qemu/block/sheepdog.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 2009-2010 Nippon Telegraph and Telephone Corporation.
   3 *
   4 * This program is free software; you can redistribute it and/or
   5 * modify it under the terms of the GNU General Public License version
   6 * 2 as published by the Free Software Foundation.
   7 *
   8 * You should have received a copy of the GNU General Public License
   9 * along with this program. If not, see <http://www.gnu.org/licenses/>.
  10 *
  11 * Contributions after 2012-01-13 are licensed under the terms of the
  12 * GNU GPL, version 2 or (at your option) any later version.
  13 */
  14
  15#include "qemu/osdep.h"
  16#include "qapi/error.h"
  17#include "qapi/qapi-visit-sockets.h"
  18#include "qapi/qapi-visit-block-core.h"
  19#include "qapi/qmp/qdict.h"
  20#include "qapi/qobject-input-visitor.h"
  21#include "qapi/qobject-output-visitor.h"
  22#include "qemu/uri.h"
  23#include "qemu/error-report.h"
  24#include "qemu/option.h"
  25#include "qemu/sockets.h"
  26#include "block/block_int.h"
  27#include "block/qdict.h"
  28#include "sysemu/block-backend.h"
  29#include "qemu/bitops.h"
  30#include "qemu/cutils.h"
  31
  32#define SD_PROTO_VER 0x01
  33
  34#define SD_DEFAULT_ADDR "localhost"
  35#define SD_DEFAULT_PORT 7000
  36
  37#define SD_OP_CREATE_AND_WRITE_OBJ  0x01
  38#define SD_OP_READ_OBJ       0x02
  39#define SD_OP_WRITE_OBJ      0x03
  40/* 0x04 is used internally by Sheepdog */
  41
  42#define SD_OP_NEW_VDI        0x11
  43#define SD_OP_LOCK_VDI       0x12
  44#define SD_OP_RELEASE_VDI    0x13
  45#define SD_OP_GET_VDI_INFO   0x14
  46#define SD_OP_READ_VDIS      0x15
  47#define SD_OP_FLUSH_VDI      0x16
  48#define SD_OP_DEL_VDI        0x17
  49#define SD_OP_GET_CLUSTER_DEFAULT   0x18
  50
  51#define SD_FLAG_CMD_WRITE    0x01
  52#define SD_FLAG_CMD_COW      0x02
  53#define SD_FLAG_CMD_CACHE    0x04 /* Writeback mode for cache */
  54#define SD_FLAG_CMD_DIRECT   0x08 /* Don't use cache */
  55
  56#define SD_RES_SUCCESS       0x00 /* Success */
  57#define SD_RES_UNKNOWN       0x01 /* Unknown error */
  58#define SD_RES_NO_OBJ        0x02 /* No object found */
  59#define SD_RES_EIO           0x03 /* I/O error */
  60#define SD_RES_VDI_EXIST     0x04 /* Vdi exists already */
  61#define SD_RES_INVALID_PARMS 0x05 /* Invalid parameters */
  62#define SD_RES_SYSTEM_ERROR  0x06 /* System error */
  63#define SD_RES_VDI_LOCKED    0x07 /* Vdi is locked */
  64#define SD_RES_NO_VDI        0x08 /* No vdi found */
  65#define SD_RES_NO_BASE_VDI   0x09 /* No base vdi found */
  66#define SD_RES_VDI_READ      0x0A /* Cannot read requested vdi */
  67#define SD_RES_VDI_WRITE     0x0B /* Cannot write requested vdi */
  68#define SD_RES_BASE_VDI_READ 0x0C /* Cannot read base vdi */
  69#define SD_RES_BASE_VDI_WRITE   0x0D /* Cannot write base vdi */
  70#define SD_RES_NO_TAG        0x0E /* Requested tag is not found */
  71#define SD_RES_STARTUP       0x0F /* Sheepdog is on starting up */
  72#define SD_RES_VDI_NOT_LOCKED   0x10 /* Vdi is not locked */
  73#define SD_RES_SHUTDOWN      0x11 /* Sheepdog is shutting down */
  74#define SD_RES_NO_MEM        0x12 /* Cannot allocate memory */
  75#define SD_RES_FULL_VDI      0x13 /* we already have the maximum vdis */
  76#define SD_RES_VER_MISMATCH  0x14 /* Protocol version mismatch */
  77#define SD_RES_NO_SPACE      0x15 /* Server has no room for new objects */
  78#define SD_RES_WAIT_FOR_FORMAT  0x16 /* Waiting for a format operation */
  79#define SD_RES_WAIT_FOR_JOIN    0x17 /* Waiting for other nodes joining */
  80#define SD_RES_JOIN_FAILED   0x18 /* Target node had failed to join sheepdog */
  81#define SD_RES_HALT          0x19 /* Sheepdog is stopped serving IO request */
  82#define SD_RES_READONLY      0x1A /* Object is read-only */
  83
  84/*
  85 * Object ID rules
  86 *
  87 *  0 - 19 (20 bits): data object space
  88 * 20 - 31 (12 bits): reserved data object space
  89 * 32 - 55 (24 bits): vdi object space
  90 * 56 - 59 ( 4 bits): reserved vdi object space
  91 * 60 - 63 ( 4 bits): object type identifier space
  92 */
  93
  94#define VDI_SPACE_SHIFT   32
  95#define VDI_BIT (UINT64_C(1) << 63)
  96#define VMSTATE_BIT (UINT64_C(1) << 62)
  97#define MAX_DATA_OBJS (UINT64_C(1) << 20)
  98#define MAX_CHILDREN 1024
  99#define SD_MAX_VDI_LEN 256
 100#define SD_MAX_VDI_TAG_LEN 256
 101#define SD_NR_VDIS   (1U << 24)
 102#define SD_DATA_OBJ_SIZE (UINT64_C(1) << 22)
 103#define SD_MAX_VDI_SIZE (SD_DATA_OBJ_SIZE * MAX_DATA_OBJS)
 104#define SD_DEFAULT_BLOCK_SIZE_SHIFT 22
 105/*
 106 * For erasure coding, we use at most SD_EC_MAX_STRIP for data strips and
 107 * (SD_EC_MAX_STRIP - 1) for parity strips
 108 *
 109 * SD_MAX_COPIES is sum of number of data strips and parity strips.
 110 */
 111#define SD_EC_MAX_STRIP 16
 112#define SD_MAX_COPIES (SD_EC_MAX_STRIP * 2 - 1)
 113
 114#define SD_INODE_SIZE (sizeof(SheepdogInode))
 115#define CURRENT_VDI_ID 0
 116
 117#define LOCK_TYPE_NORMAL 0
 118#define LOCK_TYPE_SHARED 1      /* for iSCSI multipath */
 119
 120typedef struct SheepdogReq {
 121    uint8_t proto_ver;
 122    uint8_t opcode;
 123    uint16_t flags;
 124    uint32_t epoch;
 125    uint32_t id;
 126    uint32_t data_length;
 127    uint32_t opcode_specific[8];
 128} SheepdogReq;
 129
 130typedef struct SheepdogRsp {
 131    uint8_t proto_ver;
 132    uint8_t opcode;
 133    uint16_t flags;
 134    uint32_t epoch;
 135    uint32_t id;
 136    uint32_t data_length;
 137    uint32_t result;
 138    uint32_t opcode_specific[7];
 139} SheepdogRsp;
 140
 141typedef struct SheepdogObjReq {
 142    uint8_t proto_ver;
 143    uint8_t opcode;
 144    uint16_t flags;
 145    uint32_t epoch;
 146    uint32_t id;
 147    uint32_t data_length;
 148    uint64_t oid;
 149    uint64_t cow_oid;
 150    uint8_t copies;
 151    uint8_t copy_policy;
 152    uint8_t reserved[6];
 153    uint64_t offset;
 154} SheepdogObjReq;
 155
 156typedef struct SheepdogObjRsp {
 157    uint8_t proto_ver;
 158    uint8_t opcode;
 159    uint16_t flags;
 160    uint32_t epoch;
 161    uint32_t id;
 162    uint32_t data_length;
 163    uint32_t result;
 164    uint8_t copies;
 165    uint8_t copy_policy;
 166    uint8_t reserved[2];
 167    uint32_t pad[6];
 168} SheepdogObjRsp;
 169
 170typedef struct SheepdogVdiReq {
 171    uint8_t proto_ver;
 172    uint8_t opcode;
 173    uint16_t flags;
 174    uint32_t epoch;
 175    uint32_t id;
 176    uint32_t data_length;
 177    uint64_t vdi_size;
 178    uint32_t base_vdi_id;
 179    uint8_t copies;
 180    uint8_t copy_policy;
 181    uint8_t store_policy;
 182    uint8_t block_size_shift;
 183    uint32_t snapid;
 184    uint32_t type;
 185    uint32_t pad[2];
 186} SheepdogVdiReq;
 187
 188typedef struct SheepdogVdiRsp {
 189    uint8_t proto_ver;
 190    uint8_t opcode;
 191    uint16_t flags;
 192    uint32_t epoch;
 193    uint32_t id;
 194    uint32_t data_length;
 195    uint32_t result;
 196    uint32_t rsvd;
 197    uint32_t vdi_id;
 198    uint32_t pad[5];
 199} SheepdogVdiRsp;
 200
 201typedef struct SheepdogClusterRsp {
 202    uint8_t proto_ver;
 203    uint8_t opcode;
 204    uint16_t flags;
 205    uint32_t epoch;
 206    uint32_t id;
 207    uint32_t data_length;
 208    uint32_t result;
 209    uint8_t nr_copies;
 210    uint8_t copy_policy;
 211    uint8_t block_size_shift;
 212    uint8_t __pad1;
 213    uint32_t __pad2[6];
 214} SheepdogClusterRsp;
 215
 216typedef struct SheepdogInode {
 217    char name[SD_MAX_VDI_LEN];
 218    char tag[SD_MAX_VDI_TAG_LEN];
 219    uint64_t ctime;
 220    uint64_t snap_ctime;
 221    uint64_t vm_clock_nsec;
 222    uint64_t vdi_size;
 223    uint64_t vm_state_size;
 224    uint16_t copy_policy;
 225    uint8_t nr_copies;
 226    uint8_t block_size_shift;
 227    uint32_t snap_id;
 228    uint32_t vdi_id;
 229    uint32_t parent_vdi_id;
 230    uint32_t child_vdi_id[MAX_CHILDREN];
 231    uint32_t data_vdi_id[MAX_DATA_OBJS];
 232} SheepdogInode;
 233
 234#define SD_INODE_HEADER_SIZE offsetof(SheepdogInode, data_vdi_id)
 235
 236/*
 237 * 64 bit FNV-1a non-zero initial basis
 238 */
 239#define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
 240
 241/*
 242 * 64 bit Fowler/Noll/Vo FNV-1a hash code
 243 */
 244static inline uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval)
 245{
 246    unsigned char *bp = buf;
 247    unsigned char *be = bp + len;
 248    while (bp < be) {
 249        hval ^= (uint64_t) *bp++;
 250        hval += (hval << 1) + (hval << 4) + (hval << 5) +
 251            (hval << 7) + (hval << 8) + (hval << 40);
 252    }
 253    return hval;
 254}
 255
 256static inline bool is_data_obj_writable(SheepdogInode *inode, unsigned int idx)
 257{
 258    return inode->vdi_id == inode->data_vdi_id[idx];
 259}
 260
 261static inline bool is_data_obj(uint64_t oid)
 262{
 263    return !(VDI_BIT & oid);
 264}
 265
 266static inline uint64_t data_oid_to_idx(uint64_t oid)
 267{
 268    return oid & (MAX_DATA_OBJS - 1);
 269}
 270
 271static inline uint32_t oid_to_vid(uint64_t oid)
 272{
 273    return (oid & ~VDI_BIT) >> VDI_SPACE_SHIFT;
 274}
 275
 276static inline uint64_t vid_to_vdi_oid(uint32_t vid)
 277{
 278    return VDI_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT);
 279}
 280
 281static inline uint64_t vid_to_vmstate_oid(uint32_t vid, uint32_t idx)
 282{
 283    return VMSTATE_BIT | ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
 284}
 285
 286static inline uint64_t vid_to_data_oid(uint32_t vid, uint32_t idx)
 287{
 288    return ((uint64_t)vid << VDI_SPACE_SHIFT) | idx;
 289}
 290
 291static inline bool is_snapshot(struct SheepdogInode *inode)
 292{
 293    return !!inode->snap_ctime;
 294}
 295
 296static inline size_t count_data_objs(const struct SheepdogInode *inode)
 297{
 298    return DIV_ROUND_UP(inode->vdi_size,
 299                        (1UL << inode->block_size_shift));
 300}
 301
 302#undef DPRINTF
 303#ifdef DEBUG_SDOG
 304#define DEBUG_SDOG_PRINT 1
 305#else
 306#define DEBUG_SDOG_PRINT 0
 307#endif
 308#define DPRINTF(fmt, args...)                                           \
 309    do {                                                                \
 310        if (DEBUG_SDOG_PRINT) {                                         \
 311            fprintf(stderr, "%s %d: " fmt, __func__, __LINE__, ##args); \
 312        }                                                               \
 313    } while (0)
 314
 315typedef struct SheepdogAIOCB SheepdogAIOCB;
 316typedef struct BDRVSheepdogState BDRVSheepdogState;
 317
 318typedef struct AIOReq {
 319    SheepdogAIOCB *aiocb;
 320    unsigned int iov_offset;
 321
 322    uint64_t oid;
 323    uint64_t base_oid;
 324    uint64_t offset;
 325    unsigned int data_len;
 326    uint8_t flags;
 327    uint32_t id;
 328    bool create;
 329
 330    QLIST_ENTRY(AIOReq) aio_siblings;
 331} AIOReq;
 332
 333enum AIOCBState {
 334    AIOCB_WRITE_UDATA,
 335    AIOCB_READ_UDATA,
 336    AIOCB_FLUSH_CACHE,
 337    AIOCB_DISCARD_OBJ,
 338};
 339
 340#define AIOCBOverlapping(x, y)                                 \
 341    (!(x->max_affect_data_idx < y->min_affect_data_idx          \
 342       || y->max_affect_data_idx < x->min_affect_data_idx))
 343
 344struct SheepdogAIOCB {
 345    BDRVSheepdogState *s;
 346
 347    QEMUIOVector *qiov;
 348
 349    int64_t sector_num;
 350    int nb_sectors;
 351
 352    int ret;
 353    enum AIOCBState aiocb_type;
 354
 355    Coroutine *coroutine;
 356    int nr_pending;
 357
 358    uint32_t min_affect_data_idx;
 359    uint32_t max_affect_data_idx;
 360
 361    /*
 362     * The difference between affect_data_idx and dirty_data_idx:
 363     * affect_data_idx represents range of index of all request types.
 364     * dirty_data_idx represents range of index updated by COW requests.
 365     * dirty_data_idx is used for updating an inode object.
 366     */
 367    uint32_t min_dirty_data_idx;
 368    uint32_t max_dirty_data_idx;
 369
 370    QLIST_ENTRY(SheepdogAIOCB) aiocb_siblings;
 371};
 372
 373struct BDRVSheepdogState {
 374    BlockDriverState *bs;
 375    AioContext *aio_context;
 376
 377    SheepdogInode inode;
 378
 379    char name[SD_MAX_VDI_LEN];
 380    bool is_snapshot;
 381    uint32_t cache_flags;
 382    bool discard_supported;
 383
 384    SocketAddress *addr;
 385    int fd;
 386
 387    CoMutex lock;
 388    Coroutine *co_send;
 389    Coroutine *co_recv;
 390
 391    uint32_t aioreq_seq_num;
 392
 393    /* Every aio request must be linked to either of these queues. */
 394    QLIST_HEAD(inflight_aio_head, AIOReq) inflight_aio_head;
 395    QLIST_HEAD(failed_aio_head, AIOReq) failed_aio_head;
 396
 397    CoMutex queue_lock;
 398    CoQueue overlapping_queue;
 399    QLIST_HEAD(inflight_aiocb_head, SheepdogAIOCB) inflight_aiocb_head;
 400};
 401
 402typedef struct BDRVSheepdogReopenState {
 403    int fd;
 404    int cache_flags;
 405} BDRVSheepdogReopenState;
 406
 407static const char *sd_strerror(int err)
 408{
 409    int i;
 410
 411    static const struct {
 412        int err;
 413        const char *desc;
 414    } errors[] = {
 415        {SD_RES_SUCCESS, "Success"},
 416        {SD_RES_UNKNOWN, "Unknown error"},
 417        {SD_RES_NO_OBJ, "No object found"},
 418        {SD_RES_EIO, "I/O error"},
 419        {SD_RES_VDI_EXIST, "VDI exists already"},
 420        {SD_RES_INVALID_PARMS, "Invalid parameters"},
 421        {SD_RES_SYSTEM_ERROR, "System error"},
 422        {SD_RES_VDI_LOCKED, "VDI is already locked"},
 423        {SD_RES_NO_VDI, "No vdi found"},
 424        {SD_RES_NO_BASE_VDI, "No base VDI found"},
 425        {SD_RES_VDI_READ, "Failed read the requested VDI"},
 426        {SD_RES_VDI_WRITE, "Failed to write the requested VDI"},
 427        {SD_RES_BASE_VDI_READ, "Failed to read the base VDI"},
 428        {SD_RES_BASE_VDI_WRITE, "Failed to write the base VDI"},
 429        {SD_RES_NO_TAG, "Failed to find the requested tag"},
 430        {SD_RES_STARTUP, "The system is still booting"},
 431        {SD_RES_VDI_NOT_LOCKED, "VDI isn't locked"},
 432        {SD_RES_SHUTDOWN, "The system is shutting down"},
 433        {SD_RES_NO_MEM, "Out of memory on the server"},
 434        {SD_RES_FULL_VDI, "We already have the maximum vdis"},
 435        {SD_RES_VER_MISMATCH, "Protocol version mismatch"},
 436        {SD_RES_NO_SPACE, "Server has no space for new objects"},
 437        {SD_RES_WAIT_FOR_FORMAT, "Sheepdog is waiting for a format operation"},
 438        {SD_RES_WAIT_FOR_JOIN, "Sheepdog is waiting for other nodes joining"},
 439        {SD_RES_JOIN_FAILED, "Target node had failed to join sheepdog"},
 440        {SD_RES_HALT, "Sheepdog is stopped serving IO request"},
 441        {SD_RES_READONLY, "Object is read-only"},
 442    };
 443
 444    for (i = 0; i < ARRAY_SIZE(errors); ++i) {
 445        if (errors[i].err == err) {
 446            return errors[i].desc;
 447        }
 448    }
 449
 450    return "Invalid error code";
 451}
 452
 453/*
 454 * Sheepdog I/O handling:
 455 *
 456 * 1. In sd_co_rw_vector, we send the I/O requests to the server and
 457 *    link the requests to the inflight_list in the
 458 *    BDRVSheepdogState.  The function yields while waiting for
 459 *    receiving the response.
 460 *
 461 * 2. We receive the response in aio_read_response, the fd handler to
 462 *    the sheepdog connection.  We switch back to sd_co_readv/sd_writev
 463 *    after all the requests belonging to the AIOCB are finished.  If
 464 *    needed, sd_co_writev will send another requests for the vdi object.
 465 */
 466
 467static inline AIOReq *alloc_aio_req(BDRVSheepdogState *s, SheepdogAIOCB *acb,
 468                                    uint64_t oid, unsigned int data_len,
 469                                    uint64_t offset, uint8_t flags, bool create,
 470                                    uint64_t base_oid, unsigned int iov_offset)
 471{
 472    AIOReq *aio_req;
 473
 474    aio_req = g_malloc(sizeof(*aio_req));
 475    aio_req->aiocb = acb;
 476    aio_req->iov_offset = iov_offset;
 477    aio_req->oid = oid;
 478    aio_req->base_oid = base_oid;
 479    aio_req->offset = offset;
 480    aio_req->data_len = data_len;
 481    aio_req->flags = flags;
 482    aio_req->id = s->aioreq_seq_num++;
 483    aio_req->create = create;
 484
 485    acb->nr_pending++;
 486    return aio_req;
 487}
 488
 489static void wait_for_overlapping_aiocb(BDRVSheepdogState *s, SheepdogAIOCB *acb)
 490{
 491    SheepdogAIOCB *cb;
 492
 493retry:
 494    QLIST_FOREACH(cb, &s->inflight_aiocb_head, aiocb_siblings) {
 495        if (AIOCBOverlapping(acb, cb)) {
 496            qemu_co_queue_wait(&s->overlapping_queue, &s->queue_lock);
 497            goto retry;
 498        }
 499    }
 500}
 501
 502static void sd_aio_setup(SheepdogAIOCB *acb, BDRVSheepdogState *s,
 503                         QEMUIOVector *qiov, int64_t sector_num, int nb_sectors,
 504                         int type)
 505{
 506    uint32_t object_size;
 507
 508    object_size = (UINT32_C(1) << s->inode.block_size_shift);
 509
 510    acb->s = s;
 511
 512    acb->qiov = qiov;
 513
 514    acb->sector_num = sector_num;
 515    acb->nb_sectors = nb_sectors;
 516
 517    acb->coroutine = qemu_coroutine_self();
 518    acb->ret = 0;
 519    acb->nr_pending = 0;
 520
 521    acb->min_affect_data_idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size;
 522    acb->max_affect_data_idx = (acb->sector_num * BDRV_SECTOR_SIZE +
 523                              acb->nb_sectors * BDRV_SECTOR_SIZE) / object_size;
 524
 525    acb->min_dirty_data_idx = UINT32_MAX;
 526    acb->max_dirty_data_idx = 0;
 527    acb->aiocb_type = type;
 528
 529    if (type == AIOCB_FLUSH_CACHE) {
 530        return;
 531    }
 532
 533    qemu_co_mutex_lock(&s->queue_lock);
 534    wait_for_overlapping_aiocb(s, acb);
 535    QLIST_INSERT_HEAD(&s->inflight_aiocb_head, acb, aiocb_siblings);
 536    qemu_co_mutex_unlock(&s->queue_lock);
 537}
 538
 539static SocketAddress *sd_server_config(QDict *options, Error **errp)
 540{
 541    QDict *server = NULL;
 542    Visitor *iv = NULL;
 543    SocketAddress *saddr = NULL;
 544    Error *local_err = NULL;
 545
 546    qdict_extract_subqdict(options, &server, "server.");
 547
 548    iv = qobject_input_visitor_new_flat_confused(server, errp);
 549    if (!iv) {
 550        goto done;
 551    }
 552
 553    visit_type_SocketAddress(iv, NULL, &saddr, &local_err);
 554    if (local_err) {
 555        error_propagate(errp, local_err);
 556        goto done;
 557    }
 558
 559done:
 560    visit_free(iv);
 561    qobject_unref(server);
 562    return saddr;
 563}
 564
 565/* Return -EIO in case of error, file descriptor on success */
 566static int connect_to_sdog(BDRVSheepdogState *s, Error **errp)
 567{
 568    int fd;
 569
 570    fd = socket_connect(s->addr, errp);
 571
 572    if (s->addr->type == SOCKET_ADDRESS_TYPE_INET && fd >= 0) {
 573        int ret = socket_set_nodelay(fd);
 574        if (ret < 0) {
 575            warn_report("can't set TCP_NODELAY: %s", strerror(errno));
 576        }
 577    }
 578
 579    if (fd >= 0) {
 580        qemu_set_nonblock(fd);
 581    } else {
 582        fd = -EIO;
 583    }
 584
 585    return fd;
 586}
 587
 588/* Return 0 on success and -errno in case of error */
 589static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data,
 590                                    unsigned int *wlen)
 591{
 592    int ret;
 593
 594    ret = qemu_co_send(sockfd, hdr, sizeof(*hdr));
 595    if (ret != sizeof(*hdr)) {
 596        error_report("failed to send a req, %s", strerror(errno));
 597        return -errno;
 598    }
 599
 600    ret = qemu_co_send(sockfd, data, *wlen);
 601    if (ret != *wlen) {
 602        error_report("failed to send a req, %s", strerror(errno));
 603        return -errno;
 604    }
 605
 606    return ret;
 607}
 608
 609typedef struct SheepdogReqCo {
 610    int sockfd;
 611    BlockDriverState *bs;
 612    AioContext *aio_context;
 613    SheepdogReq *hdr;
 614    void *data;
 615    unsigned int *wlen;
 616    unsigned int *rlen;
 617    int ret;
 618    bool finished;
 619    Coroutine *co;
 620} SheepdogReqCo;
 621
 622static void restart_co_req(void *opaque)
 623{
 624    SheepdogReqCo *srco = opaque;
 625
 626    aio_co_wake(srco->co);
 627}
 628
 629static coroutine_fn void do_co_req(void *opaque)
 630{
 631    int ret;
 632    SheepdogReqCo *srco = opaque;
 633    int sockfd = srco->sockfd;
 634    SheepdogReq *hdr = srco->hdr;
 635    void *data = srco->data;
 636    unsigned int *wlen = srco->wlen;
 637    unsigned int *rlen = srco->rlen;
 638
 639    srco->co = qemu_coroutine_self();
 640    aio_set_fd_handler(srco->aio_context, sockfd, false,
 641                       NULL, restart_co_req, NULL, srco);
 642
 643    ret = send_co_req(sockfd, hdr, data, wlen);
 644    if (ret < 0) {
 645        goto out;
 646    }
 647
 648    aio_set_fd_handler(srco->aio_context, sockfd, false,
 649                       restart_co_req, NULL, NULL, srco);
 650
 651    ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
 652    if (ret != sizeof(*hdr)) {
 653        error_report("failed to get a rsp, %s", strerror(errno));
 654        ret = -errno;
 655        goto out;
 656    }
 657
 658    if (*rlen > hdr->data_length) {
 659        *rlen = hdr->data_length;
 660    }
 661
 662    if (*rlen) {
 663        ret = qemu_co_recv(sockfd, data, *rlen);
 664        if (ret != *rlen) {
 665            error_report("failed to get the data, %s", strerror(errno));
 666            ret = -errno;
 667            goto out;
 668        }
 669    }
 670    ret = 0;
 671out:
 672    /* there is at most one request for this sockfd, so it is safe to
 673     * set each handler to NULL. */
 674    aio_set_fd_handler(srco->aio_context, sockfd, false,
 675                       NULL, NULL, NULL, NULL);
 676
 677    srco->co = NULL;
 678    srco->ret = ret;
 679    /* Set srco->finished before reading bs->wakeup.  */
 680    atomic_mb_set(&srco->finished, true);
 681    if (srco->bs) {
 682        bdrv_wakeup(srco->bs);
 683    }
 684}
 685
 686/*
 687 * Send the request to the sheep in a synchronous manner.
 688 *
 689 * Return 0 on success, -errno in case of error.
 690 */
 691static int do_req(int sockfd, BlockDriverState *bs, SheepdogReq *hdr,
 692                  void *data, unsigned int *wlen, unsigned int *rlen)
 693{
 694    Coroutine *co;
 695    SheepdogReqCo srco = {
 696        .sockfd = sockfd,
 697        .aio_context = bs ? bdrv_get_aio_context(bs) : qemu_get_aio_context(),
 698        .bs = bs,
 699        .hdr = hdr,
 700        .data = data,
 701        .wlen = wlen,
 702        .rlen = rlen,
 703        .ret = 0,
 704        .finished = false,
 705    };
 706
 707    if (qemu_in_coroutine()) {
 708        do_co_req(&srco);
 709    } else {
 710        co = qemu_coroutine_create(do_co_req, &srco);
 711        if (bs) {
 712            bdrv_coroutine_enter(bs, co);
 713            BDRV_POLL_WHILE(bs, !srco.finished);
 714        } else {
 715            qemu_coroutine_enter(co);
 716            while (!srco.finished) {
 717                aio_poll(qemu_get_aio_context(), true);
 718            }
 719        }
 720    }
 721
 722    return srco.ret;
 723}
 724
 725static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
 726                                         struct iovec *iov, int niov,
 727                                         enum AIOCBState aiocb_type);
 728static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req);
 729static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag);
 730static int get_sheep_fd(BDRVSheepdogState *s, Error **errp);
 731static void co_write_request(void *opaque);
 732
 733static coroutine_fn void reconnect_to_sdog(void *opaque)
 734{
 735    BDRVSheepdogState *s = opaque;
 736    AIOReq *aio_req, *next;
 737
 738    aio_set_fd_handler(s->aio_context, s->fd, false, NULL,
 739                       NULL, NULL, NULL);
 740    close(s->fd);
 741    s->fd = -1;
 742
 743    /* Wait for outstanding write requests to be completed. */
 744    while (s->co_send != NULL) {
 745        co_write_request(opaque);
 746    }
 747
 748    /* Try to reconnect the sheepdog server every one second. */
 749    while (s->fd < 0) {
 750        Error *local_err = NULL;
 751        s->fd = get_sheep_fd(s, &local_err);
 752        if (s->fd < 0) {
 753            DPRINTF("Wait for connection to be established\n");
 754            error_report_err(local_err);
 755            qemu_co_sleep_ns(QEMU_CLOCK_REALTIME, 1000000000ULL);
 756        }
 757    };
 758
 759    /*
 760     * Now we have to resend all the request in the inflight queue.  However,
 761     * resend_aioreq() can yield and newly created requests can be added to the
 762     * inflight queue before the coroutine is resumed.  To avoid mixing them, we
 763     * have to move all the inflight requests to the failed queue before
 764     * resend_aioreq() is called.
 765     */
 766    qemu_co_mutex_lock(&s->queue_lock);
 767    QLIST_FOREACH_SAFE(aio_req, &s->inflight_aio_head, aio_siblings, next) {
 768        QLIST_REMOVE(aio_req, aio_siblings);
 769        QLIST_INSERT_HEAD(&s->failed_aio_head, aio_req, aio_siblings);
 770    }
 771
 772    /* Resend all the failed aio requests. */
 773    while (!QLIST_EMPTY(&s->failed_aio_head)) {
 774        aio_req = QLIST_FIRST(&s->failed_aio_head);
 775        QLIST_REMOVE(aio_req, aio_siblings);
 776        qemu_co_mutex_unlock(&s->queue_lock);
 777        resend_aioreq(s, aio_req);
 778        qemu_co_mutex_lock(&s->queue_lock);
 779    }
 780    qemu_co_mutex_unlock(&s->queue_lock);
 781}
 782
 783/*
 784 * Receive responses of the I/O requests.
 785 *
 786 * This function is registered as a fd handler, and called from the
 787 * main loop when s->fd is ready for reading responses.
 788 */
 789static void coroutine_fn aio_read_response(void *opaque)
 790{
 791    SheepdogObjRsp rsp;
 792    BDRVSheepdogState *s = opaque;
 793    int fd = s->fd;
 794    int ret;
 795    AIOReq *aio_req = NULL;
 796    SheepdogAIOCB *acb;
 797    uint64_t idx;
 798
 799    /* read a header */
 800    ret = qemu_co_recv(fd, &rsp, sizeof(rsp));
 801    if (ret != sizeof(rsp)) {
 802        error_report("failed to get the header, %s", strerror(errno));
 803        goto err;
 804    }
 805
 806    /* find the right aio_req from the inflight aio list */
 807    QLIST_FOREACH(aio_req, &s->inflight_aio_head, aio_siblings) {
 808        if (aio_req->id == rsp.id) {
 809            break;
 810        }
 811    }
 812    if (!aio_req) {
 813        error_report("cannot find aio_req %x", rsp.id);
 814        goto err;
 815    }
 816
 817    acb = aio_req->aiocb;
 818
 819    switch (acb->aiocb_type) {
 820    case AIOCB_WRITE_UDATA:
 821        if (!is_data_obj(aio_req->oid)) {
 822            break;
 823        }
 824        idx = data_oid_to_idx(aio_req->oid);
 825
 826        if (aio_req->create) {
 827            /*
 828             * If the object is newly created one, we need to update
 829             * the vdi object (metadata object).  min_dirty_data_idx
 830             * and max_dirty_data_idx are changed to include updated
 831             * index between them.
 832             */
 833            if (rsp.result == SD_RES_SUCCESS) {
 834                s->inode.data_vdi_id[idx] = s->inode.vdi_id;
 835                acb->max_dirty_data_idx = MAX(idx, acb->max_dirty_data_idx);
 836                acb->min_dirty_data_idx = MIN(idx, acb->min_dirty_data_idx);
 837            }
 838        }
 839        break;
 840    case AIOCB_READ_UDATA:
 841        ret = qemu_co_recvv(fd, acb->qiov->iov, acb->qiov->niov,
 842                            aio_req->iov_offset, rsp.data_length);
 843        if (ret != rsp.data_length) {
 844            error_report("failed to get the data, %s", strerror(errno));
 845            goto err;
 846        }
 847        break;
 848    case AIOCB_FLUSH_CACHE:
 849        if (rsp.result == SD_RES_INVALID_PARMS) {
 850            DPRINTF("disable cache since the server doesn't support it\n");
 851            s->cache_flags = SD_FLAG_CMD_DIRECT;
 852            rsp.result = SD_RES_SUCCESS;
 853        }
 854        break;
 855    case AIOCB_DISCARD_OBJ:
 856        switch (rsp.result) {
 857        case SD_RES_INVALID_PARMS:
 858            error_report("server doesn't support discard command");
 859            rsp.result = SD_RES_SUCCESS;
 860            s->discard_supported = false;
 861            break;
 862        default:
 863            break;
 864        }
 865    }
 866
 867    /* No more data for this aio_req (reload_inode below uses its own file
 868     * descriptor handler which doesn't use co_recv).
 869    */
 870    s->co_recv = NULL;
 871
 872    qemu_co_mutex_lock(&s->queue_lock);
 873    QLIST_REMOVE(aio_req, aio_siblings);
 874    qemu_co_mutex_unlock(&s->queue_lock);
 875
 876    switch (rsp.result) {
 877    case SD_RES_SUCCESS:
 878        break;
 879    case SD_RES_READONLY:
 880        if (s->inode.vdi_id == oid_to_vid(aio_req->oid)) {
 881            ret = reload_inode(s, 0, "");
 882            if (ret < 0) {
 883                goto err;
 884            }
 885        }
 886        if (is_data_obj(aio_req->oid)) {
 887            aio_req->oid = vid_to_data_oid(s->inode.vdi_id,
 888                                           data_oid_to_idx(aio_req->oid));
 889        } else {
 890            aio_req->oid = vid_to_vdi_oid(s->inode.vdi_id);
 891        }
 892        resend_aioreq(s, aio_req);
 893        return;
 894    default:
 895        acb->ret = -EIO;
 896        error_report("%s", sd_strerror(rsp.result));
 897        break;
 898    }
 899
 900    g_free(aio_req);
 901
 902    if (!--acb->nr_pending) {
 903        /*
 904         * We've finished all requests which belong to the AIOCB, so
 905         * we can switch back to sd_co_readv/writev now.
 906         */
 907        aio_co_wake(acb->coroutine);
 908    }
 909
 910    return;
 911
 912err:
 913    reconnect_to_sdog(opaque);
 914}
 915
 916static void co_read_response(void *opaque)
 917{
 918    BDRVSheepdogState *s = opaque;
 919
 920    if (!s->co_recv) {
 921        s->co_recv = qemu_coroutine_create(aio_read_response, opaque);
 922    }
 923
 924    aio_co_enter(s->aio_context, s->co_recv);
 925}
 926
 927static void co_write_request(void *opaque)
 928{
 929    BDRVSheepdogState *s = opaque;
 930
 931    aio_co_wake(s->co_send);
 932}
 933
 934/*
 935 * Return a socket descriptor to read/write objects.
 936 *
 937 * We cannot use this descriptor for other operations because
 938 * the block driver may be on waiting response from the server.
 939 */
 940static int get_sheep_fd(BDRVSheepdogState *s, Error **errp)
 941{
 942    int fd;
 943
 944    fd = connect_to_sdog(s, errp);
 945    if (fd < 0) {
 946        return fd;
 947    }
 948
 949    aio_set_fd_handler(s->aio_context, fd, false,
 950                       co_read_response, NULL, NULL, s);
 951    return fd;
 952}
 953
 954/*
 955 * Parse numeric snapshot ID in @str
 956 * If @str can't be parsed as number, return false.
 957 * Else, if the number is zero or too large, set *@snapid to zero and
 958 * return true.
 959 * Else, set *@snapid to the number and return true.
 960 */
 961static bool sd_parse_snapid(const char *str, uint32_t *snapid)
 962{
 963    unsigned long ul;
 964    int ret;
 965
 966    ret = qemu_strtoul(str, NULL, 10, &ul);
 967    if (ret == -ERANGE) {
 968        ul = ret = 0;
 969    }
 970    if (ret) {
 971        return false;
 972    }
 973    if (ul > UINT32_MAX) {
 974        ul = 0;
 975    }
 976
 977    *snapid = ul;
 978    return true;
 979}
 980
 981static bool sd_parse_snapid_or_tag(const char *str,
 982                                   uint32_t *snapid, char tag[])
 983{
 984    if (!sd_parse_snapid(str, snapid)) {
 985        *snapid = 0;
 986        if (g_strlcpy(tag, str, SD_MAX_VDI_TAG_LEN) >= SD_MAX_VDI_TAG_LEN) {
 987            return false;
 988        }
 989    } else if (!*snapid) {
 990        return false;
 991    } else {
 992        tag[0] = 0;
 993    }
 994    return true;
 995}
 996
 997typedef struct {
 998    const char *path;           /* non-null iff transport is tcp */
 999    const char *host;           /* valid when transport is tcp */
1000    int port;                   /* valid when transport is tcp */
1001    char vdi[SD_MAX_VDI_LEN];
1002    char tag[SD_MAX_VDI_TAG_LEN];
1003    uint32_t snap_id;
1004    /* Remainder is only for sd_config_done() */
1005    URI *uri;
1006    QueryParams *qp;
1007} SheepdogConfig;
1008
1009static void sd_config_done(SheepdogConfig *cfg)
1010{
1011    if (cfg->qp) {
1012        query_params_free(cfg->qp);
1013    }
1014    uri_free(cfg->uri);
1015}
1016
1017static void sd_parse_uri(SheepdogConfig *cfg, const char *filename,
1018                         Error **errp)
1019{
1020    Error *err = NULL;
1021    QueryParams *qp = NULL;
1022    bool is_unix;
1023    URI *uri;
1024
1025    memset(cfg, 0, sizeof(*cfg));
1026
1027    cfg->uri = uri = uri_parse(filename);
1028    if (!uri) {
1029        error_setg(&err, "invalid URI '%s'", filename);
1030        goto out;
1031    }
1032
1033    /* transport */
1034    if (!g_strcmp0(uri->scheme, "sheepdog")) {
1035        is_unix = false;
1036    } else if (!g_strcmp0(uri->scheme, "sheepdog+tcp")) {
1037        is_unix = false;
1038    } else if (!g_strcmp0(uri->scheme, "sheepdog+unix")) {
1039        is_unix = true;
1040    } else {
1041        error_setg(&err, "URI scheme must be 'sheepdog', 'sheepdog+tcp',"
1042                   " or 'sheepdog+unix'");
1043        goto out;
1044    }
1045
1046    if (uri->path == NULL || !strcmp(uri->path, "/")) {
1047        error_setg(&err, "missing file path in URI");
1048        goto out;
1049    }
1050    if (g_strlcpy(cfg->vdi, uri->path + 1, SD_MAX_VDI_LEN)
1051        >= SD_MAX_VDI_LEN) {
1052        error_setg(&err, "VDI name is too long");
1053        goto out;
1054    }
1055
1056    cfg->qp = qp = query_params_parse(uri->query);
1057
1058    if (is_unix) {
1059        /* sheepdog+unix:///vdiname?socket=path */
1060        if (uri->server || uri->port) {
1061            error_setg(&err, "URI scheme %s doesn't accept a server address",
1062                       uri->scheme);
1063            goto out;
1064        }
1065        if (!qp->n) {
1066            error_setg(&err,
1067                       "URI scheme %s requires query parameter 'socket'",
1068                       uri->scheme);
1069            goto out;
1070        }
1071        if (qp->n != 1 || strcmp(qp->p[0].name, "socket")) {
1072            error_setg(&err, "unexpected query parameters");
1073            goto out;
1074        }
1075        cfg->path = qp->p[0].value;
1076    } else {
1077        /* sheepdog[+tcp]://[host:port]/vdiname */
1078        if (qp->n) {
1079            error_setg(&err, "unexpected query parameters");
1080            goto out;
1081        }
1082        cfg->host = uri->server;
1083        cfg->port = uri->port;
1084    }
1085
1086    /* snapshot tag */
1087    if (uri->fragment) {
1088        if (!sd_parse_snapid_or_tag(uri->fragment,
1089                                    &cfg->snap_id, cfg->tag)) {
1090            error_setg(&err, "'%s' is not a valid snapshot ID",
1091                       uri->fragment);
1092            goto out;
1093        }
1094    } else {
1095        cfg->snap_id = CURRENT_VDI_ID; /* search current vdi */
1096    }
1097
1098out:
1099    if (err) {
1100        error_propagate(errp, err);
1101        sd_config_done(cfg);
1102    }
1103}
1104
1105/*
1106 * Parse a filename (old syntax)
1107 *
1108 * filename must be one of the following formats:
1109 *   1. [vdiname]
1110 *   2. [vdiname]:[snapid]
1111 *   3. [vdiname]:[tag]
1112 *   4. [hostname]:[port]:[vdiname]
1113 *   5. [hostname]:[port]:[vdiname]:[snapid]
1114 *   6. [hostname]:[port]:[vdiname]:[tag]
1115 *
1116 * You can boot from the snapshot images by specifying `snapid` or
1117 * `tag'.
1118 *
1119 * You can run VMs outside the Sheepdog cluster by specifying
1120 * `hostname' and `port' (experimental).
1121 */
1122static void parse_vdiname(SheepdogConfig *cfg, const char *filename,
1123                          Error **errp)
1124{
1125    Error *err = NULL;
1126    char *p, *q, *uri;
1127    const char *host_spec, *vdi_spec;
1128    int nr_sep;
1129
1130    strstart(filename, "sheepdog:", &filename);
1131    p = q = g_strdup(filename);
1132
1133    /* count the number of separators */
1134    nr_sep = 0;
1135    while (*p) {
1136        if (*p == ':') {
1137            nr_sep++;
1138        }
1139        p++;
1140    }
1141    p = q;
1142
1143    /* use the first two tokens as host_spec. */
1144    if (nr_sep >= 2) {
1145        host_spec = p;
1146        p = strchr(p, ':');
1147        p++;
1148        p = strchr(p, ':');
1149        *p++ = '\0';
1150    } else {
1151        host_spec = "";
1152    }
1153
1154    vdi_spec = p;
1155
1156    p = strchr(vdi_spec, ':');
1157    if (p) {
1158        *p++ = '#';
1159    }
1160
1161    uri = g_strdup_printf("sheepdog://%s/%s", host_spec, vdi_spec);
1162
1163    /*
1164     * FIXME We to escape URI meta-characters, e.g. "x?y=z"
1165     * produces "sheepdog://x?y=z".  Because of that ...
1166     */
1167    sd_parse_uri(cfg, uri, &err);
1168    if (err) {
1169        /*
1170         * ... this can fail, but the error message is misleading.
1171         * Replace it by the traditional useless one until the
1172         * escaping is fixed.
1173         */
1174        error_free(err);
1175        error_setg(errp, "Can't parse filename");
1176    }
1177
1178    g_free(q);
1179    g_free(uri);
1180}
1181
1182static void sd_parse_filename(const char *filename, QDict *options,
1183                              Error **errp)
1184{
1185    Error *err = NULL;
1186    SheepdogConfig cfg;
1187    char buf[32];
1188
1189    if (strstr(filename, "://")) {
1190        sd_parse_uri(&cfg, filename, &err);
1191    } else {
1192        parse_vdiname(&cfg, filename, &err);
1193    }
1194    if (err) {
1195        error_propagate(errp, err);
1196        return;
1197    }
1198
1199    if (cfg.path) {
1200        qdict_set_default_str(options, "server.path", cfg.path);
1201        qdict_set_default_str(options, "server.type", "unix");
1202    } else {
1203        qdict_set_default_str(options, "server.type", "inet");
1204        qdict_set_default_str(options, "server.host",
1205                              cfg.host ?: SD_DEFAULT_ADDR);
1206        snprintf(buf, sizeof(buf), "%d", cfg.port ?: SD_DEFAULT_PORT);
1207        qdict_set_default_str(options, "server.port", buf);
1208    }
1209    qdict_set_default_str(options, "vdi", cfg.vdi);
1210    qdict_set_default_str(options, "tag", cfg.tag);
1211    if (cfg.snap_id) {
1212        snprintf(buf, sizeof(buf), "%d", cfg.snap_id);
1213        qdict_set_default_str(options, "snap-id", buf);
1214    }
1215
1216    sd_config_done(&cfg);
1217}
1218
1219static int find_vdi_name(BDRVSheepdogState *s, const char *filename,
1220                         uint32_t snapid, const char *tag, uint32_t *vid,
1221                         bool lock, Error **errp)
1222{
1223    int ret, fd;
1224    SheepdogVdiReq hdr;
1225    SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1226    unsigned int wlen, rlen = 0;
1227    char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
1228
1229    fd = connect_to_sdog(s, errp);
1230    if (fd < 0) {
1231        return fd;
1232    }
1233
1234    /* This pair of strncpy calls ensures that the buffer is zero-filled,
1235     * which is desirable since we'll soon be sending those bytes, and
1236     * don't want the send_req to read uninitialized data.
1237     */
1238    strncpy(buf, filename, SD_MAX_VDI_LEN);
1239    strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN);
1240
1241    memset(&hdr, 0, sizeof(hdr));
1242    if (lock) {
1243        hdr.opcode = SD_OP_LOCK_VDI;
1244        hdr.type = LOCK_TYPE_NORMAL;
1245    } else {
1246        hdr.opcode = SD_OP_GET_VDI_INFO;
1247    }
1248    wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN;
1249    hdr.proto_ver = SD_PROTO_VER;
1250    hdr.data_length = wlen;
1251    hdr.snapid = snapid;
1252    hdr.flags = SD_FLAG_CMD_WRITE;
1253
1254    ret = do_req(fd, s->bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1255    if (ret) {
1256        error_setg_errno(errp, -ret, "cannot get vdi info");
1257        goto out;
1258    }
1259
1260    if (rsp->result != SD_RES_SUCCESS) {
1261        error_setg(errp, "cannot get vdi info, %s, %s %" PRIu32 " %s",
1262                   sd_strerror(rsp->result), filename, snapid, tag);
1263        if (rsp->result == SD_RES_NO_VDI) {
1264            ret = -ENOENT;
1265        } else if (rsp->result == SD_RES_VDI_LOCKED) {
1266            ret = -EBUSY;
1267        } else {
1268            ret = -EIO;
1269        }
1270        goto out;
1271    }
1272    *vid = rsp->vdi_id;
1273
1274    ret = 0;
1275out:
1276    closesocket(fd);
1277    return ret;
1278}
1279
1280static void coroutine_fn add_aio_request(BDRVSheepdogState *s, AIOReq *aio_req,
1281                                         struct iovec *iov, int niov,
1282                                         enum AIOCBState aiocb_type)
1283{
1284    int nr_copies = s->inode.nr_copies;
1285    SheepdogObjReq hdr;
1286    unsigned int wlen = 0;
1287    int ret;
1288    uint64_t oid = aio_req->oid;
1289    unsigned int datalen = aio_req->data_len;
1290    uint64_t offset = aio_req->offset;
1291    uint8_t flags = aio_req->flags;
1292    uint64_t old_oid = aio_req->base_oid;
1293    bool create = aio_req->create;
1294
1295    qemu_co_mutex_lock(&s->queue_lock);
1296    QLIST_INSERT_HEAD(&s->inflight_aio_head, aio_req, aio_siblings);
1297    qemu_co_mutex_unlock(&s->queue_lock);
1298
1299    if (!nr_copies) {
1300        error_report("bug");
1301    }
1302
1303    memset(&hdr, 0, sizeof(hdr));
1304
1305    switch (aiocb_type) {
1306    case AIOCB_FLUSH_CACHE:
1307        hdr.opcode = SD_OP_FLUSH_VDI;
1308        break;
1309    case AIOCB_READ_UDATA:
1310        hdr.opcode = SD_OP_READ_OBJ;
1311        hdr.flags = flags;
1312        break;
1313    case AIOCB_WRITE_UDATA:
1314        if (create) {
1315            hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1316        } else {
1317            hdr.opcode = SD_OP_WRITE_OBJ;
1318        }
1319        wlen = datalen;
1320        hdr.flags = SD_FLAG_CMD_WRITE | flags;
1321        break;
1322    case AIOCB_DISCARD_OBJ:
1323        hdr.opcode = SD_OP_WRITE_OBJ;
1324        hdr.flags = SD_FLAG_CMD_WRITE | flags;
1325        s->inode.data_vdi_id[data_oid_to_idx(oid)] = 0;
1326        offset = offsetof(SheepdogInode,
1327                          data_vdi_id[data_oid_to_idx(oid)]);
1328        oid = vid_to_vdi_oid(s->inode.vdi_id);
1329        wlen = datalen = sizeof(uint32_t);
1330        break;
1331    }
1332
1333    if (s->cache_flags) {
1334        hdr.flags |= s->cache_flags;
1335    }
1336
1337    hdr.oid = oid;
1338    hdr.cow_oid = old_oid;
1339    hdr.copies = s->inode.nr_copies;
1340
1341    hdr.data_length = datalen;
1342    hdr.offset = offset;
1343
1344    hdr.id = aio_req->id;
1345
1346    qemu_co_mutex_lock(&s->lock);
1347    s->co_send = qemu_coroutine_self();
1348    aio_set_fd_handler(s->aio_context, s->fd, false,
1349                       co_read_response, co_write_request, NULL, s);
1350    socket_set_cork(s->fd, 1);
1351
1352    /* send a header */
1353    ret = qemu_co_send(s->fd, &hdr, sizeof(hdr));
1354    if (ret != sizeof(hdr)) {
1355        error_report("failed to send a req, %s", strerror(errno));
1356        goto out;
1357    }
1358
1359    if (wlen) {
1360        ret = qemu_co_sendv(s->fd, iov, niov, aio_req->iov_offset, wlen);
1361        if (ret != wlen) {
1362            error_report("failed to send a data, %s", strerror(errno));
1363        }
1364    }
1365out:
1366    socket_set_cork(s->fd, 0);
1367    aio_set_fd_handler(s->aio_context, s->fd, false,
1368                       co_read_response, NULL, NULL, s);
1369    s->co_send = NULL;
1370    qemu_co_mutex_unlock(&s->lock);
1371}
1372
1373static int read_write_object(int fd, BlockDriverState *bs, char *buf,
1374                             uint64_t oid, uint8_t copies,
1375                             unsigned int datalen, uint64_t offset,
1376                             bool write, bool create, uint32_t cache_flags)
1377{
1378    SheepdogObjReq hdr;
1379    SheepdogObjRsp *rsp = (SheepdogObjRsp *)&hdr;
1380    unsigned int wlen, rlen;
1381    int ret;
1382
1383    memset(&hdr, 0, sizeof(hdr));
1384
1385    if (write) {
1386        wlen = datalen;
1387        rlen = 0;
1388        hdr.flags = SD_FLAG_CMD_WRITE;
1389        if (create) {
1390            hdr.opcode = SD_OP_CREATE_AND_WRITE_OBJ;
1391        } else {
1392            hdr.opcode = SD_OP_WRITE_OBJ;
1393        }
1394    } else {
1395        wlen = 0;
1396        rlen = datalen;
1397        hdr.opcode = SD_OP_READ_OBJ;
1398    }
1399
1400    hdr.flags |= cache_flags;
1401
1402    hdr.oid = oid;
1403    hdr.data_length = datalen;
1404    hdr.offset = offset;
1405    hdr.copies = copies;
1406
1407    ret = do_req(fd, bs, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1408    if (ret) {
1409        error_report("failed to send a request to the sheep");
1410        return ret;
1411    }
1412
1413    switch (rsp->result) {
1414    case SD_RES_SUCCESS:
1415        return 0;
1416    default:
1417        error_report("%s", sd_strerror(rsp->result));
1418        return -EIO;
1419    }
1420}
1421
1422static int read_object(int fd, BlockDriverState *bs, char *buf,
1423                       uint64_t oid, uint8_t copies,
1424                       unsigned int datalen, uint64_t offset,
1425                       uint32_t cache_flags)
1426{
1427    return read_write_object(fd, bs, buf, oid, copies,
1428                             datalen, offset, false,
1429                             false, cache_flags);
1430}
1431
1432static int write_object(int fd, BlockDriverState *bs, char *buf,
1433                        uint64_t oid, uint8_t copies,
1434                        unsigned int datalen, uint64_t offset, bool create,
1435                        uint32_t cache_flags)
1436{
1437    return read_write_object(fd, bs, buf, oid, copies,
1438                             datalen, offset, true,
1439                             create, cache_flags);
1440}
1441
1442/* update inode with the latest state */
1443static int reload_inode(BDRVSheepdogState *s, uint32_t snapid, const char *tag)
1444{
1445    Error *local_err = NULL;
1446    SheepdogInode *inode;
1447    int ret = 0, fd;
1448    uint32_t vid = 0;
1449
1450    fd = connect_to_sdog(s, &local_err);
1451    if (fd < 0) {
1452        error_report_err(local_err);
1453        return -EIO;
1454    }
1455
1456    inode = g_malloc(SD_INODE_HEADER_SIZE);
1457
1458    ret = find_vdi_name(s, s->name, snapid, tag, &vid, false, &local_err);
1459    if (ret) {
1460        error_report_err(local_err);
1461        goto out;
1462    }
1463
1464    ret = read_object(fd, s->bs, (char *)inode, vid_to_vdi_oid(vid),
1465                      s->inode.nr_copies, SD_INODE_HEADER_SIZE, 0,
1466                      s->cache_flags);
1467    if (ret < 0) {
1468        goto out;
1469    }
1470
1471    if (inode->vdi_id != s->inode.vdi_id) {
1472        memcpy(&s->inode, inode, SD_INODE_HEADER_SIZE);
1473    }
1474
1475out:
1476    g_free(inode);
1477    closesocket(fd);
1478
1479    return ret;
1480}
1481
1482static void coroutine_fn resend_aioreq(BDRVSheepdogState *s, AIOReq *aio_req)
1483{
1484    SheepdogAIOCB *acb = aio_req->aiocb;
1485
1486    aio_req->create = false;
1487
1488    /* check whether this request becomes a CoW one */
1489    if (acb->aiocb_type == AIOCB_WRITE_UDATA && is_data_obj(aio_req->oid)) {
1490        int idx = data_oid_to_idx(aio_req->oid);
1491
1492        if (is_data_obj_writable(&s->inode, idx)) {
1493            goto out;
1494        }
1495
1496        if (s->inode.data_vdi_id[idx]) {
1497            aio_req->base_oid = vid_to_data_oid(s->inode.data_vdi_id[idx], idx);
1498            aio_req->flags |= SD_FLAG_CMD_COW;
1499        }
1500        aio_req->create = true;
1501    }
1502out:
1503    if (is_data_obj(aio_req->oid)) {
1504        add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
1505                        acb->aiocb_type);
1506    } else {
1507        struct iovec iov;
1508        iov.iov_base = &s->inode;
1509        iov.iov_len = sizeof(s->inode);
1510        add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
1511    }
1512}
1513
1514static void sd_detach_aio_context(BlockDriverState *bs)
1515{
1516    BDRVSheepdogState *s = bs->opaque;
1517
1518    aio_set_fd_handler(s->aio_context, s->fd, false, NULL,
1519                       NULL, NULL, NULL);
1520}
1521
1522static void sd_attach_aio_context(BlockDriverState *bs,
1523                                  AioContext *new_context)
1524{
1525    BDRVSheepdogState *s = bs->opaque;
1526
1527    s->aio_context = new_context;
1528    aio_set_fd_handler(new_context, s->fd, false,
1529                       co_read_response, NULL, NULL, s);
1530}
1531
1532static QemuOptsList runtime_opts = {
1533    .name = "sheepdog",
1534    .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1535    .desc = {
1536        {
1537            .name = "vdi",
1538            .type = QEMU_OPT_STRING,
1539        },
1540        {
1541            .name = "snap-id",
1542            .type = QEMU_OPT_NUMBER,
1543        },
1544        {
1545            .name = "tag",
1546            .type = QEMU_OPT_STRING,
1547        },
1548        { /* end of list */ }
1549    },
1550};
1551
1552static int sd_open(BlockDriverState *bs, QDict *options, int flags,
1553                   Error **errp)
1554{
1555    int ret, fd;
1556    uint32_t vid = 0;
1557    BDRVSheepdogState *s = bs->opaque;
1558    const char *vdi, *snap_id_str, *tag;
1559    uint64_t snap_id;
1560    char *buf = NULL;
1561    QemuOpts *opts;
1562    Error *local_err = NULL;
1563
1564    s->bs = bs;
1565    s->aio_context = bdrv_get_aio_context(bs);
1566
1567    opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1568    qemu_opts_absorb_qdict(opts, options, &local_err);
1569    if (local_err) {
1570        error_propagate(errp, local_err);
1571        ret = -EINVAL;
1572        goto err_no_fd;
1573    }
1574
1575    s->addr = sd_server_config(options, errp);
1576    if (!s->addr) {
1577        ret = -EINVAL;
1578        goto err_no_fd;
1579    }
1580
1581    vdi = qemu_opt_get(opts, "vdi");
1582    snap_id_str = qemu_opt_get(opts, "snap-id");
1583    snap_id = qemu_opt_get_number(opts, "snap-id", CURRENT_VDI_ID);
1584    tag = qemu_opt_get(opts, "tag");
1585
1586    if (!vdi) {
1587        error_setg(errp, "parameter 'vdi' is missing");
1588        ret = -EINVAL;
1589        goto err_no_fd;
1590    }
1591    if (strlen(vdi) >= SD_MAX_VDI_LEN) {
1592        error_setg(errp, "value of parameter 'vdi' is too long");
1593        ret = -EINVAL;
1594        goto err_no_fd;
1595    }
1596
1597    if (snap_id > UINT32_MAX) {
1598        snap_id = 0;
1599    }
1600    if (snap_id_str && !snap_id) {
1601        error_setg(errp, "'snap-id=%s' is not a valid snapshot ID",
1602                   snap_id_str);
1603        ret = -EINVAL;
1604        goto err_no_fd;
1605    }
1606
1607    if (!tag) {
1608        tag = "";
1609    }
1610    if (strlen(tag) >= SD_MAX_VDI_TAG_LEN) {
1611        error_setg(errp, "value of parameter 'tag' is too long");
1612        ret = -EINVAL;
1613        goto err_no_fd;
1614    }
1615
1616    QLIST_INIT(&s->inflight_aio_head);
1617    QLIST_INIT(&s->failed_aio_head);
1618    QLIST_INIT(&s->inflight_aiocb_head);
1619
1620    s->fd = get_sheep_fd(s, errp);
1621    if (s->fd < 0) {
1622        ret = s->fd;
1623        goto err_no_fd;
1624    }
1625
1626    ret = find_vdi_name(s, vdi, (uint32_t)snap_id, tag, &vid, true, errp);
1627    if (ret) {
1628        goto err;
1629    }
1630
1631    /*
1632     * QEMU block layer emulates writethrough cache as 'writeback + flush', so
1633     * we always set SD_FLAG_CMD_CACHE (writeback cache) as default.
1634     */
1635    s->cache_flags = SD_FLAG_CMD_CACHE;
1636    if (flags & BDRV_O_NOCACHE) {
1637        s->cache_flags = SD_FLAG_CMD_DIRECT;
1638    }
1639    s->discard_supported = true;
1640
1641    if (snap_id || tag[0]) {
1642        DPRINTF("%" PRIx32 " snapshot inode was open.\n", vid);
1643        s->is_snapshot = true;
1644    }
1645
1646    fd = connect_to_sdog(s, errp);
1647    if (fd < 0) {
1648        ret = fd;
1649        goto err;
1650    }
1651
1652    buf = g_malloc(SD_INODE_SIZE);
1653    ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid),
1654                      0, SD_INODE_SIZE, 0, s->cache_flags);
1655
1656    closesocket(fd);
1657
1658    if (ret) {
1659        error_setg(errp, "Can't read snapshot inode");
1660        goto err;
1661    }
1662
1663    memcpy(&s->inode, buf, sizeof(s->inode));
1664
1665    bs->total_sectors = s->inode.vdi_size / BDRV_SECTOR_SIZE;
1666    pstrcpy(s->name, sizeof(s->name), vdi);
1667    qemu_co_mutex_init(&s->lock);
1668    qemu_co_mutex_init(&s->queue_lock);
1669    qemu_co_queue_init(&s->overlapping_queue);
1670    qemu_opts_del(opts);
1671    g_free(buf);
1672    return 0;
1673
1674err:
1675    aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd,
1676                       false, NULL, NULL, NULL, NULL);
1677    closesocket(s->fd);
1678err_no_fd:
1679    qemu_opts_del(opts);
1680    g_free(buf);
1681    return ret;
1682}
1683
1684static int sd_reopen_prepare(BDRVReopenState *state, BlockReopenQueue *queue,
1685                             Error **errp)
1686{
1687    BDRVSheepdogState *s = state->bs->opaque;
1688    BDRVSheepdogReopenState *re_s;
1689    int ret = 0;
1690
1691    re_s = state->opaque = g_new0(BDRVSheepdogReopenState, 1);
1692
1693    re_s->cache_flags = SD_FLAG_CMD_CACHE;
1694    if (state->flags & BDRV_O_NOCACHE) {
1695        re_s->cache_flags = SD_FLAG_CMD_DIRECT;
1696    }
1697
1698    re_s->fd = get_sheep_fd(s, errp);
1699    if (re_s->fd < 0) {
1700        ret = re_s->fd;
1701        return ret;
1702    }
1703
1704    return ret;
1705}
1706
1707static void sd_reopen_commit(BDRVReopenState *state)
1708{
1709    BDRVSheepdogReopenState *re_s = state->opaque;
1710    BDRVSheepdogState *s = state->bs->opaque;
1711
1712    if (s->fd) {
1713        aio_set_fd_handler(s->aio_context, s->fd, false,
1714                           NULL, NULL, NULL, NULL);
1715        closesocket(s->fd);
1716    }
1717
1718    s->fd = re_s->fd;
1719    s->cache_flags = re_s->cache_flags;
1720
1721    g_free(state->opaque);
1722    state->opaque = NULL;
1723
1724    return;
1725}
1726
1727static void sd_reopen_abort(BDRVReopenState *state)
1728{
1729    BDRVSheepdogReopenState *re_s = state->opaque;
1730    BDRVSheepdogState *s = state->bs->opaque;
1731
1732    if (re_s == NULL) {
1733        return;
1734    }
1735
1736    if (re_s->fd) {
1737        aio_set_fd_handler(s->aio_context, re_s->fd, false,
1738                           NULL, NULL, NULL, NULL);
1739        closesocket(re_s->fd);
1740    }
1741
1742    g_free(state->opaque);
1743    state->opaque = NULL;
1744
1745    return;
1746}
1747
1748static int do_sd_create(BDRVSheepdogState *s, uint32_t *vdi_id, int snapshot,
1749                        Error **errp)
1750{
1751    SheepdogVdiReq hdr;
1752    SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
1753    int fd, ret;
1754    unsigned int wlen, rlen = 0;
1755    char buf[SD_MAX_VDI_LEN];
1756
1757    fd = connect_to_sdog(s, errp);
1758    if (fd < 0) {
1759        return fd;
1760    }
1761
1762    /* FIXME: would it be better to fail (e.g., return -EIO) when filename
1763     * does not fit in buf?  For now, just truncate and avoid buffer overrun.
1764     */
1765    memset(buf, 0, sizeof(buf));
1766    pstrcpy(buf, sizeof(buf), s->name);
1767
1768    memset(&hdr, 0, sizeof(hdr));
1769    hdr.opcode = SD_OP_NEW_VDI;
1770    hdr.base_vdi_id = s->inode.vdi_id;
1771
1772    wlen = SD_MAX_VDI_LEN;
1773
1774    hdr.flags = SD_FLAG_CMD_WRITE;
1775    hdr.snapid = snapshot;
1776
1777    hdr.data_length = wlen;
1778    hdr.vdi_size = s->inode.vdi_size;
1779    hdr.copy_policy = s->inode.copy_policy;
1780    hdr.copies = s->inode.nr_copies;
1781    hdr.block_size_shift = s->inode.block_size_shift;
1782
1783    ret = do_req(fd, NULL, (SheepdogReq *)&hdr, buf, &wlen, &rlen);
1784
1785    closesocket(fd);
1786
1787    if (ret) {
1788        error_setg_errno(errp, -ret, "create failed");
1789        return ret;
1790    }
1791
1792    if (rsp->result != SD_RES_SUCCESS) {
1793        error_setg(errp, "%s, %s", sd_strerror(rsp->result), s->inode.name);
1794        return -EIO;
1795    }
1796
1797    if (vdi_id) {
1798        *vdi_id = rsp->vdi_id;
1799    }
1800
1801    return 0;
1802}
1803
1804static int sd_prealloc(BlockDriverState *bs, int64_t old_size, int64_t new_size,
1805                       Error **errp)
1806{
1807    BlockBackend *blk = NULL;
1808    BDRVSheepdogState *base = bs->opaque;
1809    unsigned long buf_size;
1810    uint32_t idx, max_idx;
1811    uint32_t object_size;
1812    void *buf = NULL;
1813    int ret;
1814
1815    blk = blk_new(BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | BLK_PERM_RESIZE,
1816                  BLK_PERM_ALL);
1817
1818    ret = blk_insert_bs(blk, bs, errp);
1819    if (ret < 0) {
1820        goto out_with_err_set;
1821    }
1822
1823    blk_set_allow_write_beyond_eof(blk, true);
1824
1825    object_size = (UINT32_C(1) << base->inode.block_size_shift);
1826    buf_size = MIN(object_size, SD_DATA_OBJ_SIZE);
1827    buf = g_malloc0(buf_size);
1828
1829    max_idx = DIV_ROUND_UP(new_size, buf_size);
1830
1831    for (idx = old_size / buf_size; idx < max_idx; idx++) {
1832        /*
1833         * The created image can be a cloned image, so we need to read
1834         * a data from the source image.
1835         */
1836        ret = blk_pread(blk, idx * buf_size, buf, buf_size);
1837        if (ret < 0) {
1838            goto out;
1839        }
1840        ret = blk_pwrite(blk, idx * buf_size, buf, buf_size, 0);
1841        if (ret < 0) {
1842            goto out;
1843        }
1844    }
1845
1846    ret = 0;
1847out:
1848    if (ret < 0) {
1849        error_setg_errno(errp, -ret, "Can't pre-allocate");
1850    }
1851out_with_err_set:
1852    blk_unref(blk);
1853    g_free(buf);
1854
1855    return ret;
1856}
1857
1858static int sd_create_prealloc(BlockdevOptionsSheepdog *location, int64_t size,
1859                              Error **errp)
1860{
1861    BlockDriverState *bs;
1862    Visitor *v;
1863    QObject *obj = NULL;
1864    QDict *qdict;
1865    Error *local_err = NULL;
1866    int ret;
1867
1868    v = qobject_output_visitor_new(&obj);
1869    visit_type_BlockdevOptionsSheepdog(v, NULL, &location, &local_err);
1870    visit_free(v);
1871
1872    if (local_err) {
1873        error_propagate(errp, local_err);
1874        qobject_unref(obj);
1875        return -EINVAL;
1876    }
1877
1878    qdict = qobject_to(QDict, obj);
1879    qdict_flatten(qdict);
1880
1881    qdict_put_str(qdict, "driver", "sheepdog");
1882
1883    bs = bdrv_open(NULL, NULL, qdict, BDRV_O_PROTOCOL | BDRV_O_RDWR, errp);
1884    if (bs == NULL) {
1885        ret = -EIO;
1886        goto fail;
1887    }
1888
1889    ret = sd_prealloc(bs, 0, size, errp);
1890fail:
1891    bdrv_unref(bs);
1892    qobject_unref(qdict);
1893    return ret;
1894}
1895
1896static int parse_redundancy(BDRVSheepdogState *s, SheepdogRedundancy *opt)
1897{
1898    struct SheepdogInode *inode = &s->inode;
1899
1900    switch (opt->type) {
1901    case SHEEPDOG_REDUNDANCY_TYPE_FULL:
1902        if (opt->u.full.copies > SD_MAX_COPIES || opt->u.full.copies < 1) {
1903            return -EINVAL;
1904        }
1905        inode->copy_policy = 0;
1906        inode->nr_copies = opt->u.full.copies;
1907        return 0;
1908
1909    case SHEEPDOG_REDUNDANCY_TYPE_ERASURE_CODED:
1910    {
1911        int64_t copy = opt->u.erasure_coded.data_strips;
1912        int64_t parity = opt->u.erasure_coded.parity_strips;
1913
1914        if (copy != 2 && copy != 4 && copy != 8 && copy != 16) {
1915            return -EINVAL;
1916        }
1917
1918        if (parity >= SD_EC_MAX_STRIP || parity < 1) {
1919            return -EINVAL;
1920        }
1921
1922        /*
1923         * 4 bits for parity and 4 bits for data.
1924         * We have to compress upper data bits because it can't represent 16
1925         */
1926        inode->copy_policy = ((copy / 2) << 4) + parity;
1927        inode->nr_copies = copy + parity;
1928        return 0;
1929    }
1930
1931    default:
1932        g_assert_not_reached();
1933    }
1934
1935    return -EINVAL;
1936}
1937
1938/*
1939 * Sheepdog support two kinds of redundancy, full replication and erasure
1940 * coding.
1941 *
1942 * # create a fully replicated vdi with x copies
1943 * -o redundancy=x (1 <= x <= SD_MAX_COPIES)
1944 *
1945 * # create a erasure coded vdi with x data strips and y parity strips
1946 * -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)
1947 */
1948static SheepdogRedundancy *parse_redundancy_str(const char *opt)
1949{
1950    SheepdogRedundancy *redundancy;
1951    const char *n1, *n2;
1952    long copy, parity;
1953    char p[10];
1954    int ret;
1955
1956    pstrcpy(p, sizeof(p), opt);
1957    n1 = strtok(p, ":");
1958    n2 = strtok(NULL, ":");
1959
1960    if (!n1) {
1961        return NULL;
1962    }
1963
1964    ret = qemu_strtol(n1, NULL, 10, &copy);
1965    if (ret < 0) {
1966        return NULL;
1967    }
1968
1969    redundancy = g_new0(SheepdogRedundancy, 1);
1970    if (!n2) {
1971        *redundancy = (SheepdogRedundancy) {
1972            .type               = SHEEPDOG_REDUNDANCY_TYPE_FULL,
1973            .u.full.copies      = copy,
1974        };
1975    } else {
1976        ret = qemu_strtol(n2, NULL, 10, &parity);
1977        if (ret < 0) {
1978            g_free(redundancy);
1979            return NULL;
1980        }
1981
1982        *redundancy = (SheepdogRedundancy) {
1983            .type               = SHEEPDOG_REDUNDANCY_TYPE_ERASURE_CODED,
1984            .u.erasure_coded    = {
1985                .data_strips    = copy,
1986                .parity_strips  = parity,
1987            },
1988        };
1989    }
1990
1991    return redundancy;
1992}
1993
1994static int parse_block_size_shift(BDRVSheepdogState *s,
1995                                  BlockdevCreateOptionsSheepdog *opts)
1996{
1997    struct SheepdogInode *inode = &s->inode;
1998    uint64_t object_size;
1999    int obj_order;
2000
2001    if (opts->has_object_size) {
2002        object_size = opts->object_size;
2003
2004        if ((object_size - 1) & object_size) {    /* not a power of 2? */
2005            return -EINVAL;
2006        }
2007        obj_order = ctz32(object_size);
2008        if (obj_order < 20 || obj_order > 31) {
2009            return -EINVAL;
2010        }
2011        inode->block_size_shift = (uint8_t)obj_order;
2012    }
2013
2014    return 0;
2015}
2016
2017static int sd_co_create(BlockdevCreateOptions *options, Error **errp)
2018{
2019    BlockdevCreateOptionsSheepdog *opts = &options->u.sheepdog;
2020    int ret = 0;
2021    uint32_t vid = 0;
2022    char *backing_file = NULL;
2023    char *buf = NULL;
2024    BDRVSheepdogState *s;
2025    uint64_t max_vdi_size;
2026    bool prealloc = false;
2027
2028    assert(options->driver == BLOCKDEV_DRIVER_SHEEPDOG);
2029
2030    s = g_new0(BDRVSheepdogState, 1);
2031
2032    /* Steal SocketAddress from QAPI, set NULL to prevent double free */
2033    s->addr = opts->location->server;
2034    opts->location->server = NULL;
2035
2036    if (strlen(opts->location->vdi) >= sizeof(s->name)) {
2037        error_setg(errp, "'vdi' string too long");
2038        ret = -EINVAL;
2039        goto out;
2040    }
2041    pstrcpy(s->name, sizeof(s->name), opts->location->vdi);
2042
2043    s->inode.vdi_size = opts->size;
2044    backing_file = opts->backing_file;
2045
2046    if (!opts->has_preallocation) {
2047        opts->preallocation = PREALLOC_MODE_OFF;
2048    }
2049    switch (opts->preallocation) {
2050    case PREALLOC_MODE_OFF:
2051        prealloc = false;
2052        break;
2053    case PREALLOC_MODE_FULL:
2054        prealloc = true;
2055        break;
2056    default:
2057        error_setg(errp, "Preallocation mode not supported for Sheepdog");
2058        ret = -EINVAL;
2059        goto out;
2060    }
2061
2062    if (opts->has_redundancy) {
2063        ret = parse_redundancy(s, opts->redundancy);
2064        if (ret < 0) {
2065            error_setg(errp, "Invalid redundancy mode");
2066            goto out;
2067        }
2068    }
2069    ret = parse_block_size_shift(s, opts);
2070    if (ret < 0) {
2071        error_setg(errp, "Invalid object_size."
2072                         " obect_size needs to be power of 2"
2073                         " and be limited from 2^20 to 2^31");
2074        goto out;
2075    }
2076
2077    if (opts->has_backing_file) {
2078        BlockBackend *blk;
2079        BDRVSheepdogState *base;
2080        BlockDriver *drv;
2081
2082        /* Currently, only Sheepdog backing image is supported. */
2083        drv = bdrv_find_protocol(opts->backing_file, true, NULL);
2084        if (!drv || strcmp(drv->protocol_name, "sheepdog") != 0) {
2085            error_setg(errp, "backing_file must be a sheepdog image");
2086            ret = -EINVAL;
2087            goto out;
2088        }
2089
2090        blk = blk_new_open(opts->backing_file, NULL, NULL,
2091                           BDRV_O_PROTOCOL, errp);
2092        if (blk == NULL) {
2093            ret = -EIO;
2094            goto out;
2095        }
2096
2097        base = blk_bs(blk)->opaque;
2098
2099        if (!is_snapshot(&base->inode)) {
2100            error_setg(errp, "cannot clone from a non snapshot vdi");
2101            blk_unref(blk);
2102            ret = -EINVAL;
2103            goto out;
2104        }
2105        s->inode.vdi_id = base->inode.vdi_id;
2106        blk_unref(blk);
2107    }
2108
2109    s->aio_context = qemu_get_aio_context();
2110
2111    /* if block_size_shift is not specified, get cluster default value */
2112    if (s->inode.block_size_shift == 0) {
2113        SheepdogVdiReq hdr;
2114        SheepdogClusterRsp *rsp = (SheepdogClusterRsp *)&hdr;
2115        int fd;
2116        unsigned int wlen = 0, rlen = 0;
2117
2118        fd = connect_to_sdog(s, errp);
2119        if (fd < 0) {
2120            ret = fd;
2121            goto out;
2122        }
2123
2124        memset(&hdr, 0, sizeof(hdr));
2125        hdr.opcode = SD_OP_GET_CLUSTER_DEFAULT;
2126        hdr.proto_ver = SD_PROTO_VER;
2127
2128        ret = do_req(fd, NULL, (SheepdogReq *)&hdr,
2129                     NULL, &wlen, &rlen);
2130        closesocket(fd);
2131        if (ret) {
2132            error_setg_errno(errp, -ret, "failed to get cluster default");
2133            goto out;
2134        }
2135        if (rsp->result == SD_RES_SUCCESS) {
2136            s->inode.block_size_shift = rsp->block_size_shift;
2137        } else {
2138            s->inode.block_size_shift = SD_DEFAULT_BLOCK_SIZE_SHIFT;
2139        }
2140    }
2141
2142    max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS;
2143
2144    if (s->inode.vdi_size > max_vdi_size) {
2145        error_setg(errp, "An image is too large."
2146                         " The maximum image size is %"PRIu64 "GB",
2147                         max_vdi_size / 1024 / 1024 / 1024);
2148        ret = -EINVAL;
2149        goto out;
2150    }
2151
2152    ret = do_sd_create(s, &vid, 0, errp);
2153    if (ret) {
2154        goto out;
2155    }
2156
2157    if (prealloc) {
2158        ret = sd_create_prealloc(opts->location, opts->size, errp);
2159    }
2160out:
2161    g_free(backing_file);
2162    g_free(buf);
2163    g_free(s->addr);
2164    g_free(s);
2165    return ret;
2166}
2167
2168static int coroutine_fn sd_co_create_opts(const char *filename, QemuOpts *opts,
2169                                          Error **errp)
2170{
2171    BlockdevCreateOptions *create_options = NULL;
2172    QDict *qdict, *location_qdict;
2173    Visitor *v;
2174    char *redundancy;
2175    Error *local_err = NULL;
2176    int ret;
2177
2178    redundancy = qemu_opt_get_del(opts, BLOCK_OPT_REDUNDANCY);
2179
2180    qdict = qemu_opts_to_qdict(opts, NULL);
2181    qdict_put_str(qdict, "driver", "sheepdog");
2182
2183    location_qdict = qdict_new();
2184    qdict_put(qdict, "location", location_qdict);
2185
2186    sd_parse_filename(filename, location_qdict, &local_err);
2187    if (local_err) {
2188        error_propagate(errp, local_err);
2189        ret = -EINVAL;
2190        goto fail;
2191    }
2192
2193    qdict_flatten(qdict);
2194
2195    /* Change legacy command line options into QMP ones */
2196    static const QDictRenames opt_renames[] = {
2197        { BLOCK_OPT_BACKING_FILE,       "backing-file" },
2198        { BLOCK_OPT_OBJECT_SIZE,        "object-size" },
2199        { NULL, NULL },
2200    };
2201
2202    if (!qdict_rename_keys(qdict, opt_renames, errp)) {
2203        ret = -EINVAL;
2204        goto fail;
2205    }
2206
2207    /* Get the QAPI object */
2208    v = qobject_input_visitor_new_flat_confused(qdict, errp);
2209    if (!v) {
2210        ret = -EINVAL;
2211        goto fail;
2212    }
2213
2214    visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
2215    visit_free(v);
2216
2217    if (local_err) {
2218        error_propagate(errp, local_err);
2219        ret = -EINVAL;
2220        goto fail;
2221    }
2222
2223    assert(create_options->driver == BLOCKDEV_DRIVER_SHEEPDOG);
2224    create_options->u.sheepdog.size =
2225        ROUND_UP(create_options->u.sheepdog.size, BDRV_SECTOR_SIZE);
2226
2227    if (redundancy) {
2228        create_options->u.sheepdog.has_redundancy = true;
2229        create_options->u.sheepdog.redundancy =
2230            parse_redundancy_str(redundancy);
2231        if (create_options->u.sheepdog.redundancy == NULL) {
2232            error_setg(errp, "Invalid redundancy mode");
2233            ret = -EINVAL;
2234            goto fail;
2235        }
2236    }
2237
2238    ret = sd_co_create(create_options, errp);
2239fail:
2240    qapi_free_BlockdevCreateOptions(create_options);
2241    qobject_unref(qdict);
2242    g_free(redundancy);
2243    return ret;
2244}
2245
2246static void sd_close(BlockDriverState *bs)
2247{
2248    Error *local_err = NULL;
2249    BDRVSheepdogState *s = bs->opaque;
2250    SheepdogVdiReq hdr;
2251    SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
2252    unsigned int wlen, rlen = 0;
2253    int fd, ret;
2254
2255    DPRINTF("%s\n", s->name);
2256
2257    fd = connect_to_sdog(s, &local_err);
2258    if (fd < 0) {
2259        error_report_err(local_err);
2260        return;
2261    }
2262
2263    memset(&hdr, 0, sizeof(hdr));
2264
2265    hdr.opcode = SD_OP_RELEASE_VDI;
2266    hdr.type = LOCK_TYPE_NORMAL;
2267    hdr.base_vdi_id = s->inode.vdi_id;
2268    wlen = strlen(s->name) + 1;
2269    hdr.data_length = wlen;
2270    hdr.flags = SD_FLAG_CMD_WRITE;
2271
2272    ret = do_req(fd, s->bs, (SheepdogReq *)&hdr,
2273                 s->name, &wlen, &rlen);
2274
2275    closesocket(fd);
2276
2277    if (!ret && rsp->result != SD_RES_SUCCESS &&
2278        rsp->result != SD_RES_VDI_NOT_LOCKED) {
2279        error_report("%s, %s", sd_strerror(rsp->result), s->name);
2280    }
2281
2282    aio_set_fd_handler(bdrv_get_aio_context(bs), s->fd,
2283                       false, NULL, NULL, NULL, NULL);
2284    closesocket(s->fd);
2285    qapi_free_SocketAddress(s->addr);
2286}
2287
2288static int64_t sd_getlength(BlockDriverState *bs)
2289{
2290    BDRVSheepdogState *s = bs->opaque;
2291
2292    return s->inode.vdi_size;
2293}
2294
2295static int coroutine_fn sd_co_truncate(BlockDriverState *bs, int64_t offset,
2296                                       PreallocMode prealloc, Error **errp)
2297{
2298    BDRVSheepdogState *s = bs->opaque;
2299    int ret, fd;
2300    unsigned int datalen;
2301    uint64_t max_vdi_size;
2302    int64_t old_size = s->inode.vdi_size;
2303
2304    if (prealloc != PREALLOC_MODE_OFF && prealloc != PREALLOC_MODE_FULL) {
2305        error_setg(errp, "Unsupported preallocation mode '%s'",
2306                   PreallocMode_str(prealloc));
2307        return -ENOTSUP;
2308    }
2309
2310    max_vdi_size = (UINT64_C(1) << s->inode.block_size_shift) * MAX_DATA_OBJS;
2311    if (offset < old_size) {
2312        error_setg(errp, "shrinking is not supported");
2313        return -EINVAL;
2314    } else if (offset > max_vdi_size) {
2315        error_setg(errp, "too big image size");
2316        return -EINVAL;
2317    }
2318
2319    fd = connect_to_sdog(s, errp);
2320    if (fd < 0) {
2321        return fd;
2322    }
2323
2324    /* we don't need to update entire object */
2325    datalen = SD_INODE_HEADER_SIZE;
2326    s->inode.vdi_size = offset;
2327    ret = write_object(fd, s->bs, (char *)&s->inode,
2328                       vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies,
2329                       datalen, 0, false, s->cache_flags);
2330    close(fd);
2331
2332    if (ret < 0) {
2333        error_setg_errno(errp, -ret, "failed to update an inode");
2334        return ret;
2335    }
2336
2337    if (prealloc == PREALLOC_MODE_FULL) {
2338        ret = sd_prealloc(bs, old_size, offset, errp);
2339        if (ret < 0) {
2340            return ret;
2341        }
2342    }
2343
2344    return 0;
2345}
2346
2347/*
2348 * This function is called after writing data objects.  If we need to
2349 * update metadata, this sends a write request to the vdi object.
2350 */
2351static void coroutine_fn sd_write_done(SheepdogAIOCB *acb)
2352{
2353    BDRVSheepdogState *s = acb->s;
2354    struct iovec iov;
2355    AIOReq *aio_req;
2356    uint32_t offset, data_len, mn, mx;
2357
2358    mn = acb->min_dirty_data_idx;
2359    mx = acb->max_dirty_data_idx;
2360    if (mn <= mx) {
2361        /* we need to update the vdi object. */
2362        ++acb->nr_pending;
2363        offset = sizeof(s->inode) - sizeof(s->inode.data_vdi_id) +
2364            mn * sizeof(s->inode.data_vdi_id[0]);
2365        data_len = (mx - mn + 1) * sizeof(s->inode.data_vdi_id[0]);
2366
2367        acb->min_dirty_data_idx = UINT32_MAX;
2368        acb->max_dirty_data_idx = 0;
2369
2370        iov.iov_base = &s->inode;
2371        iov.iov_len = sizeof(s->inode);
2372        aio_req = alloc_aio_req(s, acb, vid_to_vdi_oid(s->inode.vdi_id),
2373                                data_len, offset, 0, false, 0, offset);
2374        add_aio_request(s, aio_req, &iov, 1, AIOCB_WRITE_UDATA);
2375        if (--acb->nr_pending) {
2376            qemu_coroutine_yield();
2377        }
2378    }
2379}
2380
2381/* Delete current working VDI on the snapshot chain */
2382static bool sd_delete(BDRVSheepdogState *s)
2383{
2384    Error *local_err = NULL;
2385    unsigned int wlen = SD_MAX_VDI_LEN, rlen = 0;
2386    SheepdogVdiReq hdr = {
2387        .opcode = SD_OP_DEL_VDI,
2388        .base_vdi_id = s->inode.vdi_id,
2389        .data_length = wlen,
2390        .flags = SD_FLAG_CMD_WRITE,
2391    };
2392    SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
2393    int fd, ret;
2394
2395    fd = connect_to_sdog(s, &local_err);
2396    if (fd < 0) {
2397        error_report_err(local_err);
2398        return false;
2399    }
2400
2401    ret = do_req(fd, s->bs, (SheepdogReq *)&hdr,
2402                 s->name, &wlen, &rlen);
2403    closesocket(fd);
2404    if (ret) {
2405        return false;
2406    }
2407    switch (rsp->result) {
2408    case SD_RES_NO_VDI:
2409        error_report("%s was already deleted", s->name);
2410        /* fall through */
2411    case SD_RES_SUCCESS:
2412        break;
2413    default:
2414        error_report("%s, %s", sd_strerror(rsp->result), s->name);
2415        return false;
2416    }
2417
2418    return true;
2419}
2420
2421/*
2422 * Create a writable VDI from a snapshot
2423 */
2424static int sd_create_branch(BDRVSheepdogState *s)
2425{
2426    Error *local_err = NULL;
2427    int ret, fd;
2428    uint32_t vid;
2429    char *buf;
2430    bool deleted;
2431
2432    DPRINTF("%" PRIx32 " is snapshot.\n", s->inode.vdi_id);
2433
2434    buf = g_malloc(SD_INODE_SIZE);
2435
2436    /*
2437     * Even If deletion fails, we will just create extra snapshot based on
2438     * the working VDI which was supposed to be deleted. So no need to
2439     * false bail out.
2440     */
2441    deleted = sd_delete(s);
2442    ret = do_sd_create(s, &vid, !deleted, &local_err);
2443    if (ret) {
2444        error_report_err(local_err);
2445        goto out;
2446    }
2447
2448    DPRINTF("%" PRIx32 " is created.\n", vid);
2449
2450    fd = connect_to_sdog(s, &local_err);
2451    if (fd < 0) {
2452        error_report_err(local_err);
2453        ret = fd;
2454        goto out;
2455    }
2456
2457    ret = read_object(fd, s->bs, buf, vid_to_vdi_oid(vid),
2458                      s->inode.nr_copies, SD_INODE_SIZE, 0, s->cache_flags);
2459
2460    closesocket(fd);
2461
2462    if (ret < 0) {
2463        goto out;
2464    }
2465
2466    memcpy(&s->inode, buf, sizeof(s->inode));
2467
2468    s->is_snapshot = false;
2469    ret = 0;
2470    DPRINTF("%" PRIx32 " was newly created.\n", s->inode.vdi_id);
2471
2472out:
2473    g_free(buf);
2474
2475    return ret;
2476}
2477
2478/*
2479 * Send I/O requests to the server.
2480 *
2481 * This function sends requests to the server, links the requests to
2482 * the inflight_list in BDRVSheepdogState, and exits without
2483 * waiting the response.  The responses are received in the
2484 * `aio_read_response' function which is called from the main loop as
2485 * a fd handler.
2486 *
2487 * Returns 1 when we need to wait a response, 0 when there is no sent
2488 * request and -errno in error cases.
2489 */
2490static void coroutine_fn sd_co_rw_vector(SheepdogAIOCB *acb)
2491{
2492    int ret = 0;
2493    unsigned long len, done = 0, total = acb->nb_sectors * BDRV_SECTOR_SIZE;
2494    unsigned long idx;
2495    uint32_t object_size;
2496    uint64_t oid;
2497    uint64_t offset;
2498    BDRVSheepdogState *s = acb->s;
2499    SheepdogInode *inode = &s->inode;
2500    AIOReq *aio_req;
2501
2502    if (acb->aiocb_type == AIOCB_WRITE_UDATA && s->is_snapshot) {
2503        /*
2504         * In the case we open the snapshot VDI, Sheepdog creates the
2505         * writable VDI when we do a write operation first.
2506         */
2507        ret = sd_create_branch(s);
2508        if (ret) {
2509            acb->ret = -EIO;
2510            return;
2511        }
2512    }
2513
2514    object_size = (UINT32_C(1) << inode->block_size_shift);
2515    idx = acb->sector_num * BDRV_SECTOR_SIZE / object_size;
2516    offset = (acb->sector_num * BDRV_SECTOR_SIZE) % object_size;
2517
2518    /*
2519     * Make sure we don't free the aiocb before we are done with all requests.
2520     * This additional reference is dropped at the end of this function.
2521     */
2522    acb->nr_pending++;
2523
2524    while (done != total) {
2525        uint8_t flags = 0;
2526        uint64_t old_oid = 0;
2527        bool create = false;
2528
2529        oid = vid_to_data_oid(inode->data_vdi_id[idx], idx);
2530
2531        len = MIN(total - done, object_size - offset);
2532
2533        switch (acb->aiocb_type) {
2534        case AIOCB_READ_UDATA:
2535            if (!inode->data_vdi_id[idx]) {
2536                qemu_iovec_memset(acb->qiov, done, 0, len);
2537                goto done;
2538            }
2539            break;
2540        case AIOCB_WRITE_UDATA:
2541            if (!inode->data_vdi_id[idx]) {
2542                create = true;
2543            } else if (!is_data_obj_writable(inode, idx)) {
2544                /* Copy-On-Write */
2545                create = true;
2546                old_oid = oid;
2547                flags = SD_FLAG_CMD_COW;
2548            }
2549            break;
2550        case AIOCB_DISCARD_OBJ:
2551            /*
2552             * We discard the object only when the whole object is
2553             * 1) allocated 2) trimmed. Otherwise, simply skip it.
2554             */
2555            if (len != object_size || inode->data_vdi_id[idx] == 0) {
2556                goto done;
2557            }
2558            break;
2559        default:
2560            break;
2561        }
2562
2563        if (create) {
2564            DPRINTF("update ino (%" PRIu32 ") %" PRIu64 " %" PRIu64 " %ld\n",
2565                    inode->vdi_id, oid,
2566                    vid_to_data_oid(inode->data_vdi_id[idx], idx), idx);
2567            oid = vid_to_data_oid(inode->vdi_id, idx);
2568            DPRINTF("new oid %" PRIx64 "\n", oid);
2569        }
2570
2571        aio_req = alloc_aio_req(s, acb, oid, len, offset, flags, create,
2572                                old_oid,
2573                                acb->aiocb_type == AIOCB_DISCARD_OBJ ?
2574                                0 : done);
2575        add_aio_request(s, aio_req, acb->qiov->iov, acb->qiov->niov,
2576                        acb->aiocb_type);
2577    done:
2578        offset = 0;
2579        idx++;
2580        done += len;
2581    }
2582    if (--acb->nr_pending) {
2583        qemu_coroutine_yield();
2584    }
2585}
2586
2587static void sd_aio_complete(SheepdogAIOCB *acb)
2588{
2589    BDRVSheepdogState *s;
2590    if (acb->aiocb_type == AIOCB_FLUSH_CACHE) {
2591        return;
2592    }
2593
2594    s = acb->s;
2595    qemu_co_mutex_lock(&s->queue_lock);
2596    QLIST_REMOVE(acb, aiocb_siblings);
2597    qemu_co_queue_restart_all(&s->overlapping_queue);
2598    qemu_co_mutex_unlock(&s->queue_lock);
2599}
2600
2601static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,
2602                                     int nb_sectors, QEMUIOVector *qiov,
2603                                     int flags)
2604{
2605    SheepdogAIOCB acb;
2606    int ret;
2607    int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE;
2608    BDRVSheepdogState *s = bs->opaque;
2609
2610    assert(!flags);
2611    if (offset > s->inode.vdi_size) {
2612        ret = sd_co_truncate(bs, offset, PREALLOC_MODE_OFF, NULL);
2613        if (ret < 0) {
2614            return ret;
2615        }
2616    }
2617
2618    sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_WRITE_UDATA);
2619    sd_co_rw_vector(&acb);
2620    sd_write_done(&acb);
2621    sd_aio_complete(&acb);
2622
2623    return acb.ret;
2624}
2625
2626static coroutine_fn int sd_co_readv(BlockDriverState *bs, int64_t sector_num,
2627                       int nb_sectors, QEMUIOVector *qiov)
2628{
2629    SheepdogAIOCB acb;
2630    BDRVSheepdogState *s = bs->opaque;
2631
2632    sd_aio_setup(&acb, s, qiov, sector_num, nb_sectors, AIOCB_READ_UDATA);
2633    sd_co_rw_vector(&acb);
2634    sd_aio_complete(&acb);
2635
2636    return acb.ret;
2637}
2638
2639static int coroutine_fn sd_co_flush_to_disk(BlockDriverState *bs)
2640{
2641    BDRVSheepdogState *s = bs->opaque;
2642    SheepdogAIOCB acb;
2643    AIOReq *aio_req;
2644
2645    if (s->cache_flags != SD_FLAG_CMD_CACHE) {
2646        return 0;
2647    }
2648
2649    sd_aio_setup(&acb, s, NULL, 0, 0, AIOCB_FLUSH_CACHE);
2650
2651    acb.nr_pending++;
2652    aio_req = alloc_aio_req(s, &acb, vid_to_vdi_oid(s->inode.vdi_id),
2653                            0, 0, 0, false, 0, 0);
2654    add_aio_request(s, aio_req, NULL, 0, acb.aiocb_type);
2655
2656    if (--acb.nr_pending) {
2657        qemu_coroutine_yield();
2658    }
2659
2660    sd_aio_complete(&acb);
2661    return acb.ret;
2662}
2663
2664static int sd_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
2665{
2666    Error *local_err = NULL;
2667    BDRVSheepdogState *s = bs->opaque;
2668    int ret, fd;
2669    uint32_t new_vid;
2670    SheepdogInode *inode;
2671    unsigned int datalen;
2672
2673    DPRINTF("sn_info: name %s id_str %s s: name %s vm_state_size %" PRId64 " "
2674            "is_snapshot %d\n", sn_info->name, sn_info->id_str,
2675            s->name, sn_info->vm_state_size, s->is_snapshot);
2676
2677    if (s->is_snapshot) {
2678        error_report("You can't create a snapshot of a snapshot VDI, "
2679                     "%s (%" PRIu32 ").", s->name, s->inode.vdi_id);
2680
2681        return -EINVAL;
2682    }
2683
2684    DPRINTF("%s %s\n", sn_info->name, sn_info->id_str);
2685
2686    s->inode.vm_state_size = sn_info->vm_state_size;
2687    s->inode.vm_clock_nsec = sn_info->vm_clock_nsec;
2688    /* It appears that inode.tag does not require a NUL terminator,
2689     * which means this use of strncpy is ok.
2690     */
2691    strncpy(s->inode.tag, sn_info->name, sizeof(s->inode.tag));
2692    /* we don't need to update entire object */
2693    datalen = SD_INODE_HEADER_SIZE;
2694    inode = g_malloc(datalen);
2695
2696    /* refresh inode. */
2697    fd = connect_to_sdog(s, &local_err);
2698    if (fd < 0) {
2699        error_report_err(local_err);
2700        ret = fd;
2701        goto cleanup;
2702    }
2703
2704    ret = write_object(fd, s->bs, (char *)&s->inode,
2705                       vid_to_vdi_oid(s->inode.vdi_id), s->inode.nr_copies,
2706                       datalen, 0, false, s->cache_flags);
2707    if (ret < 0) {
2708        error_report("failed to write snapshot's inode.");
2709        goto cleanup;
2710    }
2711
2712    ret = do_sd_create(s, &new_vid, 1, &local_err);
2713    if (ret < 0) {
2714        error_reportf_err(local_err,
2715                          "failed to create inode for snapshot: ");
2716        goto cleanup;
2717    }
2718
2719    ret = read_object(fd, s->bs, (char *)inode,
2720                      vid_to_vdi_oid(new_vid), s->inode.nr_copies, datalen, 0,
2721                      s->cache_flags);
2722
2723    if (ret < 0) {
2724        error_report("failed to read new inode info. %s", strerror(errno));
2725        goto cleanup;
2726    }
2727
2728    memcpy(&s->inode, inode, datalen);
2729    DPRINTF("s->inode: name %s snap_id %x oid %x\n",
2730            s->inode.name, s->inode.snap_id, s->inode.vdi_id);
2731
2732cleanup:
2733    g_free(inode);
2734    closesocket(fd);
2735    return ret;
2736}
2737
2738/*
2739 * We implement rollback(loadvm) operation to the specified snapshot by
2740 * 1) switch to the snapshot
2741 * 2) rely on sd_create_branch to delete working VDI and
2742 * 3) create a new working VDI based on the specified snapshot
2743 */
2744static int sd_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
2745{
2746    BDRVSheepdogState *s = bs->opaque;
2747    BDRVSheepdogState *old_s;
2748    char tag[SD_MAX_VDI_TAG_LEN];
2749    uint32_t snapid = 0;
2750    int ret;
2751
2752    if (!sd_parse_snapid_or_tag(snapshot_id, &snapid, tag)) {
2753        return -EINVAL;
2754    }
2755
2756    old_s = g_new(BDRVSheepdogState, 1);
2757
2758    memcpy(old_s, s, sizeof(BDRVSheepdogState));
2759
2760    ret = reload_inode(s, snapid, tag);
2761    if (ret) {
2762        goto out;
2763    }
2764
2765    ret = sd_create_branch(s);
2766    if (ret) {
2767        goto out;
2768    }
2769
2770    g_free(old_s);
2771
2772    return 0;
2773out:
2774    /* recover bdrv_sd_state */
2775    memcpy(s, old_s, sizeof(BDRVSheepdogState));
2776    g_free(old_s);
2777
2778    error_report("failed to open. recover old bdrv_sd_state.");
2779
2780    return ret;
2781}
2782
2783#define NR_BATCHED_DISCARD 128
2784
2785static int remove_objects(BDRVSheepdogState *s, Error **errp)
2786{
2787    int fd, i = 0, nr_objs = 0;
2788    int ret;
2789    SheepdogInode *inode = &s->inode;
2790
2791    fd = connect_to_sdog(s, errp);
2792    if (fd < 0) {
2793        return fd;
2794    }
2795
2796    nr_objs = count_data_objs(inode);
2797    while (i < nr_objs) {
2798        int start_idx, nr_filled_idx;
2799
2800        while (i < nr_objs && !inode->data_vdi_id[i]) {
2801            i++;
2802        }
2803        start_idx = i;
2804
2805        nr_filled_idx = 0;
2806        while (i < nr_objs && nr_filled_idx < NR_BATCHED_DISCARD) {
2807            if (inode->data_vdi_id[i]) {
2808                inode->data_vdi_id[i] = 0;
2809                nr_filled_idx++;
2810            }
2811
2812            i++;
2813        }
2814
2815        ret = write_object(fd, s->bs,
2816                           (char *)&inode->data_vdi_id[start_idx],
2817                           vid_to_vdi_oid(s->inode.vdi_id), inode->nr_copies,
2818                           (i - start_idx) * sizeof(uint32_t),
2819                           offsetof(struct SheepdogInode,
2820                                    data_vdi_id[start_idx]),
2821                           false, s->cache_flags);
2822        if (ret < 0) {
2823            error_setg(errp, "Failed to discard snapshot inode");
2824            goto out;
2825        }
2826    }
2827
2828    ret = 0;
2829out:
2830    closesocket(fd);
2831    return ret;
2832}
2833
2834static int sd_snapshot_delete(BlockDriverState *bs,
2835                              const char *snapshot_id,
2836                              const char *name,
2837                              Error **errp)
2838{
2839    /*
2840     * FIXME should delete the snapshot matching both @snapshot_id and
2841     * @name, but @name not used here
2842     */
2843    unsigned long snap_id = 0;
2844    char snap_tag[SD_MAX_VDI_TAG_LEN];
2845    int fd, ret;
2846    char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
2847    BDRVSheepdogState *s = bs->opaque;
2848    unsigned int wlen = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN, rlen = 0;
2849    uint32_t vid;
2850    SheepdogVdiReq hdr = {
2851        .opcode = SD_OP_DEL_VDI,
2852        .data_length = wlen,
2853        .flags = SD_FLAG_CMD_WRITE,
2854    };
2855    SheepdogVdiRsp *rsp = (SheepdogVdiRsp *)&hdr;
2856
2857    ret = remove_objects(s, errp);
2858    if (ret) {
2859        return ret;
2860    }
2861
2862    memset(buf, 0, sizeof(buf));
2863    memset(snap_tag, 0, sizeof(snap_tag));
2864    pstrcpy(buf, SD_MAX_VDI_LEN, s->name);
2865    /* TODO Use sd_parse_snapid() once this mess is cleaned up */
2866    ret = qemu_strtoul(snapshot_id, NULL, 10, &snap_id);
2867    if (ret || snap_id > UINT32_MAX) {
2868        /*
2869         * FIXME Since qemu_strtoul() returns -EINVAL when
2870         * @snapshot_id is null, @snapshot_id is mandatory.  Correct
2871         * would be to require at least one of @snapshot_id and @name.
2872         */
2873        error_setg(errp, "Invalid snapshot ID: %s",
2874                         snapshot_id ? snapshot_id : "<null>");
2875        return -EINVAL;
2876    }
2877
2878    if (snap_id) {
2879        hdr.snapid = (uint32_t) snap_id;
2880    } else {
2881        /* FIXME I suspect we should use @name here */
2882        /* FIXME don't truncate silently */
2883        pstrcpy(snap_tag, sizeof(snap_tag), snapshot_id);
2884        pstrcpy(buf + SD_MAX_VDI_LEN, SD_MAX_VDI_TAG_LEN, snap_tag);
2885    }
2886
2887    ret = find_vdi_name(s, s->name, snap_id, snap_tag, &vid, true, errp);
2888    if (ret) {
2889        return ret;
2890    }
2891
2892    fd = connect_to_sdog(s, errp);
2893    if (fd < 0) {
2894        return fd;
2895    }
2896
2897    ret = do_req(fd, s->bs, (SheepdogReq *)&hdr,
2898                 buf, &wlen, &rlen);
2899    closesocket(fd);
2900    if (ret) {
2901        error_setg_errno(errp, -ret, "Couldn't send request to server");
2902        return ret;
2903    }
2904
2905    switch (rsp->result) {
2906    case SD_RES_NO_VDI:
2907        error_setg(errp, "Can't find the snapshot");
2908        return -ENOENT;
2909    case SD_RES_SUCCESS:
2910        break;
2911    default:
2912        error_setg(errp, "%s", sd_strerror(rsp->result));
2913        return -EIO;
2914    }
2915
2916    return 0;
2917}
2918
2919static int sd_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
2920{
2921    Error *local_err = NULL;
2922    BDRVSheepdogState *s = bs->opaque;
2923    SheepdogReq req;
2924    int fd, nr = 1024, ret, max = BITS_TO_LONGS(SD_NR_VDIS) * sizeof(long);
2925    QEMUSnapshotInfo *sn_tab = NULL;
2926    unsigned wlen, rlen;
2927    int found = 0;
2928    SheepdogInode *inode;
2929    unsigned long *vdi_inuse;
2930    unsigned int start_nr;
2931    uint64_t hval;
2932    uint32_t vid;
2933
2934    vdi_inuse = g_malloc(max);
2935    inode = g_malloc(SD_INODE_HEADER_SIZE);
2936
2937    fd = connect_to_sdog(s, &local_err);
2938    if (fd < 0) {
2939        error_report_err(local_err);
2940        ret = fd;
2941        goto out;
2942    }
2943
2944    rlen = max;
2945    wlen = 0;
2946
2947    memset(&req, 0, sizeof(req));
2948
2949    req.opcode = SD_OP_READ_VDIS;
2950    req.data_length = max;
2951
2952    ret = do_req(fd, s->bs, &req, vdi_inuse, &wlen, &rlen);
2953
2954    closesocket(fd);
2955    if (ret) {
2956        goto out;
2957    }
2958
2959    sn_tab = g_new0(QEMUSnapshotInfo, nr);
2960
2961    /* calculate a vdi id with hash function */
2962    hval = fnv_64a_buf(s->name, strlen(s->name), FNV1A_64_INIT);
2963    start_nr = hval & (SD_NR_VDIS - 1);
2964
2965    fd = connect_to_sdog(s, &local_err);
2966    if (fd < 0) {
2967        error_report_err(local_err);
2968        ret = fd;
2969        goto out;
2970    }
2971
2972    for (vid = start_nr; found < nr; vid = (vid + 1) % SD_NR_VDIS) {
2973        if (!test_bit(vid, vdi_inuse)) {
2974            break;
2975        }
2976
2977        /* we don't need to read entire object */
2978        ret = read_object(fd, s->bs, (char *)inode,
2979                          vid_to_vdi_oid(vid),
2980                          0, SD_INODE_HEADER_SIZE, 0,
2981                          s->cache_flags);
2982
2983        if (ret) {
2984            continue;
2985        }
2986
2987        if (!strcmp(inode->name, s->name) && is_snapshot(inode)) {
2988            sn_tab[found].date_sec = inode->snap_ctime >> 32;
2989            sn_tab[found].date_nsec = inode->snap_ctime & 0xffffffff;
2990            sn_tab[found].vm_state_size = inode->vm_state_size;
2991            sn_tab[found].vm_clock_nsec = inode->vm_clock_nsec;
2992
2993            snprintf(sn_tab[found].id_str, sizeof(sn_tab[found].id_str),
2994                     "%" PRIu32, inode->snap_id);
2995            pstrcpy(sn_tab[found].name,
2996                    MIN(sizeof(sn_tab[found].name), sizeof(inode->tag)),
2997                    inode->tag);
2998            found++;
2999        }
3000    }
3001
3002    closesocket(fd);
3003out:
3004    *psn_tab = sn_tab;
3005
3006    g_free(vdi_inuse);
3007    g_free(inode);
3008
3009    if (ret < 0) {
3010        return ret;
3011    }
3012
3013    return found;
3014}
3015
3016static int do_load_save_vmstate(BDRVSheepdogState *s, uint8_t *data,
3017                                int64_t pos, int size, int load)
3018{
3019    Error *local_err = NULL;
3020    bool create;
3021    int fd, ret = 0, remaining = size;
3022    unsigned int data_len;
3023    uint64_t vmstate_oid;
3024    uint64_t offset;
3025    uint32_t vdi_index;
3026    uint32_t vdi_id = load ? s->inode.parent_vdi_id : s->inode.vdi_id;
3027    uint32_t object_size = (UINT32_C(1) << s->inode.block_size_shift);
3028
3029    fd = connect_to_sdog(s, &local_err);
3030    if (fd < 0) {
3031        error_report_err(local_err);
3032        return fd;
3033    }
3034
3035    while (remaining) {
3036        vdi_index = pos / object_size;
3037        offset = pos % object_size;
3038
3039        data_len = MIN(remaining, object_size - offset);
3040
3041        vmstate_oid = vid_to_vmstate_oid(vdi_id, vdi_index);
3042
3043        create = (offset == 0);
3044        if (load) {
3045            ret = read_object(fd, s->bs, (char *)data, vmstate_oid,
3046                              s->inode.nr_copies, data_len, offset,
3047                              s->cache_flags);
3048        } else {
3049            ret = write_object(fd, s->bs, (char *)data, vmstate_oid,
3050                               s->inode.nr_copies, data_len, offset, create,
3051                               s->cache_flags);
3052        }
3053
3054        if (ret < 0) {
3055            error_report("failed to save vmstate %s", strerror(errno));
3056            goto cleanup;
3057        }
3058
3059        pos += data_len;
3060        data += data_len;
3061        remaining -= data_len;
3062    }
3063    ret = size;
3064cleanup:
3065    closesocket(fd);
3066    return ret;
3067}
3068
3069static int sd_save_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
3070                           int64_t pos)
3071{
3072    BDRVSheepdogState *s = bs->opaque;
3073    void *buf;
3074    int ret;
3075
3076    buf = qemu_blockalign(bs, qiov->size);
3077    qemu_iovec_to_buf(qiov, 0, buf, qiov->size);
3078    ret = do_load_save_vmstate(s, (uint8_t *) buf, pos, qiov->size, 0);
3079    qemu_vfree(buf);
3080
3081    return ret;
3082}
3083
3084static int sd_load_vmstate(BlockDriverState *bs, QEMUIOVector *qiov,
3085                           int64_t pos)
3086{
3087    BDRVSheepdogState *s = bs->opaque;
3088    void *buf;
3089    int ret;
3090
3091    buf = qemu_blockalign(bs, qiov->size);
3092    ret = do_load_save_vmstate(s, buf, pos, qiov->size, 1);
3093    qemu_iovec_from_buf(qiov, 0, buf, qiov->size);
3094    qemu_vfree(buf);
3095
3096    return ret;
3097}
3098
3099
3100static coroutine_fn int sd_co_pdiscard(BlockDriverState *bs, int64_t offset,
3101                                      int bytes)
3102{
3103    SheepdogAIOCB acb;
3104    BDRVSheepdogState *s = bs->opaque;
3105    QEMUIOVector discard_iov;
3106    struct iovec iov;
3107    uint32_t zero = 0;
3108
3109    if (!s->discard_supported) {
3110        return 0;
3111    }
3112
3113    memset(&discard_iov, 0, sizeof(discard_iov));
3114    memset(&iov, 0, sizeof(iov));
3115    iov.iov_base = &zero;
3116    iov.iov_len = sizeof(zero);
3117    discard_iov.iov = &iov;
3118    discard_iov.niov = 1;
3119    if (!QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE)) {
3120        return -ENOTSUP;
3121    }
3122    sd_aio_setup(&acb, s, &discard_iov, offset >> BDRV_SECTOR_BITS,
3123                 bytes >> BDRV_SECTOR_BITS, AIOCB_DISCARD_OBJ);
3124    sd_co_rw_vector(&acb);
3125    sd_aio_complete(&acb);
3126
3127    return acb.ret;
3128}
3129
3130static coroutine_fn int
3131sd_co_block_status(BlockDriverState *bs, bool want_zero, int64_t offset,
3132                   int64_t bytes, int64_t *pnum, int64_t *map,
3133                   BlockDriverState **file)
3134{
3135    BDRVSheepdogState *s = bs->opaque;
3136    SheepdogInode *inode = &s->inode;
3137    uint32_t object_size = (UINT32_C(1) << inode->block_size_shift);
3138    unsigned long start = offset / object_size,
3139                  end = DIV_ROUND_UP(offset + bytes, object_size);
3140    unsigned long idx;
3141    *map = offset;
3142    int ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
3143
3144    for (idx = start; idx < end; idx++) {
3145        if (inode->data_vdi_id[idx] == 0) {
3146            break;
3147        }
3148    }
3149    if (idx == start) {
3150        /* Get the longest length of unallocated sectors */
3151        ret = 0;
3152        for (idx = start + 1; idx < end; idx++) {
3153            if (inode->data_vdi_id[idx] != 0) {
3154                break;
3155            }
3156        }
3157    }
3158
3159    *pnum = (idx - start) * object_size;
3160    if (*pnum > bytes) {
3161        *pnum = bytes;
3162    }
3163    if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
3164        *file = bs;
3165    }
3166    return ret;
3167}
3168
3169static int64_t sd_get_allocated_file_size(BlockDriverState *bs)
3170{
3171    BDRVSheepdogState *s = bs->opaque;
3172    SheepdogInode *inode = &s->inode;
3173    uint32_t object_size = (UINT32_C(1) << inode->block_size_shift);
3174    unsigned long i, last = DIV_ROUND_UP(inode->vdi_size, object_size);
3175    uint64_t size = 0;
3176
3177    for (i = 0; i < last; i++) {
3178        if (inode->data_vdi_id[i] == 0) {
3179            continue;
3180        }
3181        size += object_size;
3182    }
3183    return size;
3184}
3185
3186static QemuOptsList sd_create_opts = {
3187    .name = "sheepdog-create-opts",
3188    .head = QTAILQ_HEAD_INITIALIZER(sd_create_opts.head),
3189    .desc = {
3190        {
3191            .name = BLOCK_OPT_SIZE,
3192            .type = QEMU_OPT_SIZE,
3193            .help = "Virtual disk size"
3194        },
3195        {
3196            .name = BLOCK_OPT_BACKING_FILE,
3197            .type = QEMU_OPT_STRING,
3198            .help = "File name of a base image"
3199        },
3200        {
3201            .name = BLOCK_OPT_PREALLOC,
3202            .type = QEMU_OPT_STRING,
3203            .help = "Preallocation mode (allowed values: off, full)"
3204        },
3205        {
3206            .name = BLOCK_OPT_REDUNDANCY,
3207            .type = QEMU_OPT_STRING,
3208            .help = "Redundancy of the image"
3209        },
3210        {
3211            .name = BLOCK_OPT_OBJECT_SIZE,
3212            .type = QEMU_OPT_SIZE,
3213            .help = "Object size of the image"
3214        },
3215        { /* end of list */ }
3216    }
3217};
3218
3219static BlockDriver bdrv_sheepdog = {
3220    .format_name                  = "sheepdog",
3221    .protocol_name                = "sheepdog",
3222    .instance_size                = sizeof(BDRVSheepdogState),
3223    .bdrv_parse_filename          = sd_parse_filename,
3224    .bdrv_file_open               = sd_open,
3225    .bdrv_reopen_prepare          = sd_reopen_prepare,
3226    .bdrv_reopen_commit           = sd_reopen_commit,
3227    .bdrv_reopen_abort            = sd_reopen_abort,
3228    .bdrv_close                   = sd_close,
3229    .bdrv_co_create               = sd_co_create,
3230    .bdrv_co_create_opts          = sd_co_create_opts,
3231    .bdrv_has_zero_init           = bdrv_has_zero_init_1,
3232    .bdrv_getlength               = sd_getlength,
3233    .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
3234    .bdrv_co_truncate             = sd_co_truncate,
3235
3236    .bdrv_co_readv                = sd_co_readv,
3237    .bdrv_co_writev               = sd_co_writev,
3238    .bdrv_co_flush_to_disk        = sd_co_flush_to_disk,
3239    .bdrv_co_pdiscard             = sd_co_pdiscard,
3240    .bdrv_co_block_status         = sd_co_block_status,
3241
3242    .bdrv_snapshot_create         = sd_snapshot_create,
3243    .bdrv_snapshot_goto           = sd_snapshot_goto,
3244    .bdrv_snapshot_delete         = sd_snapshot_delete,
3245    .bdrv_snapshot_list           = sd_snapshot_list,
3246
3247    .bdrv_save_vmstate            = sd_save_vmstate,
3248    .bdrv_load_vmstate            = sd_load_vmstate,
3249
3250    .bdrv_detach_aio_context      = sd_detach_aio_context,
3251    .bdrv_attach_aio_context      = sd_attach_aio_context,
3252
3253    .create_opts                  = &sd_create_opts,
3254};
3255
3256static BlockDriver bdrv_sheepdog_tcp = {
3257    .format_name                  = "sheepdog",
3258    .protocol_name                = "sheepdog+tcp",
3259    .instance_size                = sizeof(BDRVSheepdogState),
3260    .bdrv_parse_filename          = sd_parse_filename,
3261    .bdrv_file_open               = sd_open,
3262    .bdrv_reopen_prepare          = sd_reopen_prepare,
3263    .bdrv_reopen_commit           = sd_reopen_commit,
3264    .bdrv_reopen_abort            = sd_reopen_abort,
3265    .bdrv_close                   = sd_close,
3266    .bdrv_co_create               = sd_co_create,
3267    .bdrv_co_create_opts          = sd_co_create_opts,
3268    .bdrv_has_zero_init           = bdrv_has_zero_init_1,
3269    .bdrv_getlength               = sd_getlength,
3270    .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
3271    .bdrv_co_truncate             = sd_co_truncate,
3272
3273    .bdrv_co_readv                = sd_co_readv,
3274    .bdrv_co_writev               = sd_co_writev,
3275    .bdrv_co_flush_to_disk        = sd_co_flush_to_disk,
3276    .bdrv_co_pdiscard             = sd_co_pdiscard,
3277    .bdrv_co_block_status         = sd_co_block_status,
3278
3279    .bdrv_snapshot_create         = sd_snapshot_create,
3280    .bdrv_snapshot_goto           = sd_snapshot_goto,
3281    .bdrv_snapshot_delete         = sd_snapshot_delete,
3282    .bdrv_snapshot_list           = sd_snapshot_list,
3283
3284    .bdrv_save_vmstate            = sd_save_vmstate,
3285    .bdrv_load_vmstate            = sd_load_vmstate,
3286
3287    .bdrv_detach_aio_context      = sd_detach_aio_context,
3288    .bdrv_attach_aio_context      = sd_attach_aio_context,
3289
3290    .create_opts                  = &sd_create_opts,
3291};
3292
3293static BlockDriver bdrv_sheepdog_unix = {
3294    .format_name                  = "sheepdog",
3295    .protocol_name                = "sheepdog+unix",
3296    .instance_size                = sizeof(BDRVSheepdogState),
3297    .bdrv_parse_filename          = sd_parse_filename,
3298    .bdrv_file_open               = sd_open,
3299    .bdrv_reopen_prepare          = sd_reopen_prepare,
3300    .bdrv_reopen_commit           = sd_reopen_commit,
3301    .bdrv_reopen_abort            = sd_reopen_abort,
3302    .bdrv_close                   = sd_close,
3303    .bdrv_co_create               = sd_co_create,
3304    .bdrv_co_create_opts          = sd_co_create_opts,
3305    .bdrv_has_zero_init           = bdrv_has_zero_init_1,
3306    .bdrv_getlength               = sd_getlength,
3307    .bdrv_get_allocated_file_size = sd_get_allocated_file_size,
3308    .bdrv_co_truncate             = sd_co_truncate,
3309
3310    .bdrv_co_readv                = sd_co_readv,
3311    .bdrv_co_writev               = sd_co_writev,
3312    .bdrv_co_flush_to_disk        = sd_co_flush_to_disk,
3313    .bdrv_co_pdiscard             = sd_co_pdiscard,
3314    .bdrv_co_block_status         = sd_co_block_status,
3315
3316    .bdrv_snapshot_create         = sd_snapshot_create,
3317    .bdrv_snapshot_goto           = sd_snapshot_goto,
3318    .bdrv_snapshot_delete         = sd_snapshot_delete,
3319    .bdrv_snapshot_list           = sd_snapshot_list,
3320
3321    .bdrv_save_vmstate            = sd_save_vmstate,
3322    .bdrv_load_vmstate            = sd_load_vmstate,
3323
3324    .bdrv_detach_aio_context      = sd_detach_aio_context,
3325    .bdrv_attach_aio_context      = sd_attach_aio_context,
3326
3327    .create_opts                  = &sd_create_opts,
3328};
3329
3330static void bdrv_sheepdog_init(void)
3331{
3332    bdrv_register(&bdrv_sheepdog);
3333    bdrv_register(&bdrv_sheepdog_tcp);
3334    bdrv_register(&bdrv_sheepdog_unix);
3335}
3336block_init(bdrv_sheepdog_init);
3337