block: Support Archipelago as a QEMU block backend
authorChrysostomos Nanakos <cnanakos@grnet.gr>
Mon, 4 Aug 2014 14:35:32 +0000 (17:35 +0300)
committerKevin Wolf <kwolf@redhat.com>
Fri, 15 Aug 2014 13:07:14 +0000 (15:07 +0200)
VM Image on Archipelago volume is specified like this:

file.driver=archipelago,file.volume=<volumename>[,file.mport=<mapperd_port>[,
file.vport=<vlmcd_port>][,file.segment=<segment_name>]]

'archipelago' is the protocol.

'mport' is the port number on which mapperd is listening. This is optional
and if not specified, QEMU will make Archipelago to use the default port.

'vport' is the port number on which vlmcd is listening. This is optional
and if not specified, QEMU will make Archipelago to use the default port.

'segment' is the name of the shared memory segment Archipelago stack is using.
This is optional and if not specified, QEMU will make Archipelago to use the
default value, 'archipelago'.

Examples:

file.driver=archipelago,file.volume=my_vm_volume
file.driver=archipelago,file.volume=my_vm_volume,file.mport=123
file.driver=archipelago,file.volume=my_vm_volume,file.mport=123,
file.vport=1234
file.driver=archipelago,file.volume=my_vm_volume,file.mport=123,
file.vport=1234,file.segment=my_segment

Signed-off-by: Chrysostomos Nanakos <cnanakos@grnet.gr>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
MAINTAINERS
block/Makefile.objs
block/archipelago.c [new file with mode: 0644]
configure

index 906f252..59940f9 100644 (file)
@@ -1000,3 +1000,9 @@ SSH
 M: Richard W.M. Jones <rjones@redhat.com>
 S: Supported
 F: block/ssh.c
+
+ARCHIPELAGO
+M: Chrysostomos Nanakos <cnanakos@grnet.gr>
+M: Chrysostomos Nanakos <chris@include.gr>
+S: Maintained
+F: block/archipelago.c
index fd88c03..858d2b3 100644 (file)
@@ -17,6 +17,7 @@ block-obj-$(CONFIG_LIBNFS) += nfs.o
 block-obj-$(CONFIG_CURL) += curl.o
 block-obj-$(CONFIG_RBD) += rbd.o
 block-obj-$(CONFIG_GLUSTERFS) += gluster.o
+block-obj-$(CONFIG_ARCHIPELAGO) += archipelago.o
 block-obj-$(CONFIG_LIBSSH2) += ssh.o
 endif
 
@@ -35,5 +36,6 @@ gluster.o-cflags   := $(GLUSTERFS_CFLAGS)
 gluster.o-libs     := $(GLUSTERFS_LIBS)
 ssh.o-cflags       := $(LIBSSH2_CFLAGS)
 ssh.o-libs         := $(LIBSSH2_LIBS)
+archipelago.o-libs := $(ARCHIPELAGO_LIBS)
 qcow.o-libs        := -lz
 linux-aio.o-libs   := -laio
diff --git a/block/archipelago.c b/block/archipelago.c
new file mode 100644 (file)
index 0000000..d91872d
--- /dev/null
@@ -0,0 +1,787 @@
+/*
+ * QEMU Block driver for Archipelago
+ *
+ * Copyright (C) 2014 Chrysostomos Nanakos <cnanakos@grnet.gr>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+/*
+ * VM Image on Archipelago volume is specified like this:
+ *
+ * file.driver=archipelago,file.volume=<volumename>
+ * [,file.mport=<mapperd_port>[,file.vport=<vlmcd_port>]
+ * [,file.segment=<segment_name>]]
+ *
+ * 'archipelago' is the protocol.
+ *
+ * 'mport' is the port number on which mapperd is listening. This is optional
+ * and if not specified, QEMU will make Archipelago to use the default port.
+ *
+ * 'vport' is the port number on which vlmcd is listening. This is optional
+ * and if not specified, QEMU will make Archipelago to use the default port.
+ *
+ * 'segment' is the name of the shared memory segment Archipelago stack
+ * is using. This is optional and if not specified, QEMU will make Archipelago
+ * to use the default value, 'archipelago'.
+ *
+ * Examples:
+ *
+ * file.driver=archipelago,file.volume=my_vm_volume
+ * file.driver=archipelago,file.volume=my_vm_volume,file.mport=123
+ * file.driver=archipelago,file.volume=my_vm_volume,file.mport=123,
+ * file.vport=1234
+ * file.driver=archipelago,file.volume=my_vm_volume,file.mport=123,
+ * file.vport=1234,file.segment=my_segment
+ */
+
+#include "block/block_int.h"
+#include "qemu/error-report.h"
+#include "qemu/thread.h"
+#include "qapi/qmp/qint.h"
+#include "qapi/qmp/qstring.h"
+#include "qapi/qmp/qjson.h"
+
+#include <inttypes.h>
+#include <xseg/xseg.h>
+#include <xseg/protocol.h>
+
+#define ARCHIP_FD_READ      0
+#define ARCHIP_FD_WRITE     1
+#define MAX_REQUEST_SIZE    524288
+
+#define ARCHIPELAGO_OPT_VOLUME      "volume"
+#define ARCHIPELAGO_OPT_SEGMENT     "segment"
+#define ARCHIPELAGO_OPT_MPORT       "mport"
+#define ARCHIPELAGO_OPT_VPORT       "vport"
+#define ARCHIPELAGO_DFL_MPORT       1001
+#define ARCHIPELAGO_DFL_VPORT       501
+
+#define archipelagolog(fmt, ...) \
+    do {                         \
+        fprintf(stderr, "archipelago\t%-24s: " fmt, __func__, ##__VA_ARGS__); \
+    } while (0)
+
+typedef enum {
+    ARCHIP_OP_READ,
+    ARCHIP_OP_WRITE,
+    ARCHIP_OP_FLUSH,
+    ARCHIP_OP_VOLINFO,
+} ARCHIPCmd;
+
+typedef struct ArchipelagoAIOCB {
+    BlockDriverAIOCB common;
+    QEMUBH *bh;
+    struct BDRVArchipelagoState *s;
+    QEMUIOVector *qiov;
+    ARCHIPCmd cmd;
+    bool cancelled;
+    int status;
+    int64_t size;
+    int64_t ret;
+} ArchipelagoAIOCB;
+
+typedef struct BDRVArchipelagoState {
+    ArchipelagoAIOCB *event_acb;
+    char *volname;
+    char *segment_name;
+    uint64_t size;
+    /* Archipelago specific */
+    struct xseg *xseg;
+    struct xseg_port *port;
+    xport srcport;
+    xport sport;
+    xport mportno;
+    xport vportno;
+    QemuMutex archip_mutex;
+    QemuCond archip_cond;
+    bool is_signaled;
+    /* Request handler specific */
+    QemuThread request_th;
+    QemuCond request_cond;
+    QemuMutex request_mutex;
+    bool th_is_signaled;
+    bool stopping;
+} BDRVArchipelagoState;
+
+typedef struct ArchipelagoSegmentedRequest {
+    size_t count;
+    size_t total;
+    int ref;
+    int failed;
+} ArchipelagoSegmentedRequest;
+
+typedef struct AIORequestData {
+    const char *volname;
+    off_t offset;
+    size_t size;
+    uint64_t bufidx;
+    int ret;
+    int op;
+    ArchipelagoAIOCB *aio_cb;
+    ArchipelagoSegmentedRequest *segreq;
+} AIORequestData;
+
+static void qemu_archipelago_complete_aio(void *opaque);
+
+static void init_local_signal(struct xseg *xseg, xport sport, xport srcport)
+{
+    if (xseg && (sport != srcport)) {
+        xseg_init_local_signal(xseg, srcport);
+        sport = srcport;
+    }
+}
+
+static void archipelago_finish_aiocb(AIORequestData *reqdata)
+{
+    if (reqdata->aio_cb->ret != reqdata->segreq->total) {
+        reqdata->aio_cb->ret = -EIO;
+    } else if (reqdata->aio_cb->ret == reqdata->segreq->total) {
+        reqdata->aio_cb->ret = 0;
+    }
+    reqdata->aio_cb->bh = aio_bh_new(
+                        bdrv_get_aio_context(reqdata->aio_cb->common.bs),
+                        qemu_archipelago_complete_aio, reqdata
+                        );
+    qemu_bh_schedule(reqdata->aio_cb->bh);
+}
+
+static int wait_reply(struct xseg *xseg, xport srcport, struct xseg_port *port,
+                      struct xseg_request *expected_req)
+{
+    struct xseg_request *req;
+    xseg_prepare_wait(xseg, srcport);
+    void *psd = xseg_get_signal_desc(xseg, port);
+    while (1) {
+        req = xseg_receive(xseg, srcport, X_NONBLOCK);
+        if (req) {
+            if (req != expected_req) {
+                archipelagolog("Unknown received request\n");
+                xseg_put_request(xseg, req, srcport);
+            } else if (!(req->state & XS_SERVED)) {
+                return -1;
+            } else {
+                break;
+            }
+        }
+        xseg_wait_signal(xseg, psd, 100000UL);
+    }
+    xseg_cancel_wait(xseg, srcport);
+    return 0;
+}
+
+static void xseg_request_handler(void *state)
+{
+    BDRVArchipelagoState *s = (BDRVArchipelagoState *) state;
+    void *psd = xseg_get_signal_desc(s->xseg, s->port);
+    qemu_mutex_lock(&s->request_mutex);
+
+    while (!s->stopping) {
+        struct xseg_request *req;
+        void *data;
+        xseg_prepare_wait(s->xseg, s->srcport);
+        req = xseg_receive(s->xseg, s->srcport, X_NONBLOCK);
+        if (req) {
+            AIORequestData *reqdata;
+            ArchipelagoSegmentedRequest *segreq;
+            xseg_get_req_data(s->xseg, req, (void **)&reqdata);
+
+            switch (reqdata->op) {
+            case ARCHIP_OP_READ:
+                data = xseg_get_data(s->xseg, req);
+                segreq = reqdata->segreq;
+                segreq->count += req->serviced;
+
+                qemu_iovec_from_buf(reqdata->aio_cb->qiov, reqdata->bufidx,
+                                    data,
+                                    req->serviced);
+
+                xseg_put_request(s->xseg, req, s->srcport);
+
+                if ((__sync_add_and_fetch(&segreq->ref, -1)) == 0) {
+                    if (!segreq->failed) {
+                        reqdata->aio_cb->ret = segreq->count;
+                        archipelago_finish_aiocb(reqdata);
+                        g_free(segreq);
+                    } else {
+                        g_free(segreq);
+                        g_free(reqdata);
+                    }
+                } else {
+                    g_free(reqdata);
+                }
+                break;
+            case ARCHIP_OP_WRITE:
+            case ARCHIP_OP_FLUSH:
+                segreq = reqdata->segreq;
+                segreq->count += req->serviced;
+                xseg_put_request(s->xseg, req, s->srcport);
+
+                if ((__sync_add_and_fetch(&segreq->ref, -1)) == 0) {
+                    if (!segreq->failed) {
+                        reqdata->aio_cb->ret = segreq->count;
+                        archipelago_finish_aiocb(reqdata);
+                        g_free(segreq);
+                    } else {
+                        g_free(segreq);
+                        g_free(reqdata);
+                    }
+                } else {
+                    g_free(reqdata);
+                }
+                break;
+            case ARCHIP_OP_VOLINFO:
+                s->is_signaled = true;
+                qemu_cond_signal(&s->archip_cond);
+                break;
+            }
+        } else {
+            xseg_wait_signal(s->xseg, psd, 100000UL);
+        }
+        xseg_cancel_wait(s->xseg, s->srcport);
+    }
+
+    s->th_is_signaled = true;
+    qemu_cond_signal(&s->request_cond);
+    qemu_mutex_unlock(&s->request_mutex);
+    qemu_thread_exit(NULL);
+}
+
+static int qemu_archipelago_xseg_init(BDRVArchipelagoState *s)
+{
+    if (xseg_initialize()) {
+        archipelagolog("Cannot initialize XSEG\n");
+        goto err_exit;
+    }
+
+    s->xseg = xseg_join("posix", s->segment_name,
+                        "posixfd", NULL);
+    if (!s->xseg) {
+        archipelagolog("Cannot join XSEG shared memory segment\n");
+        goto err_exit;
+    }
+    s->port = xseg_bind_dynport(s->xseg);
+    s->srcport = s->port->portno;
+    init_local_signal(s->xseg, s->sport, s->srcport);
+    return 0;
+
+err_exit:
+    return -1;
+}
+
+static int qemu_archipelago_init(BDRVArchipelagoState *s)
+{
+    int ret;
+
+    ret = qemu_archipelago_xseg_init(s);
+    if (ret < 0) {
+        error_report("Cannot initialize XSEG. Aborting...\n");
+        goto err_exit;
+    }
+
+    qemu_cond_init(&s->archip_cond);
+    qemu_mutex_init(&s->archip_mutex);
+    qemu_cond_init(&s->request_cond);
+    qemu_mutex_init(&s->request_mutex);
+    s->th_is_signaled = false;
+    qemu_thread_create(&s->request_th, "xseg_io_th",
+                       (void *) xseg_request_handler,
+                       (void *) s, QEMU_THREAD_JOINABLE);
+
+err_exit:
+    return ret;
+}
+
+static void qemu_archipelago_complete_aio(void *opaque)
+{
+    AIORequestData *reqdata = (AIORequestData *) opaque;
+    ArchipelagoAIOCB *aio_cb = (ArchipelagoAIOCB *) reqdata->aio_cb;
+
+    qemu_bh_delete(aio_cb->bh);
+    aio_cb->common.cb(aio_cb->common.opaque, aio_cb->ret);
+    aio_cb->status = 0;
+
+    if (!aio_cb->cancelled) {
+        qemu_aio_release(aio_cb);
+    }
+    g_free(reqdata);
+}
+
+static QemuOptsList archipelago_runtime_opts = {
+    .name = "archipelago",
+    .head = QTAILQ_HEAD_INITIALIZER(archipelago_runtime_opts.head),
+    .desc = {
+        {
+            .name = ARCHIPELAGO_OPT_VOLUME,
+            .type = QEMU_OPT_STRING,
+            .help = "Name of the volume image",
+        },
+        {
+            .name = ARCHIPELAGO_OPT_SEGMENT,
+            .type = QEMU_OPT_STRING,
+            .help = "Name of the Archipelago shared memory segment",
+        },
+        {
+            .name = ARCHIPELAGO_OPT_MPORT,
+            .type = QEMU_OPT_NUMBER,
+            .help = "Archipelago mapperd port number"
+        },
+        {
+            .name = ARCHIPELAGO_OPT_VPORT,
+            .type = QEMU_OPT_NUMBER,
+            .help = "Archipelago vlmcd port number"
+
+        },
+        { /* end of list */ }
+    },
+};
+
+static int qemu_archipelago_open(BlockDriverState *bs,
+                                 QDict *options,
+                                 int bdrv_flags,
+                                 Error **errp)
+{
+    int ret = 0;
+    const char *volume, *segment_name;
+    QemuOpts *opts;
+    Error *local_err = NULL;
+    BDRVArchipelagoState *s = bs->opaque;
+
+    opts = qemu_opts_create(&archipelago_runtime_opts, NULL, 0, &error_abort);
+    qemu_opts_absorb_qdict(opts, options, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        ret = -EINVAL;
+        goto err_exit;
+    }
+
+    s->mportno = qemu_opt_get_number(opts, ARCHIPELAGO_OPT_MPORT,
+                                     ARCHIPELAGO_DFL_MPORT);
+    s->vportno = qemu_opt_get_number(opts, ARCHIPELAGO_OPT_VPORT,
+                                     ARCHIPELAGO_DFL_VPORT);
+
+    segment_name = qemu_opt_get(opts, ARCHIPELAGO_OPT_SEGMENT);
+    if (segment_name == NULL) {
+        s->segment_name = g_strdup("archipelago");
+    } else {
+        s->segment_name = g_strdup(segment_name);
+    }
+
+    volume = qemu_opt_get(opts, ARCHIPELAGO_OPT_VOLUME);
+    if (volume == NULL) {
+        error_setg(errp, "archipelago block driver requires the 'volume'"
+                   " option");
+        ret = -EINVAL;
+        goto err_exit;
+    }
+    s->volname = g_strdup(volume);
+
+    /* Initialize XSEG, join shared memory segment */
+    ret = qemu_archipelago_init(s);
+    if (ret < 0) {
+        error_setg(errp, "cannot initialize XSEG and join shared "
+                   "memory segment");
+        goto err_exit;
+    }
+
+    qemu_opts_del(opts);
+    return 0;
+
+err_exit:
+    g_free(s->volname);
+    g_free(s->segment_name);
+    qemu_opts_del(opts);
+    return ret;
+}
+
+static void qemu_archipelago_close(BlockDriverState *bs)
+{
+    int r, targetlen;
+    char *target;
+    struct xseg_request *req;
+    BDRVArchipelagoState *s = bs->opaque;
+
+    s->stopping = true;
+
+    qemu_mutex_lock(&s->request_mutex);
+    while (!s->th_is_signaled) {
+        qemu_cond_wait(&s->request_cond,
+                       &s->request_mutex);
+    }
+    qemu_mutex_unlock(&s->request_mutex);
+    qemu_thread_join(&s->request_th);
+    qemu_cond_destroy(&s->request_cond);
+    qemu_mutex_destroy(&s->request_mutex);
+
+    qemu_cond_destroy(&s->archip_cond);
+    qemu_mutex_destroy(&s->archip_mutex);
+
+    targetlen = strlen(s->volname);
+    req = xseg_get_request(s->xseg, s->srcport, s->vportno, X_ALLOC);
+    if (!req) {
+        archipelagolog("Cannot get XSEG request\n");
+        goto err_exit;
+    }
+    r = xseg_prep_request(s->xseg, req, targetlen, 0);
+    if (r < 0) {
+        xseg_put_request(s->xseg, req, s->srcport);
+        archipelagolog("Cannot prepare XSEG close request\n");
+        goto err_exit;
+    }
+
+    target = xseg_get_target(s->xseg, req);
+    memcpy(target, s->volname, targetlen);
+    req->size = req->datalen;
+    req->offset = 0;
+    req->op = X_CLOSE;
+
+    xport p = xseg_submit(s->xseg, req, s->srcport, X_ALLOC);
+    if (p == NoPort) {
+        xseg_put_request(s->xseg, req, s->srcport);
+        archipelagolog("Cannot submit XSEG close request\n");
+        goto err_exit;
+    }
+
+    xseg_signal(s->xseg, p);
+    wait_reply(s->xseg, s->srcport, s->port, req);
+
+    xseg_put_request(s->xseg, req, s->srcport);
+
+err_exit:
+    g_free(s->volname);
+    g_free(s->segment_name);
+    xseg_quit_local_signal(s->xseg, s->srcport);
+    xseg_leave_dynport(s->xseg, s->port);
+    xseg_leave(s->xseg);
+}
+
+static void qemu_archipelago_aio_cancel(BlockDriverAIOCB *blockacb)
+{
+    ArchipelagoAIOCB *aio_cb = (ArchipelagoAIOCB *) blockacb;
+    aio_cb->cancelled = true;
+    while (aio_cb->status == -EINPROGRESS) {
+        aio_poll(bdrv_get_aio_context(aio_cb->common.bs), true);
+    }
+    qemu_aio_release(aio_cb);
+}
+
+static const AIOCBInfo archipelago_aiocb_info = {
+    .aiocb_size = sizeof(ArchipelagoAIOCB),
+    .cancel = qemu_archipelago_aio_cancel,
+};
+
+static int archipelago_submit_request(BDRVArchipelagoState *s,
+                                        uint64_t bufidx,
+                                        size_t count,
+                                        off_t offset,
+                                        ArchipelagoAIOCB *aio_cb,
+                                        ArchipelagoSegmentedRequest *segreq,
+                                        int op)
+{
+    int ret, targetlen;
+    char *target;
+    void *data = NULL;
+    struct xseg_request *req;
+    AIORequestData *reqdata = g_malloc(sizeof(AIORequestData));
+
+    targetlen = strlen(s->volname);
+    req = xseg_get_request(s->xseg, s->srcport, s->vportno, X_ALLOC);
+    if (!req) {
+        archipelagolog("Cannot get XSEG request\n");
+        goto err_exit2;
+    }
+    ret = xseg_prep_request(s->xseg, req, targetlen, count);
+    if (ret < 0) {
+        archipelagolog("Cannot prepare XSEG request\n");
+        goto err_exit;
+    }
+    target = xseg_get_target(s->xseg, req);
+    if (!target) {
+        archipelagolog("Cannot get XSEG target\n");
+        goto err_exit;
+    }
+    memcpy(target, s->volname, targetlen);
+    req->size = count;
+    req->offset = offset;
+
+    switch (op) {
+    case ARCHIP_OP_READ:
+        req->op = X_READ;
+        break;
+    case ARCHIP_OP_WRITE:
+        req->op = X_WRITE;
+        break;
+    case ARCHIP_OP_FLUSH:
+        req->op = X_FLUSH;
+        break;
+    }
+    reqdata->volname = s->volname;
+    reqdata->offset = offset;
+    reqdata->size = count;
+    reqdata->bufidx = bufidx;
+    reqdata->aio_cb = aio_cb;
+    reqdata->segreq = segreq;
+    reqdata->op = op;
+
+    xseg_set_req_data(s->xseg, req, reqdata);
+    if (op == ARCHIP_OP_WRITE) {
+        data = xseg_get_data(s->xseg, req);
+        if (!data) {
+            archipelagolog("Cannot get XSEG data\n");
+            goto err_exit;
+        }
+        qemu_iovec_to_buf(aio_cb->qiov, bufidx, data, count);
+    }
+
+    xport p = xseg_submit(s->xseg, req, s->srcport, X_ALLOC);
+    if (p == NoPort) {
+        archipelagolog("Could not submit XSEG request\n");
+        goto err_exit;
+    }
+    xseg_signal(s->xseg, p);
+    return 0;
+
+err_exit:
+    g_free(reqdata);
+    xseg_put_request(s->xseg, req, s->srcport);
+    return -EIO;
+err_exit2:
+    g_free(reqdata);
+    return -EIO;
+}
+
+static int archipelago_aio_segmented_rw(BDRVArchipelagoState *s,
+                                        size_t count,
+                                        off_t offset,
+                                        ArchipelagoAIOCB *aio_cb,
+                                        int op)
+{
+    int i, ret, segments_nr, last_segment_size;
+    ArchipelagoSegmentedRequest *segreq;
+
+    segreq = g_malloc(sizeof(ArchipelagoSegmentedRequest));
+
+    if (op == ARCHIP_OP_FLUSH) {
+        segments_nr = 1;
+        segreq->ref = segments_nr;
+        segreq->total = count;
+        segreq->count = 0;
+        segreq->failed = 0;
+        ret = archipelago_submit_request(s, 0, count, offset, aio_cb,
+                                           segreq, ARCHIP_OP_FLUSH);
+        if (ret < 0) {
+            goto err_exit;
+        }
+        return 0;
+    }
+
+    segments_nr = (int)(count / MAX_REQUEST_SIZE) + \
+                  ((count % MAX_REQUEST_SIZE) ? 1 : 0);
+    last_segment_size = (int)(count % MAX_REQUEST_SIZE);
+
+    segreq->ref = segments_nr;
+    segreq->total = count;
+    segreq->count = 0;
+    segreq->failed = 0;
+
+    for (i = 0; i < segments_nr - 1; i++) {
+        ret = archipelago_submit_request(s, i * MAX_REQUEST_SIZE,
+                                           MAX_REQUEST_SIZE,
+                                           offset + i * MAX_REQUEST_SIZE,
+                                           aio_cb, segreq, op);
+
+        if (ret < 0) {
+            goto err_exit;
+        }
+    }
+
+    if ((segments_nr > 1) && last_segment_size) {
+        ret = archipelago_submit_request(s, i * MAX_REQUEST_SIZE,
+                                           last_segment_size,
+                                           offset + i * MAX_REQUEST_SIZE,
+                                           aio_cb, segreq, op);
+    } else if ((segments_nr > 1) && !last_segment_size) {
+        ret = archipelago_submit_request(s, i * MAX_REQUEST_SIZE,
+                                           MAX_REQUEST_SIZE,
+                                           offset + i * MAX_REQUEST_SIZE,
+                                           aio_cb, segreq, op);
+    } else if (segments_nr == 1) {
+            ret = archipelago_submit_request(s, 0, count, offset, aio_cb,
+                                               segreq, op);
+    }
+
+    if (ret < 0) {
+        goto err_exit;
+    }
+
+    return 0;
+
+err_exit:
+    __sync_add_and_fetch(&segreq->failed, 1);
+    if (segments_nr == 1) {
+        if (__sync_add_and_fetch(&segreq->ref, -1) == 0) {
+            g_free(segreq);
+        }
+    } else {
+        if ((__sync_add_and_fetch(&segreq->ref, -segments_nr + i)) == 0) {
+            g_free(segreq);
+        }
+    }
+
+    return ret;
+}
+
+static BlockDriverAIOCB *qemu_archipelago_aio_rw(BlockDriverState *bs,
+                                                 int64_t sector_num,
+                                                 QEMUIOVector *qiov,
+                                                 int nb_sectors,
+                                                 BlockDriverCompletionFunc *cb,
+                                                 void *opaque,
+                                                 int op)
+{
+    ArchipelagoAIOCB *aio_cb;
+    BDRVArchipelagoState *s = bs->opaque;
+    int64_t size, off;
+    int ret;
+
+    aio_cb = qemu_aio_get(&archipelago_aiocb_info, bs, cb, opaque);
+    aio_cb->cmd = op;
+    aio_cb->qiov = qiov;
+
+    aio_cb->ret = 0;
+    aio_cb->s = s;
+    aio_cb->cancelled = false;
+    aio_cb->status = -EINPROGRESS;
+
+    off = sector_num * BDRV_SECTOR_SIZE;
+    size = nb_sectors * BDRV_SECTOR_SIZE;
+    aio_cb->size = size;
+
+    ret = archipelago_aio_segmented_rw(s, size, off,
+                                       aio_cb, op);
+    if (ret < 0) {
+        goto err_exit;
+    }
+    return &aio_cb->common;
+
+err_exit:
+    error_report("qemu_archipelago_aio_rw(): I/O Error\n");
+    qemu_aio_release(aio_cb);
+    return NULL;
+}
+
+static BlockDriverAIOCB *qemu_archipelago_aio_readv(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
+        BlockDriverCompletionFunc *cb, void *opaque)
+{
+    return qemu_archipelago_aio_rw(bs, sector_num, qiov, nb_sectors, cb,
+                                   opaque, ARCHIP_OP_READ);
+}
+
+static BlockDriverAIOCB *qemu_archipelago_aio_writev(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
+        BlockDriverCompletionFunc *cb, void *opaque)
+{
+    return qemu_archipelago_aio_rw(bs, sector_num, qiov, nb_sectors, cb,
+                                   opaque, ARCHIP_OP_WRITE);
+}
+
+static int64_t archipelago_volume_info(BDRVArchipelagoState *s)
+{
+    uint64_t size;
+    int ret, targetlen;
+    struct xseg_request *req;
+    struct xseg_reply_info *xinfo;
+    AIORequestData *reqdata = g_malloc(sizeof(AIORequestData));
+
+    const char *volname = s->volname;
+    targetlen = strlen(volname);
+    req = xseg_get_request(s->xseg, s->srcport, s->mportno, X_ALLOC);
+    if (!req) {
+        archipelagolog("Cannot get XSEG request\n");
+        goto err_exit2;
+    }
+    ret = xseg_prep_request(s->xseg, req, targetlen,
+                            sizeof(struct xseg_reply_info));
+    if (ret < 0) {
+        archipelagolog("Cannot prepare XSEG request\n");
+        goto err_exit;
+    }
+    char *target = xseg_get_target(s->xseg, req);
+    if (!target) {
+        archipelagolog("Cannot get XSEG target\n");
+        goto err_exit;
+    }
+    memcpy(target, volname, targetlen);
+    req->size = req->datalen;
+    req->offset = 0;
+    req->op = X_INFO;
+
+    reqdata->op = ARCHIP_OP_VOLINFO;
+    reqdata->volname = volname;
+    xseg_set_req_data(s->xseg, req, reqdata);
+
+    xport p = xseg_submit(s->xseg, req, s->srcport, X_ALLOC);
+    if (p == NoPort) {
+        archipelagolog("Cannot submit XSEG request\n");
+        goto err_exit;
+    }
+    xseg_signal(s->xseg, p);
+    qemu_mutex_lock(&s->archip_mutex);
+    while (!s->is_signaled) {
+        qemu_cond_wait(&s->archip_cond, &s->archip_mutex);
+    }
+    s->is_signaled = false;
+    qemu_mutex_unlock(&s->archip_mutex);
+
+    xinfo = (struct xseg_reply_info *) xseg_get_data(s->xseg, req);
+    size = xinfo->size;
+    xseg_put_request(s->xseg, req, s->srcport);
+    g_free(reqdata);
+    s->size = size;
+    return size;
+
+err_exit:
+    xseg_put_request(s->xseg, req, s->srcport);
+err_exit2:
+    g_free(reqdata);
+    return -EIO;
+}
+
+static int64_t qemu_archipelago_getlength(BlockDriverState *bs)
+{
+    int64_t ret;
+    BDRVArchipelagoState *s = bs->opaque;
+
+    ret = archipelago_volume_info(s);
+    return ret;
+}
+
+static BlockDriverAIOCB *qemu_archipelago_aio_flush(BlockDriverState *bs,
+        BlockDriverCompletionFunc *cb, void *opaque)
+{
+    return qemu_archipelago_aio_rw(bs, 0, NULL, 0, cb, opaque,
+                                   ARCHIP_OP_FLUSH);
+}
+
+static BlockDriver bdrv_archipelago = {
+    .format_name         = "archipelago",
+    .protocol_name       = "archipelago",
+    .instance_size       = sizeof(BDRVArchipelagoState),
+    .bdrv_file_open      = qemu_archipelago_open,
+    .bdrv_close          = qemu_archipelago_close,
+    .bdrv_getlength      = qemu_archipelago_getlength,
+    .bdrv_aio_readv      = qemu_archipelago_aio_readv,
+    .bdrv_aio_writev     = qemu_archipelago_aio_writev,
+    .bdrv_aio_flush      = qemu_archipelago_aio_flush,
+    .bdrv_has_zero_init  = bdrv_has_zero_init_1,
+};
+
+static void bdrv_archipelago_init(void)
+{
+    bdrv_register(&bdrv_archipelago);
+}
+
+block_init(bdrv_archipelago_init);
index 5715fe6..09e4f98 100755 (executable)
--- a/configure
+++ b/configure
@@ -326,6 +326,7 @@ seccomp=""
 glusterfs=""
 glusterfs_discard="no"
 glusterfs_zerofill="no"
+archipelago=""
 virtio_blk_data_plane=""
 gtk=""
 gtkabi=""
@@ -1087,6 +1088,10 @@ for opt do
   ;;
   --enable-glusterfs) glusterfs="yes"
   ;;
+  --disable-archipelago) archipelago="no"
+  ;;
+  --enable-archipelago) archipelago="yes"
+  ;;
   --disable-virtio-blk-data-plane) virtio_blk_data_plane="no"
   ;;
   --enable-virtio-blk-data-plane) virtio_blk_data_plane="yes"
@@ -1382,6 +1387,8 @@ Advanced options (experts only):
   --enable-coroutine-pool  enable coroutine freelist (better performance)
   --enable-glusterfs       enable GlusterFS backend
   --disable-glusterfs      disable GlusterFS backend
+  --enable-archipelago     enable Archipelago backend
+  --disable-archipelago    disable Archipelago backend
   --enable-gcov            enable test coverage analysis with gcov
   --gcov=GCOV              use specified gcov [$gcov_tool]
   --disable-tpm            disable TPM support
@@ -3071,6 +3078,33 @@ EOF
   fi
 fi
 
+
+##########################################
+# archipelago probe
+if test "$archipelago" != "no" ; then
+    cat > $TMPC <<EOF
+#include <stdio.h>
+#include <xseg/xseg.h>
+#include <xseg/protocol.h>
+int main(void) {
+    xseg_initialize();
+    return 0;
+}
+EOF
+    archipelago_libs=-lxseg
+    if compile_prog "" "$archipelago_libs"; then
+        archipelago="yes"
+        libs_tools="$archipelago_libs $libs_tools"
+        libs_softmmu="$archipelago_libs $libs_softmmu"
+    else
+      if test "$archipelago" = "yes" ; then
+        feature_not_found "Archipelago backend support" "Install libxseg devel"
+      fi
+      archipelago="no"
+    fi
+fi
+
+
 ##########################################
 # glusterfs probe
 if test "$glusterfs" != "no" ; then
@@ -4252,6 +4286,7 @@ echo "seccomp support   $seccomp"
 echo "coroutine backend $coroutine"
 echo "coroutine pool    $coroutine_pool"
 echo "GlusterFS support $glusterfs"
+echo "Archipelago support $archipelago"
 echo "virtio-blk-data-plane $virtio_blk_data_plane"
 echo "gcov              $gcov_tool"
 echo "gcov enabled      $gcov"
@@ -4690,6 +4725,11 @@ if test "$glusterfs_zerofill" = "yes" ; then
   echo "CONFIG_GLUSTERFS_ZEROFILL=y" >> $config_host_mak
 fi
 
+if test "$archipelago" = "yes" ; then
+  echo "CONFIG_ARCHIPELAGO=m" >> $config_host_mak
+  echo "ARCHIPELAGO_LIBS=$archipelago_libs" >> $config_host_mak
+fi
+
 if test "$libssh2" = "yes" ; then
   echo "CONFIG_LIBSSH2=m" >> $config_host_mak
   echo "LIBSSH2_CFLAGS=$libssh2_cflags" >> $config_host_mak