include/block/nbd.h

   1 /*
   2  *  Copyright Red Hat
   3  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
   4  *
   5  *  Network Block Device
   6  *
   7  *  This program is free software; you can redistribute it and/or modify
   8  *  it under the terms of the GNU General Public License as published by
   9  *  the Free Software Foundation; under version 2 of the License.
  10  *
  11  *  This program is distributed in the hope that it will be useful,
  12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  *  GNU General Public License for more details.
  15  *
  16  *  You should have received a copy of the GNU General Public License
  17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
  18  */
  19
  20 #ifndef NBD_H
  21 #define NBD_H
  22
  23 #include "block/export.h"
  24 #include "io/channel-socket.h"
  25 #include "crypto/tlscreds.h"
  26 #include "qapi/error.h"
  27 #include "qemu/bswap.h"
  28
  29 typedef struct NBDExport NBDExport;
  30 typedef struct NBDClient NBDClient;
  31 typedef struct NBDClientConnection NBDClientConnection;
  32
  33 extern const BlockExportDriver blk_exp_nbd;
  34
  35 /* Handshake phase structs - this struct is passed on the wire */
  36
  37 typedef struct NBDOption {
  38     uint64_t magic; /* NBD_OPTS_MAGIC */
  39     uint32_t option; /* NBD_OPT_* */
  40     uint32_t length;
  41 } QEMU_PACKED NBDOption;
  42
  43 typedef struct NBDOptionReply {
  44     uint64_t magic; /* NBD_REP_MAGIC */
  45     uint32_t option; /* NBD_OPT_* */
  46     uint32_t type; /* NBD_REP_* */
  47     uint32_t length;
  48 } QEMU_PACKED NBDOptionReply;
  49
  50 typedef struct NBDOptionReplyMetaContext {
  51     NBDOptionReply h; /* h.type = NBD_REP_META_CONTEXT, h.length > 4 */
  52     uint32_t context_id;
  53     /* metadata context name follows */
  54 } QEMU_PACKED NBDOptionReplyMetaContext;
  55
  56 /* Transmission phase structs
  57  *
  58  * Note: these are _NOT_ the same as the network representation of an NBD
  59  * request and reply!
  60  */
  61 typedef struct NBDRequest {
  62     uint64_t handle;
  63     uint64_t from;
  64     uint32_t len;
  65     uint16_t flags; /* NBD_CMD_FLAG_* */
  66     uint16_t type; /* NBD_CMD_* */
  67 } NBDRequest;
  68
  69 typedef struct NBDSimpleReply {
  70     uint32_t magic;  /* NBD_SIMPLE_REPLY_MAGIC */
  71     uint32_t error;
  72     uint64_t handle;
  73 } QEMU_PACKED NBDSimpleReply;
  74
  75 /* Header of all structured replies */
  76 typedef struct NBDStructuredReplyChunk {
  77     uint32_t magic;  /* NBD_STRUCTURED_REPLY_MAGIC */
  78     uint16_t flags;  /* combination of NBD_REPLY_FLAG_* */
  79     uint16_t type;   /* NBD_REPLY_TYPE_* */
  80     uint64_t handle; /* request handle */
  81     uint32_t length; /* length of payload */
  82 } QEMU_PACKED NBDStructuredReplyChunk;
  83
  84 typedef union NBDReply {
  85     NBDSimpleReply simple;
  86     NBDStructuredReplyChunk structured;
  87     struct {
  88         /* @magic and @handle fields have the same offset and size both in
  89          * simple reply and structured reply chunk, so let them be accessible
  90          * without ".simple." or ".structured." specification
  91          */
  92         uint32_t magic;
  93         uint32_t _skip;
  94         uint64_t handle;
  95     } QEMU_PACKED;
  96 } NBDReply;
  97
  98 /* Header of chunk for NBD_REPLY_TYPE_OFFSET_DATA */
  99 typedef struct NBDStructuredReadData {
 100     /* header's .length >= 9 */
 101     uint64_t offset;
 102     /* At least one byte of data payload follows, calculated from h.length */
 103 } QEMU_PACKED NBDStructuredReadData;
 104
 105 /* Complete chunk for NBD_REPLY_TYPE_OFFSET_HOLE */
 106 typedef struct NBDStructuredReadHole {
 107     /* header's length == 12 */
 108     uint64_t offset;
 109     uint32_t length;
 110 } QEMU_PACKED NBDStructuredReadHole;
 111
 112 /* Header of all NBD_REPLY_TYPE_ERROR* errors */
 113 typedef struct NBDStructuredError {
 114     /* header's length >= 6 */
 115     uint32_t error;
 116     uint16_t message_length;
 117 } QEMU_PACKED NBDStructuredError;
 118
 119 /* Header of NBD_REPLY_TYPE_BLOCK_STATUS */
 120 typedef struct NBDStructuredMeta {
 121     /* header's length >= 12 (at least one extent) */
 122     uint32_t context_id;
 123     /* extents follows */
 124 } QEMU_PACKED NBDStructuredMeta;
 125
 126 /* Extent chunk for NBD_REPLY_TYPE_BLOCK_STATUS */
 127 typedef struct NBDExtent {
 128     uint32_t length;
 129     uint32_t flags; /* NBD_STATE_* */
 130 } QEMU_PACKED NBDExtent;
 131
 132 /* Transmission (export) flags: sent from server to client during handshake,
 133    but describe what will happen during transmission */
 134 enum {
 135     NBD_FLAG_HAS_FLAGS_BIT          =  0, /* Flags are there */
 136     NBD_FLAG_READ_ONLY_BIT          =  1, /* Device is read-only */
 137     NBD_FLAG_SEND_FLUSH_BIT         =  2, /* Send FLUSH */
 138     NBD_FLAG_SEND_FUA_BIT           =  3, /* Send FUA (Force Unit Access) */
 139     NBD_FLAG_ROTATIONAL_BIT         =  4, /* Use elevator algorithm -
 140                                              rotational media */
 141     NBD_FLAG_SEND_TRIM_BIT          =  5, /* Send TRIM (discard) */
 142     NBD_FLAG_SEND_WRITE_ZEROES_BIT  =  6, /* Send WRITE_ZEROES */
 143     NBD_FLAG_SEND_DF_BIT            =  7, /* Send DF (Do not Fragment) */
 144     NBD_FLAG_CAN_MULTI_CONN_BIT     =  8, /* Multi-client cache consistent */
 145     NBD_FLAG_SEND_RESIZE_BIT        =  9, /* Send resize */
 146     NBD_FLAG_SEND_CACHE_BIT         = 10, /* Send CACHE (prefetch) */
 147     NBD_FLAG_SEND_FAST_ZERO_BIT     = 11, /* FAST_ZERO flag for WRITE_ZEROES */
 148 };
 149
 150 #define NBD_FLAG_HAS_FLAGS         (1 << NBD_FLAG_HAS_FLAGS_BIT)
 151 #define NBD_FLAG_READ_ONLY         (1 << NBD_FLAG_READ_ONLY_BIT)
 152 #define NBD_FLAG_SEND_FLUSH        (1 << NBD_FLAG_SEND_FLUSH_BIT)
 153 #define NBD_FLAG_SEND_FUA          (1 << NBD_FLAG_SEND_FUA_BIT)
 154 #define NBD_FLAG_ROTATIONAL        (1 << NBD_FLAG_ROTATIONAL_BIT)
 155 #define NBD_FLAG_SEND_TRIM         (1 << NBD_FLAG_SEND_TRIM_BIT)
 156 #define NBD_FLAG_SEND_WRITE_ZEROES (1 << NBD_FLAG_SEND_WRITE_ZEROES_BIT)
 157 #define NBD_FLAG_SEND_DF           (1 << NBD_FLAG_SEND_DF_BIT)
 158 #define NBD_FLAG_CAN_MULTI_CONN    (1 << NBD_FLAG_CAN_MULTI_CONN_BIT)
 159 #define NBD_FLAG_SEND_RESIZE       (1 << NBD_FLAG_SEND_RESIZE_BIT)
 160 #define NBD_FLAG_SEND_CACHE        (1 << NBD_FLAG_SEND_CACHE_BIT)
 161 #define NBD_FLAG_SEND_FAST_ZERO    (1 << NBD_FLAG_SEND_FAST_ZERO_BIT)
 162
 163 /* New-style handshake (global) flags, sent from server to client, and
 164    control what will happen during handshake phase. */
 165 #define NBD_FLAG_FIXED_NEWSTYLE   (1 << 0) /* Fixed newstyle protocol. */
 166 #define NBD_FLAG_NO_ZEROES        (1 << 1) /* End handshake without zeroes. */
 167
 168 /* New-style client flags, sent from client to server to control what happens
 169    during handshake phase. */
 170 #define NBD_FLAG_C_FIXED_NEWSTYLE (1 << 0) /* Fixed newstyle protocol. */
 171 #define NBD_FLAG_C_NO_ZEROES      (1 << 1) /* End handshake without zeroes. */
 172
 173 /* Option requests. */
 174 #define NBD_OPT_EXPORT_NAME       (1)
 175 #define NBD_OPT_ABORT             (2)
 176 #define NBD_OPT_LIST              (3)
 177 /* #define NBD_OPT_PEEK_EXPORT    (4) not in use */
 178 #define NBD_OPT_STARTTLS          (5)
 179 #define NBD_OPT_INFO              (6)
 180 #define NBD_OPT_GO                (7)
 181 #define NBD_OPT_STRUCTURED_REPLY  (8)
 182 #define NBD_OPT_LIST_META_CONTEXT (9)
 183 #define NBD_OPT_SET_META_CONTEXT  (10)
 184
 185 /* Option reply types. */
 186 #define NBD_REP_ERR(value) ((UINT32_C(1) << 31) | (value))
 187
 188 #define NBD_REP_ACK             (1)    /* Data sending finished. */
 189 #define NBD_REP_SERVER          (2)    /* Export description. */
 190 #define NBD_REP_INFO            (3)    /* NBD_OPT_INFO/GO. */
 191 #define NBD_REP_META_CONTEXT    (4)    /* NBD_OPT_{LIST,SET}_META_CONTEXT */
 192
 193 #define NBD_REP_ERR_UNSUP           NBD_REP_ERR(1)  /* Unknown option */
 194 #define NBD_REP_ERR_POLICY          NBD_REP_ERR(2)  /* Server denied */
 195 #define NBD_REP_ERR_INVALID         NBD_REP_ERR(3)  /* Invalid length */
 196 #define NBD_REP_ERR_PLATFORM        NBD_REP_ERR(4)  /* Not compiled in */
 197 #define NBD_REP_ERR_TLS_REQD        NBD_REP_ERR(5)  /* TLS required */
 198 #define NBD_REP_ERR_UNKNOWN         NBD_REP_ERR(6)  /* Export unknown */
 199 #define NBD_REP_ERR_SHUTDOWN        NBD_REP_ERR(7)  /* Server shutting down */
 200 #define NBD_REP_ERR_BLOCK_SIZE_REQD NBD_REP_ERR(8)  /* Need INFO_BLOCK_SIZE */
 201
 202 /* Info types, used during NBD_REP_INFO */
 203 #define NBD_INFO_EXPORT         0
 204 #define NBD_INFO_NAME           1
 205 #define NBD_INFO_DESCRIPTION    2
 206 #define NBD_INFO_BLOCK_SIZE     3
 207
 208 /* Request flags, sent from client to server during transmission phase */
 209 #define NBD_CMD_FLAG_FUA        (1 << 0) /* 'force unit access' during write */
 210 #define NBD_CMD_FLAG_NO_HOLE    (1 << 1) /* don't punch hole on zero run */
 211 #define NBD_CMD_FLAG_DF         (1 << 2) /* don't fragment structured read */
 212 #define NBD_CMD_FLAG_REQ_ONE    (1 << 3) /* only one extent in BLOCK_STATUS
 213                                           * reply chunk */
 214 #define NBD_CMD_FLAG_FAST_ZERO  (1 << 4) /* fail if WRITE_ZEROES is not fast */
 215
 216 /* Supported request types */
 217 enum {
 218     NBD_CMD_READ = 0,
 219     NBD_CMD_WRITE = 1,
 220     NBD_CMD_DISC = 2,
 221     NBD_CMD_FLUSH = 3,
 222     NBD_CMD_TRIM = 4,
 223     NBD_CMD_CACHE = 5,
 224     NBD_CMD_WRITE_ZEROES = 6,
 225     NBD_CMD_BLOCK_STATUS = 7,
 226 };
 227
 228 #define NBD_DEFAULT_PORT        10809
 229
 230 /* Maximum size of a single READ/WRITE data buffer */
 231 #define NBD_MAX_BUFFER_SIZE (32 * 1024 * 1024)
 232
 233 /*
 234  * Maximum size of a protocol string (export name, metadata context name,
 235  * etc.).  Use malloc rather than stack allocation for storage of a
 236  * string.
 237  */
 238 #define NBD_MAX_STRING_SIZE 4096
 239
 240 /* Two types of reply structures */
 241 #define NBD_SIMPLE_REPLY_MAGIC      0x67446698
 242 #define NBD_STRUCTURED_REPLY_MAGIC  0x668e33ef
 243
 244 /* Structured reply flags */
 245 #define NBD_REPLY_FLAG_DONE          (1 << 0) /* This reply-chunk is last */
 246
 247 /* Structured reply types */
 248 #define NBD_REPLY_ERR(value)         ((1 << 15) | (value))
 249
 250 #define NBD_REPLY_TYPE_NONE          0
 251 #define NBD_REPLY_TYPE_OFFSET_DATA   1
 252 #define NBD_REPLY_TYPE_OFFSET_HOLE   2
 253 #define NBD_REPLY_TYPE_BLOCK_STATUS  5
 254 #define NBD_REPLY_TYPE_ERROR         NBD_REPLY_ERR(1)
 255 #define NBD_REPLY_TYPE_ERROR_OFFSET  NBD_REPLY_ERR(2)
 256
 257 /* Extent flags for base:allocation in NBD_REPLY_TYPE_BLOCK_STATUS */
 258 #define NBD_STATE_HOLE (1 << 0)
 259 #define NBD_STATE_ZERO (1 << 1)
 260
 261 /* Extent flags for qemu:dirty-bitmap in NBD_REPLY_TYPE_BLOCK_STATUS */
 262 #define NBD_STATE_DIRTY (1 << 0)
 263
 264 /* No flags needed for qemu:allocation-depth in NBD_REPLY_TYPE_BLOCK_STATUS */
 265
 266 static inline bool nbd_reply_type_is_error(int type)
 267 {
 268     return type & (1 << 15);
 269 }
 270
 271 /* NBD errors are based on errno numbers, so there is a 1:1 mapping,
 272  * but only a limited set of errno values is specified in the protocol.
 273  * Everything else is squashed to EINVAL.
 274  */
 275 #define NBD_SUCCESS    0
 276 #define NBD_EPERM      1
 277 #define NBD_EIO        5
 278 #define NBD_ENOMEM     12
 279 #define NBD_EINVAL     22
 280 #define NBD_ENOSPC     28
 281 #define NBD_EOVERFLOW  75
 282 #define NBD_ENOTSUP    95
 283 #define NBD_ESHUTDOWN  108
 284
 285 /* Details collected by NBD_OPT_EXPORT_NAME and NBD_OPT_GO */
 286 typedef struct NBDExportInfo {
 287     /* Set by client before nbd_receive_negotiate() */
 288     bool request_sizes;
 289     char *x_dirty_bitmap;
 290
 291     /* Set by client before nbd_receive_negotiate(), or by server results
 292      * during nbd_receive_export_list() */
 293     char *name; /* must be non-NULL */
 294
 295     /* In-out fields, set by client before nbd_receive_negotiate() and
 296      * updated by server results during nbd_receive_negotiate() */
 297     bool structured_reply;
 298     bool base_allocation; /* base:allocation context for NBD_CMD_BLOCK_STATUS */
 299
 300     /* Set by server results during nbd_receive_negotiate() and
 301      * nbd_receive_export_list() */
 302     uint64_t size;
 303     uint16_t flags;
 304     uint32_t min_block;
 305     uint32_t opt_block;
 306     uint32_t max_block;
 307
 308     uint32_t context_id;
 309
 310     /* Set by server results during nbd_receive_export_list() */
 311     char *description;
 312     int n_contexts;
 313     char **contexts;
 314 } NBDExportInfo;
 315
 316 int nbd_receive_negotiate(AioContext *aio_context, QIOChannel *ioc,
 317                           QCryptoTLSCreds *tlscreds,
 318                           const char *hostname, QIOChannel **outioc,
 319                           NBDExportInfo *info, Error **errp);
 320 void nbd_free_export_list(NBDExportInfo *info, int count);
 321 int nbd_receive_export_list(QIOChannel *ioc, QCryptoTLSCreds *tlscreds,
 322                             const char *hostname, NBDExportInfo **info,
 323                             Error **errp);
 324 int nbd_init(int fd, QIOChannelSocket *sioc, NBDExportInfo *info,
 325              Error **errp);
 326 int nbd_send_request(QIOChannel *ioc, NBDRequest *request);
 327 int coroutine_fn nbd_receive_reply(BlockDriverState *bs, QIOChannel *ioc,
 328                                    NBDReply *reply, Error **errp);
 329 int nbd_client(int fd);
 330 int nbd_disconnect(int fd);
 331 int nbd_errno_to_system_errno(int err);
 332
 333 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk);
 334
 335 AioContext *nbd_export_aio_context(NBDExport *exp);
 336 NBDExport *nbd_export_find(const char *name);
 337
 338 void nbd_client_new(QIOChannelSocket *sioc,
 339                     QCryptoTLSCreds *tlscreds,
 340                     const char *tlsauthz,
 341                     void (*close_fn)(NBDClient *, bool));
 342 void nbd_client_get(NBDClient *client);
 343 void nbd_client_put(NBDClient *client);
 344
 345 void nbd_server_is_qemu_nbd(int max_connections);
 346 bool nbd_server_is_running(void);
 347 int nbd_server_max_connections(void);
 348 void nbd_server_start(SocketAddress *addr, const char *tls_creds,
 349                       const char *tls_authz, uint32_t max_connections,
 350                       Error **errp);
 351 void nbd_server_start_options(NbdServerOptions *arg, Error **errp);
 352
 353 /* nbd_read
 354  * Reads @size bytes from @ioc. Returns 0 on success.
 355  */
 356 static inline int nbd_read(QIOChannel *ioc, void *buffer, size_t size,
 357                            const char *desc, Error **errp)
 358 {
 359     ERRP_GUARD();
 360     int ret = qio_channel_read_all(ioc, buffer, size, errp) < 0 ? -EIO : 0;
 361
 362     if (ret < 0) {
 363         if (desc) {
 364             error_prepend(errp, "Failed to read %s: ", desc);
 365         }
 366         return ret;
 367     }
 368
 369     return 0;
 370 }
 371
 372 #define DEF_NBD_READ_N(bits)                                            \
 373 static inline int nbd_read##bits(QIOChannel *ioc,                       \
 374                                  uint##bits##_t *val,                   \
 375                                  const char *desc, Error **errp)        \
 376 {                                                                       \
 377     int ret = nbd_read(ioc, val, sizeof(*val), desc, errp);             \
 378     if (ret < 0) {                                                      \
 379         return ret;                                                     \
 380     }                                                                   \
 381     *val = be##bits##_to_cpu(*val);                                     \
 382     return 0;                                                           \
 383 }
 384
 385 DEF_NBD_READ_N(16) /* Defines nbd_read16(). */
 386 DEF_NBD_READ_N(32) /* Defines nbd_read32(). */
 387 DEF_NBD_READ_N(64) /* Defines nbd_read64(). */
 388
 389 #undef DEF_NBD_READ_N
 390
 391 static inline bool nbd_reply_is_simple(NBDReply *reply)
 392 {
 393     return reply->magic == NBD_SIMPLE_REPLY_MAGIC;
 394 }
 395
 396 static inline bool nbd_reply_is_structured(NBDReply *reply)
 397 {
 398     return reply->magic == NBD_STRUCTURED_REPLY_MAGIC;
 399 }
 400
 401 const char *nbd_reply_type_lookup(uint16_t type);
 402 const char *nbd_opt_lookup(uint32_t opt);
 403 const char *nbd_rep_lookup(uint32_t rep);
 404 const char *nbd_info_lookup(uint16_t info);
 405 const char *nbd_cmd_lookup(uint16_t info);
 406 const char *nbd_err_lookup(int err);
 407
 408 /* nbd/client-connection.c */
 409 void nbd_client_connection_enable_retry(NBDClientConnection *conn);
 410
 411 NBDClientConnection *nbd_client_connection_new(const SocketAddress *saddr,
 412                                                bool do_negotiation,
 413                                                const char *export_name,
 414                                                const char *x_dirty_bitmap,
 415                                                QCryptoTLSCreds *tlscreds,
 416                                                const char *tlshostname);
 417 void nbd_client_connection_release(NBDClientConnection *conn);
 418
 419 QIOChannel *coroutine_fn
 420 nbd_co_establish_connection(NBDClientConnection *conn, NBDExportInfo *info,
 421                             bool blocking, Error **errp);
 422
 423 void nbd_co_establish_connection_cancel(NBDClientConnection *conn);
 424
 425 #endif