | 08448d5 | 2012-11-20 13:44:55 +0100 | [diff] [blame] | 90 | int64_t ret; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 91 | } RADOSCB; |
| 92 | |
| 93 | #define RBD_FD_READ 0 |
| 94 | #define RBD_FD_WRITE 1 |
| 95 | |
| 96 | typedef struct BDRVRBDState { |
| 97 | int fds[2]; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 98 | rados_t cluster; |
| 99 | rados_ioctx_t io_ctx; |
| 100 | rbd_image_t image; |
| 101 | char name[RBD_MAX_IMAGE_NAME_SIZE]; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 102 | int qemu_aio_count; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 103 | char *snap; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 104 | int event_reader_pos; |
| 105 | RADOSCB *event_rcb; |
| 106 | } BDRVRBDState; |
| 107 | |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 108 | static void rbd_aio_bh_cb(void *opaque); |
| 109 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 110 | static int qemu_rbd_next_tok(char *dst, int dst_len, |
| 111 | char *src, char delim, |
| 112 | const char *name, |
| 113 | char **p) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 114 | { |
| 115 | int l; |
| 116 | char *end; |
| 117 | |
| 118 | *p = NULL; |
| 119 | |
| 120 | if (delim != '\0') { |
Sage Weil | 16a06b2 | 2011-09-19 13:35:26 -0700 | [diff] [blame] | 121 | for (end = src; *end; ++end) { |
| 122 | if (*end == delim) { |
| 123 | break; |
| 124 | } |
| 125 | if (*end == '\\' && end[1] != '\0') { |
| 126 | end++; |
| 127 | } |
| 128 | } |
| 129 | if (*end == delim) { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 130 | *p = end + 1; |
| 131 | *end = '\0'; |
| 132 | } |
| 133 | } |
| 134 | l = strlen(src); |
| 135 | if (l >= dst_len) { |
| 136 | error_report("%s too long", name); |
| 137 | return -EINVAL; |
| 138 | } else if (l == 0) { |
| 139 | error_report("%s too short", name); |
| 140 | return -EINVAL; |
| 141 | } |
| 142 | |
| 143 | pstrcpy(dst, dst_len, src); |
| 144 | |
| 145 | return 0; |
| 146 | } |
| 147 | |
Sage Weil | 16a06b2 | 2011-09-19 13:35:26 -0700 | [diff] [blame] | 148 | static void qemu_rbd_unescape(char *src) |
| 149 | { |
| 150 | char *p; |
| 151 | |
| 152 | for (p = src; *src; ++src, ++p) { |
| 153 | if (*src == '\\' && src[1] != '\0') { |
| 154 | src++; |
| 155 | } |
| 156 | *p = *src; |
| 157 | } |
| 158 | *p = '\0'; |
| 159 | } |
| 160 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 161 | static int qemu_rbd_parsename(const char *filename, |
| 162 | char *pool, int pool_len, |
| 163 | char *snap, int snap_len, |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 164 | char *name, int name_len, |
| 165 | char *conf, int conf_len) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 166 | { |
| 167 | const char *start; |
| 168 | char *p, *buf; |
| 169 | int ret; |
| 170 | |
| 171 | if (!strstart(filename, "rbd:", &start)) { |
| 172 | return -EINVAL; |
| 173 | } |
| 174 | |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 175 | buf = g_strdup(start); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 176 | p = buf; |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 177 | *snap = '\0'; |
| 178 | *conf = '\0'; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 179 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 180 | ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 181 | if (ret < 0 || !p) { |
| 182 | ret = -EINVAL; |
| 183 | goto done; |
| 184 | } |
Sage Weil | 16a06b2 | 2011-09-19 13:35:26 -0700 | [diff] [blame] | 185 | qemu_rbd_unescape(pool); |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 186 | |
| 187 | if (strchr(p, '@')) { |
| 188 | ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p); |
| 189 | if (ret < 0) { |
| 190 | goto done; |
| 191 | } |
| 192 | ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p); |
Sage Weil | 16a06b2 | 2011-09-19 13:35:26 -0700 | [diff] [blame] | 193 | qemu_rbd_unescape(snap); |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 194 | } else { |
| 195 | ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 196 | } |
Sage Weil | 16a06b2 | 2011-09-19 13:35:26 -0700 | [diff] [blame] | 197 | qemu_rbd_unescape(name); |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 198 | if (ret < 0 || !p) { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 199 | goto done; |
| 200 | } |
| 201 | |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 202 | ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 203 | |
| 204 | done: |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 205 | g_free(buf); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 206 | return ret; |
| 207 | } |
| 208 | |
Sage Weil | 7c7e9df | 2011-09-07 09:28:04 -0700 | [diff] [blame] | 209 | static char *qemu_rbd_parse_clientname(const char *conf, char *clientname) |
| 210 | { |
| 211 | const char *p = conf; |
| 212 | |
| 213 | while (*p) { |
| 214 | int len; |
| 215 | const char *end = strchr(p, ':'); |
| 216 | |
| 217 | if (end) { |
| 218 | len = end - p; |
| 219 | } else { |
| 220 | len = strlen(p); |
| 221 | } |
| 222 | |
| 223 | if (strncmp(p, "id=", 3) == 0) { |
| 224 | len -= 3; |
| 225 | strncpy(clientname, p + 3, len); |
| 226 | clientname[len] = '\0'; |
| 227 | return clientname; |
| 228 | } |
| 229 | if (end == NULL) { |
| 230 | break; |
| 231 | } |
| 232 | p = end + 1; |
| 233 | } |
| 234 | return NULL; |
| 235 | } |
| 236 | |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 237 | static int qemu_rbd_set_conf(rados_t cluster, const char *conf) |
| 238 | { |
| 239 | char *p, *buf; |
| 240 | char name[RBD_MAX_CONF_NAME_SIZE]; |
| 241 | char value[RBD_MAX_CONF_VAL_SIZE]; |
| 242 | int ret = 0; |
| 243 | |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 244 | buf = g_strdup(conf); |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 245 | p = buf; |
| 246 | |
| 247 | while (p) { |
| 248 | ret = qemu_rbd_next_tok(name, sizeof(name), p, |
| 249 | '=', "conf option name", &p); |
| 250 | if (ret < 0) { |
| 251 | break; |
| 252 | } |
Sage Weil | 16a06b2 | 2011-09-19 13:35:26 -0700 | [diff] [blame] | 253 | qemu_rbd_unescape(name); |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 254 | |
| 255 | if (!p) { |
| 256 | error_report("conf option %s has no value", name); |
| 257 | ret = -EINVAL; |
| 258 | break; |
| 259 | } |
| 260 | |
| 261 | ret = qemu_rbd_next_tok(value, sizeof(value), p, |
| 262 | ':', "conf option value", &p); |
| 263 | if (ret < 0) { |
| 264 | break; |
| 265 | } |
Sage Weil | 16a06b2 | 2011-09-19 13:35:26 -0700 | [diff] [blame] | 266 | qemu_rbd_unescape(value); |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 267 | |
Sage Weil | 7c7e9df | 2011-09-07 09:28:04 -0700 | [diff] [blame] | 268 | if (strcmp(name, "conf") == 0) { |
| 269 | ret = rados_conf_read_file(cluster, value); |
| 270 | if (ret < 0) { |
| 271 | error_report("error reading conf file %s", value); |
| 272 | break; |
| 273 | } |
| 274 | } else if (strcmp(name, "id") == 0) { |
| 275 | /* ignore, this is parsed by qemu_rbd_parse_clientname() */ |
| 276 | } else { |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 277 | ret = rados_conf_set(cluster, name, value); |
| 278 | if (ret < 0) { |
| 279 | error_report("invalid conf option %s", name); |
| 280 | ret = -EINVAL; |
| 281 | break; |
| 282 | } |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 283 | } |
| 284 | } |
| 285 | |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 286 | g_free(buf); |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 287 | return ret; |
| 288 | } |
| 289 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 290 | static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 291 | { |
| 292 | int64_t bytes = 0; |
| 293 | int64_t objsize; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 294 | int obj_order = 0; |
| 295 | char pool[RBD_MAX_POOL_NAME_SIZE]; |
| 296 | char name[RBD_MAX_IMAGE_NAME_SIZE]; |
| 297 | char snap_buf[RBD_MAX_SNAP_NAME_SIZE]; |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 298 | char conf[RBD_MAX_CONF_SIZE]; |
Sage Weil | 7c7e9df | 2011-09-07 09:28:04 -0700 | [diff] [blame] | 299 | char clientname_buf[RBD_MAX_CONF_SIZE]; |
| 300 | char *clientname; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 301 | rados_t cluster; |
| 302 | rados_ioctx_t io_ctx; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 303 | int ret; |
| 304 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 305 | if (qemu_rbd_parsename(filename, pool, sizeof(pool), |
| 306 | snap_buf, sizeof(snap_buf), |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 307 | name, sizeof(name), |
| 308 | conf, sizeof(conf)) < 0) { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 309 | return -EINVAL; |
| 310 | } |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 311 | |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 312 | /* Read out options */ |
| 313 | while (options && options->name) { |
| 314 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { |
| 315 | bytes = options->value.n; |
| 316 | } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { |
| 317 | if (options->value.n) { |
| 318 | objsize = options->value.n; |
| 319 | if ((objsize - 1) & objsize) { /* not a power of 2? */ |
| 320 | error_report("obj size needs to be power of 2"); |
| 321 | return -EINVAL; |
| 322 | } |
| 323 | if (objsize < 4096) { |
| 324 | error_report("obj size too small"); |
| 325 | return -EINVAL; |
| 326 | } |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 327 | obj_order = ffs(objsize) - 1; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 328 | } |
| 329 | } |
| 330 | options++; |
| 331 | } |
| 332 | |
Sage Weil | 7c7e9df | 2011-09-07 09:28:04 -0700 | [diff] [blame] | 333 | clientname = qemu_rbd_parse_clientname(conf, clientname_buf); |
| 334 | if (rados_create(&cluster, clientname) < 0) { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 335 | error_report("error initializing"); |
| 336 | return -EIO; |
| 337 | } |
| 338 | |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 339 | if (strstr(conf, "conf=") == NULL) { |
Sage Weil | f9fe18e | 2011-09-15 14:11:08 -0700 | [diff] [blame] | 340 | /* try default location, but ignore failure */ |
| 341 | rados_conf_read_file(cluster, NULL); |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 342 | } |
| 343 | |
| 344 | if (conf[0] != '\0' && |
| 345 | qemu_rbd_set_conf(cluster, conf) < 0) { |
| 346 | error_report("error setting config options"); |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 347 | rados_shutdown(cluster); |
| 348 | return -EIO; |
| 349 | } |
| 350 | |
| 351 | if (rados_connect(cluster) < 0) { |
| 352 | error_report("error connecting"); |
| 353 | rados_shutdown(cluster); |
| 354 | return -EIO; |
| 355 | } |
| 356 | |
| 357 | if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 358 | error_report("error opening pool %s", pool); |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 359 | rados_shutdown(cluster); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 360 | return -EIO; |
| 361 | } |
| 362 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 363 | ret = rbd_create(io_ctx, name, bytes, &obj_order); |
| 364 | rados_ioctx_destroy(io_ctx); |
| 365 | rados_shutdown(cluster); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 366 | |
| 367 | return ret; |
| 368 | } |
| 369 | |
| 370 | /* |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 371 | * This aio completion is being called from qemu_rbd_aio_event_reader() |
| 372 | * and runs in qemu context. It schedules a bh, but just in case the aio |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 373 | * was not cancelled before. |
| 374 | */ |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 375 | static void qemu_rbd_complete_aio(RADOSCB *rcb) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 376 | { |
| 377 | RBDAIOCB *acb = rcb->acb; |
| 378 | int64_t r; |
| 379 | |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 380 | r = rcb->ret; |
| 381 | |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 382 | if (acb->cmd == RBD_AIO_WRITE || |
| 383 | acb->cmd == RBD_AIO_DISCARD) { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 384 | if (r < 0) { |
| 385 | acb->ret = r; |
| 386 | acb->error = 1; |
| 387 | } else if (!acb->error) { |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 388 | acb->ret = rcb->size; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 389 | } |
| 390 | } else { |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 391 | if (r < 0) { |
| 392 | memset(rcb->buf, 0, rcb->size); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 393 | acb->ret = r; |
| 394 | acb->error = 1; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 395 | } else if (r < rcb->size) { |
| 396 | memset(rcb->buf + r, 0, rcb->size - r); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 397 | if (!acb->error) { |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 398 | acb->ret = rcb->size; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 399 | } |
| 400 | } else if (!acb->error) { |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 401 | acb->ret = r; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 402 | } |
| 403 | } |
| 404 | /* Note that acb->bh can be NULL in case where the aio was cancelled */ |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 405 | acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb); |
| 406 | qemu_bh_schedule(acb->bh); |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 407 | g_free(rcb); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 408 | } |
| 409 | |
| 410 | /* |
| 411 | * aio fd read handler. It runs in the qemu context and calls the |
| 412 | * completion handling of completed rados aio operations. |
| 413 | */ |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 414 | static void qemu_rbd_aio_event_reader(void *opaque) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 415 | { |
| 416 | BDRVRBDState *s = opaque; |
| 417 | |
| 418 | ssize_t ret; |
| 419 | |
| 420 | do { |
| 421 | char *p = (char *)&s->event_rcb; |
| 422 | |
| 423 | /* now read the rcb pointer that was sent from a non qemu thread */ |
Sage Weil | dfe80b0 | 2011-09-07 09:28:05 -0700 | [diff] [blame] | 424 | ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos, |
| 425 | sizeof(s->event_rcb) - s->event_reader_pos); |
| 426 | if (ret > 0) { |
| 427 | s->event_reader_pos += ret; |
| 428 | if (s->event_reader_pos == sizeof(s->event_rcb)) { |
| 429 | s->event_reader_pos = 0; |
| 430 | qemu_rbd_complete_aio(s->event_rcb); |
| 431 | s->qemu_aio_count--; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 432 | } |
| 433 | } |
| 434 | } while (ret < 0 && errno == EINTR); |
| 435 | } |
| 436 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 437 | static int qemu_rbd_aio_flush_cb(void *opaque) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 438 | { |
| 439 | BDRVRBDState *s = opaque; |
| 440 | |
| 441 | return (s->qemu_aio_count > 0); |
| 442 | } |
| 443 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 444 | static int qemu_rbd_open(BlockDriverState *bs, const char *filename, int flags) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 445 | { |
| 446 | BDRVRBDState *s = bs->opaque; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 447 | char pool[RBD_MAX_POOL_NAME_SIZE]; |
| 448 | char snap_buf[RBD_MAX_SNAP_NAME_SIZE]; |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 449 | char conf[RBD_MAX_CONF_SIZE]; |
Sage Weil | 7c7e9df | 2011-09-07 09:28:04 -0700 | [diff] [blame] | 450 | char clientname_buf[RBD_MAX_CONF_SIZE]; |
| 451 | char *clientname; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 452 | int r; |
| 453 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 454 | if (qemu_rbd_parsename(filename, pool, sizeof(pool), |
| 455 | snap_buf, sizeof(snap_buf), |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 456 | s->name, sizeof(s->name), |
| 457 | conf, sizeof(conf)) < 0) { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 458 | return -EINVAL; |
| 459 | } |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 460 | |
Sage Weil | 7c7e9df | 2011-09-07 09:28:04 -0700 | [diff] [blame] | 461 | clientname = qemu_rbd_parse_clientname(conf, clientname_buf); |
| 462 | r = rados_create(&s->cluster, clientname); |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 463 | if (r < 0) { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 464 | error_report("error initializing"); |
| 465 | return r; |
| 466 | } |
| 467 | |
Sage Weil | eb93d5d | 2011-09-07 09:28:06 -0700 | [diff] [blame] | 468 | s->snap = NULL; |
| 469 | if (snap_buf[0] != '\0') { |
| 470 | s->snap = g_strdup(snap_buf); |
| 471 | } |
| 472 | |
Josh Durgin | b11f38f | 2012-05-17 13:42:29 -0700 | [diff] [blame] | 473 | /* |
| 474 | * Fallback to more conservative semantics if setting cache |
| 475 | * options fails. Ignore errors from setting rbd_cache because the |
| 476 | * only possible error is that the option does not exist, and |
| 477 | * librbd defaults to no caching. If write through caching cannot |
| 478 | * be set up, fall back to no caching. |
| 479 | */ |
| 480 | if (flags & BDRV_O_NOCACHE) { |
| 481 | rados_conf_set(s->cluster, "rbd_cache", "false"); |
| 482 | } else { |
| 483 | rados_conf_set(s->cluster, "rbd_cache", "true"); |
Josh Durgin | b11f38f | 2012-05-17 13:42:29 -0700 | [diff] [blame] | 484 | } |
| 485 | |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 486 | if (strstr(conf, "conf=") == NULL) { |
Sage Weil | f9fe18e | 2011-09-15 14:11:08 -0700 | [diff] [blame] | 487 | /* try default location, but ignore failure */ |
| 488 | rados_conf_read_file(s->cluster, NULL); |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 489 | } |
| 490 | |
| 491 | if (conf[0] != '\0') { |
| 492 | r = qemu_rbd_set_conf(s->cluster, conf); |
| 493 | if (r < 0) { |
| 494 | error_report("error setting config options"); |
Sage Weil | eb93d5d | 2011-09-07 09:28:06 -0700 | [diff] [blame] | 495 | goto failed_shutdown; |
Josh Durgin | fab5cf5 | 2011-05-26 16:07:32 -0700 | [diff] [blame] | 496 | } |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 497 | } |
| 498 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 499 | r = rados_connect(s->cluster); |
| 500 | if (r < 0) { |
| 501 | error_report("error connecting"); |
Sage Weil | eb93d5d | 2011-09-07 09:28:06 -0700 | [diff] [blame] | 502 | goto failed_shutdown; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 503 | } |
| 504 | |
| 505 | r = rados_ioctx_create(s->cluster, pool, &s->io_ctx); |
| 506 | if (r < 0) { |
| 507 | error_report("error opening pool %s", pool); |
Sage Weil | eb93d5d | 2011-09-07 09:28:06 -0700 | [diff] [blame] | 508 | goto failed_shutdown; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 509 | } |
| 510 | |
| 511 | r = rbd_open(s->io_ctx, s->name, &s->image, s->snap); |
| 512 | if (r < 0) { |
| 513 | error_report("error reading header from %s", s->name); |
Sage Weil | eb93d5d | 2011-09-07 09:28:06 -0700 | [diff] [blame] | 514 | goto failed_open; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 515 | } |
| 516 | |
| 517 | bs->read_only = (s->snap != NULL); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 518 | |
| 519 | s->event_reader_pos = 0; |
| 520 | r = qemu_pipe(s->fds); |
| 521 | if (r < 0) { |
| 522 | error_report("error opening eventfd"); |
| 523 | goto failed; |
| 524 | } |
| 525 | fcntl(s->fds[0], F_SETFL, O_NONBLOCK); |
| 526 | fcntl(s->fds[1], F_SETFL, O_NONBLOCK); |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 527 | qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader, |
Paolo Bonzini | bafbd6a | 2012-04-12 14:00:54 +0200 | [diff] [blame] | 528 | NULL, qemu_rbd_aio_flush_cb, s); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 529 | |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 530 | |
| 531 | return 0; |
| 532 | |
| 533 | failed: |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 534 | rbd_close(s->image); |
Sage Weil | eb93d5d | 2011-09-07 09:28:06 -0700 | [diff] [blame] | 535 | failed_open: |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 536 | rados_ioctx_destroy(s->io_ctx); |
Sage Weil | eb93d5d | 2011-09-07 09:28:06 -0700 | [diff] [blame] | 537 | failed_shutdown: |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 538 | rados_shutdown(s->cluster); |
Sage Weil | eb93d5d | 2011-09-07 09:28:06 -0700 | [diff] [blame] | 539 | g_free(s->snap); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 540 | return r; |
| 541 | } |
| 542 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 543 | static void qemu_rbd_close(BlockDriverState *bs) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 544 | { |
| 545 | BDRVRBDState *s = bs->opaque; |
| 546 | |
| 547 | close(s->fds[0]); |
| 548 | close(s->fds[1]); |
Paolo Bonzini | bafbd6a | 2012-04-12 14:00:54 +0200 | [diff] [blame] | 549 | qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL, NULL, NULL, NULL); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 550 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 551 | rbd_close(s->image); |
| 552 | rados_ioctx_destroy(s->io_ctx); |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 553 | g_free(s->snap); |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 554 | rados_shutdown(s->cluster); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 555 | } |
| 556 | |
| 557 | /* |
| 558 | * Cancel aio. Since we don't reference acb in a non qemu threads, |
| 559 | * it is safe to access it here. |
| 560 | */ |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 561 | static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 562 | { |
| 563 | RBDAIOCB *acb = (RBDAIOCB *) blockacb; |
| 564 | acb->cancelled = 1; |
Stefan Priebe | 473c7f0 | 2012-11-30 09:55:46 +0100 | [diff] [blame] | 565 | |
| 566 | while (acb->status == -EINPROGRESS) { |
| 567 | qemu_aio_wait(); |
| 568 | } |
| 569 | |
| 570 | qemu_aio_release(acb); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 571 | } |
| 572 | |
Stefan Hajnoczi | d7331be | 2012-10-31 16:34:37 +0100 | [diff] [blame] | 573 | static const AIOCBInfo rbd_aiocb_info = { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 574 | .aiocb_size = sizeof(RBDAIOCB), |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 575 | .cancel = qemu_rbd_aio_cancel, |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 576 | }; |
| 577 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 578 | static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 579 | { |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 580 | int ret = 0; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 581 | while (1) { |
| 582 | fd_set wfd; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 583 | int fd = s->fds[RBD_FD_WRITE]; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 584 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 585 | /* send the op pointer to the qemu thread that is responsible |
| 586 | for the aio/op completion. Must do it in a qemu thread context */ |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 587 | ret = write(fd, (void *)&rcb, sizeof(rcb)); |
| 588 | if (ret >= 0) { |
| 589 | break; |
| 590 | } |
| 591 | if (errno == EINTR) { |
| 592 | continue; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 593 | } |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 594 | if (errno != EAGAIN) { |
| 595 | break; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 596 | } |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 597 | |
| 598 | FD_ZERO(&wfd); |
| 599 | FD_SET(fd, &wfd); |
| 600 | do { |
| 601 | ret = select(fd + 1, NULL, &wfd, NULL, NULL); |
| 602 | } while (ret < 0 && errno == EINTR); |
| 603 | } |
| 604 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 605 | return ret; |
| 606 | } |
| 607 | |
| 608 | /* |
| 609 | * This is the callback function for rbd_aio_read and _write |
| 610 | * |
| 611 | * Note: this function is being called from a non qemu thread so |
| 612 | * we need to be careful about what we do here. Generally we only |
| 613 | * write to the block notification pipe, and do the rest of the |
| 614 | * io completion handling from qemu_rbd_aio_event_reader() which |
| 615 | * runs in a qemu context. |
| 616 | */ |
| 617 | static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb) |
| 618 | { |
| 619 | int ret; |
| 620 | rcb->ret = rbd_aio_get_return_value(c); |
| 621 | rbd_aio_release(c); |
| 622 | ret = qemu_rbd_send_pipe(rcb->s, rcb); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 623 | if (ret < 0) { |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 624 | error_report("failed writing to acb->s->fds"); |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 625 | g_free(rcb); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 626 | } |
| 627 | } |
| 628 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 629 | /* Callback when all queued rbd_aio requests are complete */ |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 630 | |
| 631 | static void rbd_aio_bh_cb(void *opaque) |
| 632 | { |
| 633 | RBDAIOCB *acb = opaque; |
| 634 | |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 635 | if (acb->cmd == RBD_AIO_READ) { |
Michael Tokarev | 0339614 | 2012-06-07 20:17:55 +0400 | [diff] [blame] | 636 | qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 637 | } |
| 638 | qemu_vfree(acb->bounce); |
| 639 | acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret)); |
| 640 | qemu_bh_delete(acb->bh); |
| 641 | acb->bh = NULL; |
Stefan Priebe | 473c7f0 | 2012-11-30 09:55:46 +0100 | [diff] [blame] | 642 | acb->status = 0; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 643 | |
Stefan Priebe | 473c7f0 | 2012-11-30 09:55:46 +0100 | [diff] [blame] | 644 | if (!acb->cancelled) { |
| 645 | qemu_aio_release(acb); |
| 646 | } |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 647 | } |
| 648 | |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 649 | static int rbd_aio_discard_wrapper(rbd_image_t image, |
| 650 | uint64_t off, |
| 651 | uint64_t len, |
| 652 | rbd_completion_t comp) |
| 653 | { |
| 654 | #ifdef LIBRBD_SUPPORTS_DISCARD |
| 655 | return rbd_aio_discard(image, off, len, comp); |
| 656 | #else |
| 657 | return -ENOTSUP; |
| 658 | #endif |
| 659 | } |
| 660 | |
| 661 | static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs, |
| 662 | int64_t sector_num, |
| 663 | QEMUIOVector *qiov, |
| 664 | int nb_sectors, |
| 665 | BlockDriverCompletionFunc *cb, |
| 666 | void *opaque, |
| 667 | RBDAIOCmd cmd) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 668 | { |
| 669 | RBDAIOCB *acb; |
| 670 | RADOSCB *rcb; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 671 | rbd_completion_t c; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 672 | int64_t off, size; |
| 673 | char *buf; |
Josh Durgin | 51a1352 | 2011-05-26 16:07:33 -0700 | [diff] [blame] | 674 | int r; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 675 | |
| 676 | BDRVRBDState *s = bs->opaque; |
| 677 | |
Stefan Hajnoczi | d7331be | 2012-10-31 16:34:37 +0100 | [diff] [blame] | 678 | acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque); |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 679 | acb->cmd = cmd; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 680 | acb->qiov = qiov; |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 681 | if (cmd == RBD_AIO_DISCARD) { |
| 682 | acb->bounce = NULL; |
| 683 | } else { |
| 684 | acb->bounce = qemu_blockalign(bs, qiov->size); |
| 685 | } |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 686 | acb->ret = 0; |
| 687 | acb->error = 0; |
| 688 | acb->s = s; |
| 689 | acb->cancelled = 0; |
| 690 | acb->bh = NULL; |
Stefan Priebe | 473c7f0 | 2012-11-30 09:55:46 +0100 | [diff] [blame] | 691 | acb->status = -EINPROGRESS; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 692 | |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 693 | if (cmd == RBD_AIO_WRITE) { |
Michael Tokarev | d5e6b16 | 2012-06-07 20:21:06 +0400 | [diff] [blame] | 694 | qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 695 | } |
| 696 | |
| 697 | buf = acb->bounce; |
| 698 | |
| 699 | off = sector_num * BDRV_SECTOR_SIZE; |
| 700 | size = nb_sectors * BDRV_SECTOR_SIZE; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 701 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 702 | s->qemu_aio_count++; /* All the RADOSCB */ |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 703 | |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 704 | rcb = g_malloc(sizeof(RADOSCB)); |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 705 | rcb->done = 0; |
| 706 | rcb->acb = acb; |
| 707 | rcb->buf = buf; |
| 708 | rcb->s = acb->s; |
| 709 | rcb->size = size; |
Josh Durgin | 51a1352 | 2011-05-26 16:07:33 -0700 | [diff] [blame] | 710 | r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c); |
| 711 | if (r < 0) { |
| 712 | goto failed; |
| 713 | } |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 714 | |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 715 | switch (cmd) { |
| 716 | case RBD_AIO_WRITE: |
Josh Durgin | 51a1352 | 2011-05-26 16:07:33 -0700 | [diff] [blame] | 717 | r = rbd_aio_write(s->image, off, size, buf, c); |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 718 | break; |
| 719 | case RBD_AIO_READ: |
Josh Durgin | 51a1352 | 2011-05-26 16:07:33 -0700 | [diff] [blame] | 720 | r = rbd_aio_read(s->image, off, size, buf, c); |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 721 | break; |
| 722 | case RBD_AIO_DISCARD: |
| 723 | r = rbd_aio_discard_wrapper(s->image, off, size, c); |
| 724 | break; |
| 725 | default: |
| 726 | r = -EINVAL; |
Josh Durgin | 51a1352 | 2011-05-26 16:07:33 -0700 | [diff] [blame] | 727 | } |
| 728 | |
| 729 | if (r < 0) { |
| 730 | goto failed; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 731 | } |
| 732 | |
| 733 | return &acb->common; |
Josh Durgin | 51a1352 | 2011-05-26 16:07:33 -0700 | [diff] [blame] | 734 | |
| 735 | failed: |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 736 | g_free(rcb); |
Josh Durgin | 51a1352 | 2011-05-26 16:07:33 -0700 | [diff] [blame] | 737 | s->qemu_aio_count--; |
| 738 | qemu_aio_release(acb); |
| 739 | return NULL; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 740 | } |
| 741 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 742 | static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs, |
| 743 | int64_t sector_num, |
| 744 | QEMUIOVector *qiov, |
| 745 | int nb_sectors, |
| 746 | BlockDriverCompletionFunc *cb, |
| 747 | void *opaque) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 748 | { |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 749 | return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque, |
| 750 | RBD_AIO_READ); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 751 | } |
| 752 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 753 | static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs, |
| 754 | int64_t sector_num, |
| 755 | QEMUIOVector *qiov, |
| 756 | int nb_sectors, |
| 757 | BlockDriverCompletionFunc *cb, |
| 758 | void *opaque) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 759 | { |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 760 | return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque, |
| 761 | RBD_AIO_WRITE); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 762 | } |
| 763 | |
Paolo Bonzini | 8b94ff8 | 2011-10-20 13:16:24 +0200 | [diff] [blame] | 764 | static int qemu_rbd_co_flush(BlockDriverState *bs) |
Sage Weil | 7a3f5fe | 2011-09-15 14:11:11 -0700 | [diff] [blame] | 765 | { |
| 766 | #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1) |
| 767 | /* rbd_flush added in 0.1.1 */ |
| 768 | BDRVRBDState *s = bs->opaque; |
| 769 | return rbd_flush(s->image); |
| 770 | #else |
| 771 | return 0; |
| 772 | #endif |
| 773 | } |
| 774 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 775 | static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 776 | { |
| 777 | BDRVRBDState *s = bs->opaque; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 778 | rbd_image_info_t info; |
| 779 | int r; |
| 780 | |
| 781 | r = rbd_stat(s->image, &info, sizeof(info)); |
| 782 | if (r < 0) { |
| 783 | return r; |
| 784 | } |
| 785 | |
| 786 | bdi->cluster_size = info.obj_size; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 787 | return 0; |
| 788 | } |
| 789 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 790 | static int64_t qemu_rbd_getlength(BlockDriverState *bs) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 791 | { |
| 792 | BDRVRBDState *s = bs->opaque; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 793 | rbd_image_info_t info; |
| 794 | int r; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 795 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 796 | r = rbd_stat(s->image, &info, sizeof(info)); |
| 797 | if (r < 0) { |
| 798 | return r; |
| 799 | } |
| 800 | |
| 801 | return info.size; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 802 | } |
| 803 | |
Josh Durgin | 30cdc48 | 2011-05-26 16:07:34 -0700 | [diff] [blame] | 804 | static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset) |
| 805 | { |
| 806 | BDRVRBDState *s = bs->opaque; |
| 807 | int r; |
| 808 | |
| 809 | r = rbd_resize(s->image, offset); |
| 810 | if (r < 0) { |
| 811 | return r; |
| 812 | } |
| 813 | |
| 814 | return 0; |
| 815 | } |
| 816 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 817 | static int qemu_rbd_snap_create(BlockDriverState *bs, |
| 818 | QEMUSnapshotInfo *sn_info) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 819 | { |
| 820 | BDRVRBDState *s = bs->opaque; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 821 | int r; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 822 | |
| 823 | if (sn_info->name[0] == '\0') { |
| 824 | return -EINVAL; /* we need a name for rbd snapshots */ |
| 825 | } |
| 826 | |
| 827 | /* |
| 828 | * rbd snapshots are using the name as the user controlled unique identifier |
| 829 | * we can't use the rbd snapid for that purpose, as it can't be set |
| 830 | */ |
| 831 | if (sn_info->id_str[0] != '\0' && |
| 832 | strcmp(sn_info->id_str, sn_info->name) != 0) { |
| 833 | return -EINVAL; |
| 834 | } |
| 835 | |
| 836 | if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) { |
| 837 | return -ERANGE; |
| 838 | } |
| 839 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 840 | r = rbd_snap_create(s->image, sn_info->name); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 841 | if (r < 0) { |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 842 | error_report("failed to create snap: %s", strerror(-r)); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 843 | return r; |
| 844 | } |
| 845 | |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 846 | return 0; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 847 | } |
| 848 | |
Gregory Farnum | bd60324 | 2012-01-11 11:53:52 -0800 | [diff] [blame] | 849 | static int qemu_rbd_snap_remove(BlockDriverState *bs, |
| 850 | const char *snapshot_name) |
| 851 | { |
| 852 | BDRVRBDState *s = bs->opaque; |
| 853 | int r; |
| 854 | |
| 855 | r = rbd_snap_remove(s->image, snapshot_name); |
| 856 | return r; |
| 857 | } |
| 858 | |
| 859 | static int qemu_rbd_snap_rollback(BlockDriverState *bs, |
| 860 | const char *snapshot_name) |
| 861 | { |
| 862 | BDRVRBDState *s = bs->opaque; |
| 863 | int r; |
| 864 | |
| 865 | r = rbd_snap_rollback(s->image, snapshot_name); |
| 866 | return r; |
| 867 | } |
| 868 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 869 | static int qemu_rbd_snap_list(BlockDriverState *bs, |
| 870 | QEMUSnapshotInfo **psn_tab) |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 871 | { |
| 872 | BDRVRBDState *s = bs->opaque; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 873 | QEMUSnapshotInfo *sn_info, *sn_tab = NULL; |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 874 | int i, snap_count; |
| 875 | rbd_snap_info_t *snaps; |
| 876 | int max_snaps = RBD_MAX_SNAPS; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 877 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 878 | do { |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 879 | snaps = g_malloc(sizeof(*snaps) * max_snaps); |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 880 | snap_count = rbd_snap_list(s->image, snaps, &max_snaps); |
| 881 | if (snap_count < 0) { |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 882 | g_free(snaps); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 883 | } |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 884 | } while (snap_count == -ERANGE); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 885 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 886 | if (snap_count <= 0) { |
Josh Durgin | b9c5329 | 2011-12-06 17:05:10 -0800 | [diff] [blame] | 887 | goto done; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 888 | } |
| 889 | |
Anthony Liguori | 7267c09 | 2011-08-20 22:09:37 -0500 | [diff] [blame] | 890 | sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo)); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 891 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 892 | for (i = 0; i < snap_count; i++) { |
| 893 | const char *snap_name = snaps[i].name; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 894 | |
| 895 | sn_info = sn_tab + i; |
| 896 | pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name); |
| 897 | pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name); |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 898 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 899 | sn_info->vm_state_size = snaps[i].size; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 900 | sn_info->date_sec = 0; |
| 901 | sn_info->date_nsec = 0; |
| 902 | sn_info->vm_clock_nsec = 0; |
| 903 | } |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 904 | rbd_snap_list_end(snaps); |
| 905 | |
Josh Durgin | b9c5329 | 2011-12-06 17:05:10 -0800 | [diff] [blame] | 906 | done: |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 907 | *psn_tab = sn_tab; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 908 | return snap_count; |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 909 | } |
| 910 | |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 911 | #ifdef LIBRBD_SUPPORTS_DISCARD |
| 912 | static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs, |
| 913 | int64_t sector_num, |
| 914 | int nb_sectors, |
| 915 | BlockDriverCompletionFunc *cb, |
| 916 | void *opaque) |
| 917 | { |
| 918 | return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque, |
| 919 | RBD_AIO_DISCARD); |
| 920 | } |
| 921 | #endif |
| 922 | |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 923 | static QEMUOptionParameter qemu_rbd_create_options[] = { |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 924 | { |
| 925 | .name = BLOCK_OPT_SIZE, |
| 926 | .type = OPT_SIZE, |
| 927 | .help = "Virtual disk size" |
| 928 | }, |
| 929 | { |
| 930 | .name = BLOCK_OPT_CLUSTER_SIZE, |
| 931 | .type = OPT_SIZE, |
| 932 | .help = "RBD object size" |
| 933 | }, |
| 934 | {NULL} |
| 935 | }; |
| 936 | |
| 937 | static BlockDriver bdrv_rbd = { |
| 938 | .format_name = "rbd", |
| 939 | .instance_size = sizeof(BDRVRBDState), |
Josh Durgin | ad32e9c | 2011-05-26 16:07:31 -0700 | [diff] [blame] | 940 | .bdrv_file_open = qemu_rbd_open, |
| 941 | .bdrv_close = qemu_rbd_close, |
| 942 | .bdrv_create = qemu_rbd_create, |
| 943 | .bdrv_get_info = qemu_rbd_getinfo, |
| 944 | .create_options = qemu_rbd_create_options, |
| 945 | .bdrv_getlength = qemu_rbd_getlength, |
Josh Durgin | 30cdc48 | 2011-05-26 16:07:34 -0700 | [diff] [blame] | 946 | .bdrv_truncate = qemu_rbd_truncate, |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 947 | .protocol_name = "rbd", |
| 948 | |
Kevin Wolf | c68b89a | 2011-11-10 17:25:44 +0100 | [diff] [blame] | 949 | .bdrv_aio_readv = qemu_rbd_aio_readv, |
| 950 | .bdrv_aio_writev = qemu_rbd_aio_writev, |
| 951 | .bdrv_co_flush_to_disk = qemu_rbd_co_flush, |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 952 | |
Josh Durgin | 787f313 | 2012-04-30 23:16:45 -0700 | [diff] [blame] | 953 | #ifdef LIBRBD_SUPPORTS_DISCARD |
| 954 | .bdrv_aio_discard = qemu_rbd_aio_discard, |
| 955 | #endif |
| 956 | |
Kevin Wolf | c68b89a | 2011-11-10 17:25:44 +0100 | [diff] [blame] | 957 | .bdrv_snapshot_create = qemu_rbd_snap_create, |
Gregory Farnum | bd60324 | 2012-01-11 11:53:52 -0800 | [diff] [blame] | 958 | .bdrv_snapshot_delete = qemu_rbd_snap_remove, |
Kevin Wolf | c68b89a | 2011-11-10 17:25:44 +0100 | [diff] [blame] | 959 | .bdrv_snapshot_list = qemu_rbd_snap_list, |
Gregory Farnum | bd60324 | 2012-01-11 11:53:52 -0800 | [diff] [blame] | 960 | .bdrv_snapshot_goto = qemu_rbd_snap_rollback, |
Christian Brunner | f27aaf4 | 2010-12-06 20:53:01 +0100 | [diff] [blame] | 961 | }; |
| 962 | |
| 963 | static void bdrv_rbd_init(void) |
| 964 | { |
| 965 | bdrv_register(&bdrv_rbd); |
| 966 | } |
| 967 | |
| 968 | block_init(bdrv_rbd_init); |