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