block/vxhs.c

   1 /*
   2  * QEMU Block driver for Veritas HyperScale (VxHS)
   3  *
   4  * Copyright (c) 2017 Veritas Technologies LLC.
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
   7  * See the COPYING file in the top-level directory.
   8  *
   9  */
  10
  11 #include "qemu/osdep.h"
  12 #include <qnio/qnio_api.h>
  13 #include <sys/param.h>
  14 #include "block/block_int.h"
  15 #include "qapi/qmp/qerror.h"
  16 #include "qapi/qmp/qdict.h"
  17 #include "qapi/qmp/qstring.h"
  18 #include "trace.h"
  19 #include "qemu/uri.h"
  20 #include "qapi/error.h"
  21 #include "qemu/uuid.h"
  22 #include "crypto/tlscredsx509.h"
  23
  24 #define VXHS_OPT_FILENAME           "filename"
  25 #define VXHS_OPT_VDISK_ID           "vdisk-id"
  26 #define VXHS_OPT_SERVER             "server"
  27 #define VXHS_OPT_HOST               "host"
  28 #define VXHS_OPT_PORT               "port"
  29
  30 /* Only accessed under QEMU global mutex */
  31 static uint32_t vxhs_ref;
  32
  33 typedef enum {
  34     VDISK_AIO_READ,
  35     VDISK_AIO_WRITE,
  36 } VDISKAIOCmd;
  37
  38 /*
  39  * HyperScale AIO callbacks structure
  40  */
  41 typedef struct VXHSAIOCB {
  42     BlockAIOCB common;
  43     int err;
  44 } VXHSAIOCB;
  45
  46 typedef struct VXHSvDiskHostsInfo {
  47     void *dev_handle; /* Device handle */
  48     char *host; /* Host name or IP */
  49     int port; /* Host's port number */
  50 } VXHSvDiskHostsInfo;
  51
  52 /*
  53  * Structure per vDisk maintained for state
  54  */
  55 typedef struct BDRVVXHSState {
  56     VXHSvDiskHostsInfo vdisk_hostinfo; /* Per host info */
  57     char *vdisk_guid;
  58     char *tlscredsid; /* tlscredsid */
  59 } BDRVVXHSState;
  60
  61 static void vxhs_complete_aio_bh(void *opaque)
  62 {
  63     VXHSAIOCB *acb = opaque;
  64     BlockCompletionFunc *cb = acb->common.cb;
  65     void *cb_opaque = acb->common.opaque;
  66     int ret = 0;
  67
  68     if (acb->err != 0) {
  69         trace_vxhs_complete_aio(acb, acb->err);
  70         ret = (-EIO);
  71     }
  72
  73     qemu_aio_unref(acb);
  74     cb(cb_opaque, ret);
  75 }
  76
  77 /*
  78  * Called from a libqnio thread
  79  */
  80 static void vxhs_iio_callback(void *ctx, uint32_t opcode, uint32_t error)
  81 {
  82     VXHSAIOCB *acb = NULL;
  83
  84     switch (opcode) {
  85     case IRP_READ_REQUEST:
  86     case IRP_WRITE_REQUEST:
  87
  88         /*
  89          * ctx is VXHSAIOCB*
  90          * ctx is NULL if error is QNIOERROR_CHANNEL_HUP
  91          */
  92         if (ctx) {
  93             acb = ctx;
  94         } else {
  95             trace_vxhs_iio_callback(error);
  96             goto out;
  97         }
  98
  99         if (error) {
 100             if (!acb->err) {
 101                 acb->err = error;
 102             }
 103             trace_vxhs_iio_callback(error);
 104         }
 105
 106         aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs),
 107                                 vxhs_complete_aio_bh, acb);
 108         break;
 109
 110     default:
 111         if (error == QNIOERROR_HUP) {
 112             /*
 113              * Channel failed, spontaneous notification,
 114              * not in response to I/O
 115              */
 116             trace_vxhs_iio_callback_chnfail(error, errno);
 117         } else {
 118             trace_vxhs_iio_callback_unknwn(opcode, error);
 119         }
 120         break;
 121     }
 122 out:
 123     return;
 124 }
 125
 126 static QemuOptsList runtime_opts = {
 127     .name = "vxhs",
 128     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 129     .desc = {
 130         {
 131             .name = VXHS_OPT_FILENAME,
 132             .type = QEMU_OPT_STRING,
 133             .help = "URI to the Veritas HyperScale image",
 134         },
 135         {
 136             .name = VXHS_OPT_VDISK_ID,
 137             .type = QEMU_OPT_STRING,
 138             .help = "UUID of the VxHS vdisk",
 139         },
 140         {
 141             .name = "tls-creds",
 142             .type = QEMU_OPT_STRING,
 143             .help = "ID of the TLS/SSL credentials to use",
 144         },
 145         { /* end of list */ }
 146     },
 147 };
 148
 149 static QemuOptsList runtime_tcp_opts = {
 150     .name = "vxhs_tcp",
 151     .head = QTAILQ_HEAD_INITIALIZER(runtime_tcp_opts.head),
 152     .desc = {
 153         {
 154             .name = VXHS_OPT_HOST,
 155             .type = QEMU_OPT_STRING,
 156             .help = "host address (ipv4 addresses)",
 157         },
 158         {
 159             .name = VXHS_OPT_PORT,
 160             .type = QEMU_OPT_NUMBER,
 161             .help = "port number on which VxHSD is listening (default 9999)",
 162             .def_value_str = "9999"
 163         },
 164         { /* end of list */ }
 165     },
 166 };
 167
 168 /*
 169  * Parse incoming URI and populate *options with the host
 170  * and device information
 171  */
 172 static int vxhs_parse_uri(const char *filename, QDict *options)
 173 {
 174     URI *uri = NULL;
 175     char *port;
 176     int ret = 0;
 177
 178     trace_vxhs_parse_uri_filename(filename);
 179     uri = uri_parse(filename);
 180     if (!uri || !uri->server || !uri->path) {
 181         uri_free(uri);
 182         return -EINVAL;
 183     }
 184
 185     qdict_put_str(options, VXHS_OPT_SERVER ".host", uri->server);
 186
 187     if (uri->port) {
 188         port = g_strdup_printf("%d", uri->port);
 189         qdict_put_str(options, VXHS_OPT_SERVER ".port", port);
 190         g_free(port);
 191     }
 192
 193     qdict_put_str(options, "vdisk-id", uri->path);
 194
 195     trace_vxhs_parse_uri_hostinfo(uri->server, uri->port);
 196     uri_free(uri);
 197
 198     return ret;
 199 }
 200
 201 static void vxhs_parse_filename(const char *filename, QDict *options,
 202                                 Error **errp)
 203 {
 204     if (qdict_haskey(options, "vdisk-id") || qdict_haskey(options, "server")) {
 205         error_setg(errp, "vdisk-id/server and a file name may not be specified "
 206                          "at the same time");
 207         return;
 208     }
 209
 210     if (strstr(filename, "://")) {
 211         int ret = vxhs_parse_uri(filename, options);
 212         if (ret < 0) {
 213             error_setg(errp, "Invalid URI. URI should be of the form "
 214                        "  vxhs://<host_ip>:<port>/<vdisk-id>");
 215         }
 216     }
 217 }
 218
 219 static void vxhs_refresh_limits(BlockDriverState *bs, Error **errp)
 220 {
 221     /* XXX Does VXHS support AIO on less than 512-byte alignment? */
 222     bs->bl.request_alignment = 512;
 223 }
 224
 225 static int vxhs_init_and_ref(void)
 226 {
 227     if (vxhs_ref++ == 0) {
 228         if (iio_init(QNIO_VERSION, vxhs_iio_callback)) {
 229             return -ENODEV;
 230         }
 231     }
 232     return 0;
 233 }
 234
 235 static void vxhs_unref(void)
 236 {
 237     if (--vxhs_ref == 0) {
 238         iio_fini();
 239     }
 240 }
 241
 242 static void vxhs_get_tls_creds(const char *id, char **cacert,
 243                                char **key, char **cert, Error **errp)
 244 {
 245     Object *obj;
 246     QCryptoTLSCreds *creds;
 247     QCryptoTLSCredsX509 *creds_x509;
 248
 249     obj = object_resolve_path_component(
 250         object_get_objects_root(), id);
 251
 252     if (!obj) {
 253         error_setg(errp, "No TLS credentials with id '%s'",
 254                    id);
 255         return;
 256     }
 257
 258     creds_x509 = (QCryptoTLSCredsX509 *)
 259         object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS_X509);
 260
 261     if (!creds_x509) {
 262         error_setg(errp, "Object with id '%s' is not TLS credentials",
 263                    id);
 264         return;
 265     }
 266
 267     creds = &creds_x509->parent_obj;
 268
 269     if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT) {
 270         error_setg(errp,
 271                    "Expecting TLS credentials with a client endpoint");
 272         return;
 273     }
 274
 275     /*
 276      * Get the cacert, client_cert and client_key file names.
 277      */
 278     if (!creds->dir) {
 279         error_setg(errp, "TLS object missing 'dir' property value");
 280         return;
 281     }
 282
 283     *cacert = g_strdup_printf("%s/%s", creds->dir,
 284                               QCRYPTO_TLS_CREDS_X509_CA_CERT);
 285     *cert = g_strdup_printf("%s/%s", creds->dir,
 286                             QCRYPTO_TLS_CREDS_X509_CLIENT_CERT);
 287     *key = g_strdup_printf("%s/%s", creds->dir,
 288                            QCRYPTO_TLS_CREDS_X509_CLIENT_KEY);
 289 }
 290
 291 static int vxhs_open(BlockDriverState *bs, QDict *options,
 292                      int bdrv_flags, Error **errp)
 293 {
 294     BDRVVXHSState *s = bs->opaque;
 295     void *dev_handlep;
 296     QDict *backing_options = NULL;
 297     QemuOpts *opts = NULL;
 298     QemuOpts *tcp_opts = NULL;
 299     char *of_vsa_addr = NULL;
 300     Error *local_err = NULL;
 301     const char *vdisk_id_opt;
 302     const char *server_host_opt;
 303     int ret = 0;
 304     char *cacert = NULL;
 305     char *client_key = NULL;
 306     char *client_cert = NULL;
 307
 308     ret = vxhs_init_and_ref();
 309     if (ret < 0) {
 310         ret = -EINVAL;
 311         goto out;
 312     }
 313
 314     /* Create opts info from runtime_opts and runtime_tcp_opts list */
 315     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 316     tcp_opts = qemu_opts_create(&runtime_tcp_opts, NULL, 0, &error_abort);
 317
 318     qemu_opts_absorb_qdict(opts, options, &local_err);
 319     if (local_err) {
 320         ret = -EINVAL;
 321         goto out;
 322     }
 323
 324     /* vdisk-id is the disk UUID */
 325     vdisk_id_opt = qemu_opt_get(opts, VXHS_OPT_VDISK_ID);
 326     if (!vdisk_id_opt) {
 327         error_setg(&local_err, QERR_MISSING_PARAMETER, VXHS_OPT_VDISK_ID);
 328         ret = -EINVAL;
 329         goto out;
 330     }
 331
 332     /* vdisk-id may contain a leading '/' */
 333     if (strlen(vdisk_id_opt) > UUID_FMT_LEN + 1) {
 334         error_setg(&local_err, "vdisk-id cannot be more than %d characters",
 335                    UUID_FMT_LEN);
 336         ret = -EINVAL;
 337         goto out;
 338     }
 339
 340     s->vdisk_guid = g_strdup(vdisk_id_opt);
 341     trace_vxhs_open_vdiskid(vdisk_id_opt);
 342
 343     /* get the 'server.' arguments */
 344     qdict_extract_subqdict(options, &backing_options, VXHS_OPT_SERVER".");
 345
 346     qemu_opts_absorb_qdict(tcp_opts, backing_options, &local_err);
 347     if (local_err != NULL) {
 348         ret = -EINVAL;
 349         goto out;
 350     }
 351
 352     server_host_opt = qemu_opt_get(tcp_opts, VXHS_OPT_HOST);
 353     if (!server_host_opt) {
 354         error_setg(&local_err, QERR_MISSING_PARAMETER,
 355                    VXHS_OPT_SERVER"."VXHS_OPT_HOST);
 356         ret = -EINVAL;
 357         goto out;
 358     }
 359
 360     if (strlen(server_host_opt) > MAXHOSTNAMELEN) {
 361         error_setg(&local_err, "server.host cannot be more than %d characters",
 362                    MAXHOSTNAMELEN);
 363         ret = -EINVAL;
 364         goto out;
 365     }
 366
 367     /* check if we got tls-creds via the --object argument */
 368     s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds"));
 369     if (s->tlscredsid) {
 370         vxhs_get_tls_creds(s->tlscredsid, &cacert, &client_key,
 371                            &client_cert, &local_err);
 372         if (local_err != NULL) {
 373             ret = -EINVAL;
 374             goto out;
 375         }
 376         trace_vxhs_get_creds(cacert, client_key, client_cert);
 377     }
 378
 379     s->vdisk_hostinfo.host = g_strdup(server_host_opt);
 380     s->vdisk_hostinfo.port = g_ascii_strtoll(qemu_opt_get(tcp_opts,
 381                                                           VXHS_OPT_PORT),
 382                                                           NULL, 0);
 383
 384     trace_vxhs_open_hostinfo(s->vdisk_hostinfo.host,
 385                              s->vdisk_hostinfo.port);
 386
 387     of_vsa_addr = g_strdup_printf("of://%s:%d",
 388                                   s->vdisk_hostinfo.host,
 389                                   s->vdisk_hostinfo.port);
 390
 391     /*
 392      * Open qnio channel to storage agent if not opened before
 393      */
 394     dev_handlep = iio_open(of_vsa_addr, s->vdisk_guid, 0,
 395                            cacert, client_key, client_cert);
 396     if (dev_handlep == NULL) {
 397         trace_vxhs_open_iio_open(of_vsa_addr);
 398         ret = -ENODEV;
 399         goto out;
 400     }
 401     s->vdisk_hostinfo.dev_handle = dev_handlep;
 402
 403 out:
 404     g_free(of_vsa_addr);
 405     qobject_unref(backing_options);
 406     qemu_opts_del(tcp_opts);
 407     qemu_opts_del(opts);
 408     g_free(cacert);
 409     g_free(client_key);
 410     g_free(client_cert);
 411
 412     if (ret < 0) {
 413         vxhs_unref();
 414         error_propagate(errp, local_err);
 415         g_free(s->vdisk_hostinfo.host);
 416         g_free(s->vdisk_guid);
 417         g_free(s->tlscredsid);
 418         s->vdisk_guid = NULL;
 419     }
 420
 421     return ret;
 422 }
 423
 424 static const AIOCBInfo vxhs_aiocb_info = {
 425     .aiocb_size = sizeof(VXHSAIOCB)
 426 };
 427
 428 /*
 429  * This allocates QEMU-VXHS callback for each IO
 430  * and is passed to QNIO. When QNIO completes the work,
 431  * it will be passed back through the callback.
 432  */
 433 static BlockAIOCB *vxhs_aio_rw(BlockDriverState *bs, uint64_t offset,
 434                                QEMUIOVector *qiov, uint64_t size,
 435                                BlockCompletionFunc *cb, void *opaque,
 436                                VDISKAIOCmd iodir)
 437 {
 438     VXHSAIOCB *acb = NULL;
 439     BDRVVXHSState *s = bs->opaque;
 440     int iio_flags = 0;
 441     int ret = 0;
 442     void *dev_handle = s->vdisk_hostinfo.dev_handle;
 443
 444     acb = qemu_aio_get(&vxhs_aiocb_info, bs, cb, opaque);
 445
 446     /*
 447      * Initialize VXHSAIOCB.
 448      */
 449     acb->err = 0;
 450
 451     iio_flags = IIO_FLAG_ASYNC;
 452
 453     switch (iodir) {
 454     case VDISK_AIO_WRITE:
 455             ret = iio_writev(dev_handle, acb, qiov->iov, qiov->niov,
 456                              offset, size, iio_flags);
 457             break;
 458     case VDISK_AIO_READ:
 459             ret = iio_readv(dev_handle, acb, qiov->iov, qiov->niov,
 460                             offset, size, iio_flags);
 461             break;
 462     default:
 463             trace_vxhs_aio_rw_invalid(iodir);
 464             goto errout;
 465     }
 466
 467     if (ret != 0) {
 468         trace_vxhs_aio_rw_ioerr(s->vdisk_guid, iodir, size, offset,
 469                                 acb, ret, errno);
 470         goto errout;
 471     }
 472     return &acb->common;
 473
 474 errout:
 475     qemu_aio_unref(acb);
 476     return NULL;
 477 }
 478
 479 static BlockAIOCB *vxhs_aio_preadv(BlockDriverState *bs,
 480                                    uint64_t offset, uint64_t bytes,
 481                                    QEMUIOVector *qiov, int flags,
 482                                    BlockCompletionFunc *cb, void *opaque)
 483 {
 484     return vxhs_aio_rw(bs, offset, qiov, bytes, cb, opaque, VDISK_AIO_READ);
 485 }
 486
 487 static BlockAIOCB *vxhs_aio_pwritev(BlockDriverState *bs,
 488                                     uint64_t offset, uint64_t bytes,
 489                                     QEMUIOVector *qiov, int flags,
 490                                     BlockCompletionFunc *cb, void *opaque)
 491 {
 492     return vxhs_aio_rw(bs, offset, qiov, bytes, cb, opaque, VDISK_AIO_WRITE);
 493 }
 494
 495 static void vxhs_close(BlockDriverState *bs)
 496 {
 497     BDRVVXHSState *s = bs->opaque;
 498
 499     trace_vxhs_close(s->vdisk_guid);
 500
 501     g_free(s->vdisk_guid);
 502     s->vdisk_guid = NULL;
 503
 504     /*
 505      * Close vDisk device
 506      */
 507     if (s->vdisk_hostinfo.dev_handle) {
 508         iio_close(s->vdisk_hostinfo.dev_handle);
 509         s->vdisk_hostinfo.dev_handle = NULL;
 510     }
 511
 512     vxhs_unref();
 513
 514     /*
 515      * Free the dynamically allocated host string etc
 516      */
 517     g_free(s->vdisk_hostinfo.host);
 518     g_free(s->tlscredsid);
 519     s->tlscredsid = NULL;
 520     s->vdisk_hostinfo.host = NULL;
 521     s->vdisk_hostinfo.port = 0;
 522 }
 523
 524 static int64_t vxhs_get_vdisk_stat(BDRVVXHSState *s)
 525 {
 526     int64_t vdisk_size = -1;
 527     int ret = 0;
 528     void *dev_handle = s->vdisk_hostinfo.dev_handle;
 529
 530     ret = iio_ioctl(dev_handle, IOR_VDISK_STAT, &vdisk_size, 0);
 531     if (ret < 0) {
 532         trace_vxhs_get_vdisk_stat_err(s->vdisk_guid, ret, errno);
 533         return -EIO;
 534     }
 535
 536     trace_vxhs_get_vdisk_stat(s->vdisk_guid, vdisk_size);
 537     return vdisk_size;
 538 }
 539
 540 /*
 541  * Returns the size of vDisk in bytes. This is required
 542  * by QEMU block upper block layer so that it is visible
 543  * to guest.
 544  */
 545 static int64_t vxhs_getlength(BlockDriverState *bs)
 546 {
 547     BDRVVXHSState *s = bs->opaque;
 548     int64_t vdisk_size;
 549
 550     vdisk_size = vxhs_get_vdisk_stat(s);
 551     if (vdisk_size < 0) {
 552         return -EIO;
 553     }
 554
 555     return vdisk_size;
 556 }
 557
 558 static BlockDriver bdrv_vxhs = {
 559     .format_name                  = "vxhs",
 560     .protocol_name                = "vxhs",
 561     .instance_size                = sizeof(BDRVVXHSState),
 562     .bdrv_file_open               = vxhs_open,
 563     .bdrv_parse_filename          = vxhs_parse_filename,
 564     .bdrv_refresh_limits          = vxhs_refresh_limits,
 565     .bdrv_close                   = vxhs_close,
 566     .bdrv_getlength               = vxhs_getlength,
 567     .bdrv_aio_preadv              = vxhs_aio_preadv,
 568     .bdrv_aio_pwritev             = vxhs_aio_pwritev,
 569 };
 570
 571 static void bdrv_vxhs_init(void)
 572 {
 573     bdrv_register(&bdrv_vxhs);
 574 }
 575
 576 block_init(bdrv_vxhs_init);