include/block/nbd.h

   1 /*
   2  *  Copyright (C) 2016-2017 Red Hat, Inc.
   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 "qapi/qapi-types-block.h"
  24 #include "io/channel-socket.h"
  25 #include "crypto/tlscreds.h"
  26
  27 /* Handshake phase structs - this struct is passed on the wire */
  28
  29 struct NBDOption {
  30     uint64_t magic; /* NBD_OPTS_MAGIC */
  31     uint32_t option; /* NBD_OPT_* */
  32     uint32_t length;
  33 } QEMU_PACKED;
  34 typedef struct NBDOption NBDOption;
  35
  36 struct NBDOptionReply {
  37     uint64_t magic; /* NBD_REP_MAGIC */
  38     uint32_t option; /* NBD_OPT_* */
  39     uint32_t type; /* NBD_REP_* */
  40     uint32_t length;
  41 } QEMU_PACKED;
  42 typedef struct NBDOptionReply NBDOptionReply;
  43
  44 typedef struct NBDOptionReplyMetaContext {
  45     NBDOptionReply h; /* h.type = NBD_REP_META_CONTEXT, h.length > 4 */
  46     uint32_t context_id;
  47     /* meta context name follows */
  48 } QEMU_PACKED NBDOptionReplyMetaContext;
  49
  50 /* Transmission phase structs
  51  *
  52  * Note: these are _NOT_ the same as the network representation of an NBD
  53  * request and reply!
  54  */
  55 struct NBDRequest {
  56     uint64_t handle;
  57     uint64_t from;
  58     uint32_t len;
  59     uint16_t flags; /* NBD_CMD_FLAG_* */
  60     uint16_t type; /* NBD_CMD_* */
  61 };
  62 typedef struct NBDRequest NBDRequest;
  63
  64 typedef struct NBDSimpleReply {
  65     uint32_t magic;  /* NBD_SIMPLE_REPLY_MAGIC */
  66     uint32_t error;
  67     uint64_t handle;
  68 } QEMU_PACKED NBDSimpleReply;
  69
  70 /* Header of all structured replies */
  71 typedef struct NBDStructuredReplyChunk {
  72     uint32_t magic;  /* NBD_STRUCTURED_REPLY_MAGIC */
  73     uint16_t flags;  /* combination of NBD_REPLY_FLAG_* */
  74     uint16_t type;   /* NBD_REPLY_TYPE_* */
  75     uint64_t handle; /* request handle */
  76     uint32_t length; /* length of payload */
  77 } QEMU_PACKED NBDStructuredReplyChunk;
  78
  79 typedef union NBDReply {
  80     NBDSimpleReply simple;
  81     NBDStructuredReplyChunk structured;
  82     struct {
  83         /* @magic and @handle fields have the same offset and size both in
  84          * simple reply and structured reply chunk, so let them be accessible
  85          * without ".simple." or ".structured." specification
  86          */
  87         uint32_t magic;
  88         uint32_t _skip;
  89         uint64_t handle;
  90     } QEMU_PACKED;
  91 } NBDReply;
  92
  93 /* Header of chunk for NBD_REPLY_TYPE_OFFSET_DATA */
  94 typedef struct NBDStructuredReadData {
  95     NBDStructuredReplyChunk h; /* h.length >= 9 */
  96     uint64_t offset;
  97     /* At least one byte of data payload follows, calculated from h.length */
  98 } QEMU_PACKED NBDStructuredReadData;
  99
 100 /* Complete chunk for NBD_REPLY_TYPE_OFFSET_HOLE */
 101 typedef struct NBDStructuredReadHole {
 102     NBDStructuredReplyChunk h; /* h.length == 12 */
 103     uint64_t offset;
 104     uint32_t length;
 105 } QEMU_PACKED NBDStructuredReadHole;
 106
 107 /* Header of all NBD_REPLY_TYPE_ERROR* errors */
 108 typedef struct NBDStructuredError {
 109     NBDStructuredReplyChunk h; /* h.length >= 6 */
 110     uint32_t error;
 111     uint16_t message_length;
 112 } QEMU_PACKED NBDStructuredError;
 113
 114 /* Header of NBD_REPLY_TYPE_BLOCK_STATUS */
 115 typedef struct NBDStructuredMeta {
 116     NBDStructuredReplyChunk h; /* h.length >= 12 (at least one extent) */
 117     uint32_t context_id;
 118     /* extents follows */
 119 } QEMU_PACKED NBDStructuredMeta;
 120
 121 /* Extent chunk for NBD_REPLY_TYPE_BLOCK_STATUS */
 122 typedef struct NBDExtent {
 123     uint32_t length;
 124     uint32_t flags; /* NBD_STATE_* */
 125 } QEMU_PACKED NBDExtent;
 126
 127 /* Transmission (export) flags: sent from server to client during handshake,
 128    but describe what will happen during transmission */
 129 #define NBD_FLAG_HAS_FLAGS         (1 << 0) /* Flags are there */
 130 #define NBD_FLAG_READ_ONLY         (1 << 1) /* Device is read-only */
 131 #define NBD_FLAG_SEND_FLUSH        (1 << 2) /* Send FLUSH */
 132 #define NBD_FLAG_SEND_FUA          (1 << 3) /* Send FUA (Force Unit Access) */
 133 #define NBD_FLAG_ROTATIONAL        (1 << 4) /* Use elevator algorithm -
 134                                                rotational media */
 135 #define NBD_FLAG_SEND_TRIM         (1 << 5) /* Send TRIM (discard) */
 136 #define NBD_FLAG_SEND_WRITE_ZEROES (1 << 6) /* Send WRITE_ZEROES */
 137 #define NBD_FLAG_SEND_DF           (1 << 7) /* Send DF (Do not Fragment) */
 138 #define NBD_FLAG_CAN_MULTI_CONN    (1 << 8) /* Multi-client cache consistent */
 139 #define NBD_FLAG_SEND_RESIZE       (1 << 9) /* Send resize */
 140 #define NBD_FLAG_SEND_CACHE        (1 << 10) /* Send CACHE (prefetch) */
 141
 142 /* New-style handshake (global) flags, sent from server to client, and
 143    control what will happen during handshake phase. */
 144 #define NBD_FLAG_FIXED_NEWSTYLE   (1 << 0) /* Fixed newstyle protocol. */
 145 #define NBD_FLAG_NO_ZEROES        (1 << 1) /* End handshake without zeroes. */
 146
 147 /* New-style client flags, sent from client to server to control what happens
 148    during handshake phase. */
 149 #define NBD_FLAG_C_FIXED_NEWSTYLE (1 << 0) /* Fixed newstyle protocol. */
 150 #define NBD_FLAG_C_NO_ZEROES      (1 << 1) /* End handshake without zeroes. */
 151
 152 /* Option requests. */
 153 #define NBD_OPT_EXPORT_NAME       (1)
 154 #define NBD_OPT_ABORT             (2)
 155 #define NBD_OPT_LIST              (3)
 156 /* #define NBD_OPT_PEEK_EXPORT    (4) not in use */
 157 #define NBD_OPT_STARTTLS          (5)
 158 #define NBD_OPT_INFO              (6)
 159 #define NBD_OPT_GO                (7)
 160 #define NBD_OPT_STRUCTURED_REPLY  (8)
 161 #define NBD_OPT_LIST_META_CONTEXT (9)
 162 #define NBD_OPT_SET_META_CONTEXT  (10)
 163
 164 /* Option reply types. */
 165 #define NBD_REP_ERR(value) ((UINT32_C(1) << 31) | (value))
 166
 167 #define NBD_REP_ACK             (1)    /* Data sending finished. */
 168 #define NBD_REP_SERVER          (2)    /* Export description. */
 169 #define NBD_REP_INFO            (3)    /* NBD_OPT_INFO/GO. */
 170 #define NBD_REP_META_CONTEXT    (4)    /* NBD_OPT_{LIST,SET}_META_CONTEXT */
 171
 172 #define NBD_REP_ERR_UNSUP           NBD_REP_ERR(1)  /* Unknown option */
 173 #define NBD_REP_ERR_POLICY          NBD_REP_ERR(2)  /* Server denied */
 174 #define NBD_REP_ERR_INVALID         NBD_REP_ERR(3)  /* Invalid length */
 175 #define NBD_REP_ERR_PLATFORM        NBD_REP_ERR(4)  /* Not compiled in */
 176 #define NBD_REP_ERR_TLS_REQD        NBD_REP_ERR(5)  /* TLS required */
 177 #define NBD_REP_ERR_UNKNOWN         NBD_REP_ERR(6)  /* Export unknown */
 178 #define NBD_REP_ERR_SHUTDOWN        NBD_REP_ERR(7)  /* Server shutting down */
 179 #define NBD_REP_ERR_BLOCK_SIZE_REQD NBD_REP_ERR(8)  /* Need INFO_BLOCK_SIZE */
 180
 181 /* Info types, used during NBD_REP_INFO */
 182 #define NBD_INFO_EXPORT         0
 183 #define NBD_INFO_NAME           1
 184 #define NBD_INFO_DESCRIPTION    2
 185 #define NBD_INFO_BLOCK_SIZE     3
 186
 187 /* Request flags, sent from client to server during transmission phase */
 188 #define NBD_CMD_FLAG_FUA        (1 << 0) /* 'force unit access' during write */
 189 #define NBD_CMD_FLAG_NO_HOLE    (1 << 1) /* don't punch hole on zero run */
 190 #define NBD_CMD_FLAG_DF         (1 << 2) /* don't fragment structured read */
 191 #define NBD_CMD_FLAG_REQ_ONE    (1 << 3) /* only one extent in BLOCK_STATUS
 192                                           * reply chunk */
 193
 194 /* Supported request types */
 195 enum {
 196     NBD_CMD_READ = 0,
 197     NBD_CMD_WRITE = 1,
 198     NBD_CMD_DISC = 2,
 199     NBD_CMD_FLUSH = 3,
 200     NBD_CMD_TRIM = 4,
 201     NBD_CMD_CACHE = 5,
 202     NBD_CMD_WRITE_ZEROES = 6,
 203     NBD_CMD_BLOCK_STATUS = 7,
 204 };
 205
 206 #define NBD_DEFAULT_PORT        10809
 207
 208 /* Maximum size of a single READ/WRITE data buffer */
 209 #define NBD_MAX_BUFFER_SIZE (32 * 1024 * 1024)
 210
 211 /* Maximum size of an export name. The NBD spec requires 256 and
 212  * suggests that servers support up to 4096, but we stick to only the
 213  * required size so that we can stack-allocate the names, and because
 214  * going larger would require an audit of more code to make sure we
 215  * aren't overflowing some other buffer. */
 216 #define NBD_MAX_NAME_SIZE 256
 217
 218 /* Two types of reply structures */
 219 #define NBD_SIMPLE_REPLY_MAGIC      0x67446698
 220 #define NBD_STRUCTURED_REPLY_MAGIC  0x668e33ef
 221
 222 /* Structured reply flags */
 223 #define NBD_REPLY_FLAG_DONE          (1 << 0) /* This reply-chunk is last */
 224
 225 /* Structured reply types */
 226 #define NBD_REPLY_ERR(value)         ((1 << 15) | (value))
 227
 228 #define NBD_REPLY_TYPE_NONE          0
 229 #define NBD_REPLY_TYPE_OFFSET_DATA   1
 230 #define NBD_REPLY_TYPE_OFFSET_HOLE   2
 231 #define NBD_REPLY_TYPE_BLOCK_STATUS  5
 232 #define NBD_REPLY_TYPE_ERROR         NBD_REPLY_ERR(1)
 233 #define NBD_REPLY_TYPE_ERROR_OFFSET  NBD_REPLY_ERR(2)
 234
 235 /* Extent flags for base:allocation in NBD_REPLY_TYPE_BLOCK_STATUS */
 236 #define NBD_STATE_HOLE (1 << 0)
 237 #define NBD_STATE_ZERO (1 << 1)
 238
 239 /* Extent flags for qemu:dirty-bitmap in NBD_REPLY_TYPE_BLOCK_STATUS */
 240 #define NBD_STATE_DIRTY (1 << 0)
 241
 242 static inline bool nbd_reply_type_is_error(int type)
 243 {
 244     return type & (1 << 15);
 245 }
 246
 247 /* NBD errors are based on errno numbers, so there is a 1:1 mapping,
 248  * but only a limited set of errno values is specified in the protocol.
 249  * Everything else is squashed to EINVAL.
 250  */
 251 #define NBD_SUCCESS    0
 252 #define NBD_EPERM      1
 253 #define NBD_EIO        5
 254 #define NBD_ENOMEM     12
 255 #define NBD_EINVAL     22
 256 #define NBD_ENOSPC     28
 257 #define NBD_EOVERFLOW  75
 258 #define NBD_ESHUTDOWN  108
 259
 260 /* Details collected by NBD_OPT_EXPORT_NAME and NBD_OPT_GO */
 261 struct NBDExportInfo {
 262     /* Set by client before nbd_receive_negotiate() */
 263     bool request_sizes;
 264     char *x_dirty_bitmap;
 265
 266     /* In-out fields, set by client before nbd_receive_negotiate() and
 267      * updated by server results during nbd_receive_negotiate() */
 268     bool structured_reply;
 269     bool base_allocation; /* base:allocation context for NBD_CMD_BLOCK_STATUS */
 270
 271     /* Set by server results during nbd_receive_negotiate() */
 272     uint64_t size;
 273     uint16_t flags;
 274     uint32_t min_block;
 275     uint32_t opt_block;
 276     uint32_t max_block;
 277
 278     uint32_t meta_base_allocation_id;
 279 };
 280 typedef struct NBDExportInfo NBDExportInfo;
 281
 282 int nbd_receive_negotiate(QIOChannel *ioc, const char *name,
 283                           QCryptoTLSCreds *tlscreds, const char *hostname,
 284                           QIOChannel **outioc, NBDExportInfo *info,
 285                           Error **errp);
 286 int nbd_init(int fd, QIOChannelSocket *sioc, NBDExportInfo *info,
 287              Error **errp);
 288 int nbd_send_request(QIOChannel *ioc, NBDRequest *request);
 289 int nbd_receive_reply(QIOChannel *ioc, NBDReply *reply, Error **errp);
 290 int nbd_client(int fd);
 291 int nbd_disconnect(int fd);
 292 int nbd_errno_to_system_errno(int err);
 293
 294 typedef struct NBDExport NBDExport;
 295 typedef struct NBDClient NBDClient;
 296
 297 NBDExport *nbd_export_new(BlockDriverState *bs, uint64_t dev_offset,
 298                           uint64_t size, const char *name, const char *desc,
 299                           const char *bitmap, uint16_t nbdflags,
 300                           void (*close)(NBDExport *), bool writethrough,
 301                           BlockBackend *on_eject_blk, Error **errp);
 302 void nbd_export_close(NBDExport *exp);
 303 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp);
 304 void nbd_export_get(NBDExport *exp);
 305 void nbd_export_put(NBDExport *exp);
 306
 307 BlockBackend *nbd_export_get_blockdev(NBDExport *exp);
 308
 309 NBDExport *nbd_export_find(const char *name);
 310 void nbd_export_close_all(void);
 311
 312 void nbd_client_new(QIOChannelSocket *sioc,
 313                     QCryptoTLSCreds *tlscreds,
 314                     const char *tlsaclname,
 315                     void (*close_fn)(NBDClient *, bool));
 316 void nbd_client_get(NBDClient *client);
 317 void nbd_client_put(NBDClient *client);
 318
 319 void nbd_server_start(SocketAddress *addr, const char *tls_creds,
 320                       Error **errp);
 321
 322 /* nbd_read
 323  * Reads @size bytes from @ioc. Returns 0 on success.
 324  */
 325 static inline int nbd_read(QIOChannel *ioc, void *buffer, size_t size,
 326                            Error **errp)
 327 {
 328     return qio_channel_read_all(ioc, buffer, size, errp) < 0 ? -EIO : 0;
 329 }
 330
 331 static inline bool nbd_reply_is_simple(NBDReply *reply)
 332 {
 333     return reply->magic == NBD_SIMPLE_REPLY_MAGIC;
 334 }
 335
 336 static inline bool nbd_reply_is_structured(NBDReply *reply)
 337 {
 338     return reply->magic == NBD_STRUCTURED_REPLY_MAGIC;
 339 }
 340
 341 const char *nbd_reply_type_lookup(uint16_t type);
 342 const char *nbd_opt_lookup(uint32_t opt);
 343 const char *nbd_rep_lookup(uint32_t rep);
 344 const char *nbd_info_lookup(uint16_t info);
 345 const char *nbd_cmd_lookup(uint16_t info);
 346 const char *nbd_err_lookup(int err);
 347
 348 #endif