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