hw/9pfs/xen-9p-backend.c

   1 /*
   2  * Xen 9p backend
   3  *
   4  * Copyright Aporeto 2017
   5  *
   6  * Authors:
   7  *  Stefano Stabellini <stefano@aporeto.com>
   8  *
   9  */
  10
  11 #include "qemu/osdep.h"
  12
  13 #include "hw/9pfs/9p.h"
  14 #include "hw/xen/xen-legacy-backend.h"
  15 #include "hw/9pfs/xen-9pfs.h"
  16 #include "qapi/error.h"
  17 #include "qemu/config-file.h"
  18 #include "qemu/main-loop.h"
  19 #include "qemu/option.h"
  20 #include "fsdev/qemu-fsdev.h"
  21
  22 #define VERSIONS "1"
  23 #define MAX_RINGS 8
  24 #define MAX_RING_ORDER 8
  25
  26 typedef struct Xen9pfsRing {
  27     struct Xen9pfsDev *priv;
  28
  29     int ref;
  30     xenevtchn_handle   *evtchndev;
  31     int evtchn;
  32     int local_port;
  33     int ring_order;
  34     struct xen_9pfs_data_intf *intf;
  35     unsigned char *data;
  36     struct xen_9pfs_data ring;
  37
  38     struct iovec *sg;
  39     QEMUBH *bh;
  40
  41     /* local copies, so that we can read/write PDU data directly from
  42      * the ring */
  43     RING_IDX out_cons, out_size, in_cons;
  44     bool inprogress;
  45 } Xen9pfsRing;
  46
  47 typedef struct Xen9pfsDev {
  48     struct XenLegacyDevice xendev;  /* must be first */
  49     V9fsState state;
  50     char *path;
  51     char *security_model;
  52     char *tag;
  53     char *id;
  54
  55     int num_rings;
  56     Xen9pfsRing *rings;
  57 } Xen9pfsDev;
  58
  59 static void xen_9pfs_disconnect(struct XenLegacyDevice *xendev);
  60
  61 static void xen_9pfs_in_sg(Xen9pfsRing *ring,
  62                            struct iovec *in_sg,
  63                            int *num,
  64                            uint32_t idx,
  65                            uint32_t size)
  66 {
  67     RING_IDX cons, prod, masked_prod, masked_cons;
  68
  69     cons = ring->intf->in_cons;
  70     prod = ring->intf->in_prod;
  71     xen_rmb();
  72     masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order));
  73     masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order));
  74
  75     if (masked_prod < masked_cons) {
  76         in_sg[0].iov_base = ring->ring.in + masked_prod;
  77         in_sg[0].iov_len = masked_cons - masked_prod;
  78         *num = 1;
  79     } else {
  80         in_sg[0].iov_base = ring->ring.in + masked_prod;
  81         in_sg[0].iov_len = XEN_FLEX_RING_SIZE(ring->ring_order) - masked_prod;
  82         in_sg[1].iov_base = ring->ring.in;
  83         in_sg[1].iov_len = masked_cons;
  84         *num = 2;
  85     }
  86 }
  87
  88 static void xen_9pfs_out_sg(Xen9pfsRing *ring,
  89                             struct iovec *out_sg,
  90                             int *num,
  91                             uint32_t idx)
  92 {
  93     RING_IDX cons, prod, masked_prod, masked_cons;
  94
  95     cons = ring->intf->out_cons;
  96     prod = ring->intf->out_prod;
  97     xen_rmb();
  98     masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order));
  99     masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order));
 100
 101     if (masked_cons < masked_prod) {
 102         out_sg[0].iov_base = ring->ring.out + masked_cons;
 103         out_sg[0].iov_len = ring->out_size;
 104         *num = 1;
 105     } else {
 106         if (ring->out_size >
 107             (XEN_FLEX_RING_SIZE(ring->ring_order) - masked_cons)) {
 108             out_sg[0].iov_base = ring->ring.out + masked_cons;
 109             out_sg[0].iov_len = XEN_FLEX_RING_SIZE(ring->ring_order) -
 110                                 masked_cons;
 111             out_sg[1].iov_base = ring->ring.out;
 112             out_sg[1].iov_len = ring->out_size -
 113                                 (XEN_FLEX_RING_SIZE(ring->ring_order) -
 114                                  masked_cons);
 115             *num = 2;
 116         } else {
 117             out_sg[0].iov_base = ring->ring.out + masked_cons;
 118             out_sg[0].iov_len = ring->out_size;
 119             *num = 1;
 120         }
 121     }
 122 }
 123
 124 static ssize_t xen_9pfs_pdu_vmarshal(V9fsPDU *pdu,
 125                                      size_t offset,
 126                                      const char *fmt,
 127                                      va_list ap)
 128 {
 129     Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state);
 130     struct iovec in_sg[2];
 131     int num;
 132     ssize_t ret;
 133
 134     xen_9pfs_in_sg(&xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings],
 135                    in_sg, &num, pdu->idx, ROUND_UP(offset + 128, 512));
 136
 137     ret = v9fs_iov_vmarshal(in_sg, num, offset, 0, fmt, ap);
 138     if (ret < 0) {
 139         xen_pv_printf(&xen_9pfs->xendev, 0,
 140                       "Failed to encode VirtFS request type %d\n", pdu->id + 1);
 141         xen_be_set_state(&xen_9pfs->xendev, XenbusStateClosing);
 142         xen_9pfs_disconnect(&xen_9pfs->xendev);
 143     }
 144     return ret;
 145 }
 146
 147 static ssize_t xen_9pfs_pdu_vunmarshal(V9fsPDU *pdu,
 148                                        size_t offset,
 149                                        const char *fmt,
 150                                        va_list ap)
 151 {
 152     Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state);
 153     struct iovec out_sg[2];
 154     int num;
 155     ssize_t ret;
 156
 157     xen_9pfs_out_sg(&xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings],
 158                     out_sg, &num, pdu->idx);
 159
 160     ret = v9fs_iov_vunmarshal(out_sg, num, offset, 0, fmt, ap);
 161     if (ret < 0) {
 162         xen_pv_printf(&xen_9pfs->xendev, 0,
 163                       "Failed to decode VirtFS request type %d\n", pdu->id);
 164         xen_be_set_state(&xen_9pfs->xendev, XenbusStateClosing);
 165         xen_9pfs_disconnect(&xen_9pfs->xendev);
 166     }
 167     return ret;
 168 }
 169
 170 static void xen_9pfs_init_out_iov_from_pdu(V9fsPDU *pdu,
 171                                            struct iovec **piov,
 172                                            unsigned int *pniov,
 173                                            size_t size)
 174 {
 175     Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state);
 176     Xen9pfsRing *ring = &xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings];
 177     int num;
 178
 179     g_free(ring->sg);
 180
 181     ring->sg = g_new0(struct iovec, 2);
 182     xen_9pfs_out_sg(ring, ring->sg, &num, pdu->idx);
 183     *piov = ring->sg;
 184     *pniov = num;
 185 }
 186
 187 static void xen_9pfs_init_in_iov_from_pdu(V9fsPDU *pdu,
 188                                           struct iovec **piov,
 189                                           unsigned int *pniov,
 190                                           size_t *size)
 191 {
 192     Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state);
 193     Xen9pfsRing *ring = &xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings];
 194     int num;
 195     size_t buf_size;
 196
 197     g_free(ring->sg);
 198
 199     ring->sg = g_new0(struct iovec, 2);
 200     xen_9pfs_in_sg(ring, ring->sg, &num, pdu->idx, *size);
 201
 202     buf_size = iov_size(ring->sg, num);
 203     if (buf_size  < P9_IOHDRSZ) {
 204         xen_pv_printf(&xen_9pfs->xendev, 0, "Xen 9pfs request type %d"
 205                 "needs %zu bytes, buffer has %zu, less than minimum\n",
 206                 pdu->id, *size, buf_size);
 207         xen_be_set_state(&xen_9pfs->xendev, XenbusStateClosing);
 208         xen_9pfs_disconnect(&xen_9pfs->xendev);
 209     }
 210     if (buf_size  < *size) {
 211         *size = buf_size;
 212     }
 213
 214     *piov = ring->sg;
 215     *pniov = num;
 216 }
 217
 218 static void xen_9pfs_push_and_notify(V9fsPDU *pdu)
 219 {
 220     RING_IDX prod;
 221     Xen9pfsDev *priv = container_of(pdu->s, Xen9pfsDev, state);
 222     Xen9pfsRing *ring = &priv->rings[pdu->tag % priv->num_rings];
 223
 224     g_free(ring->sg);
 225     ring->sg = NULL;
 226
 227     ring->intf->out_cons = ring->out_cons;
 228     xen_wmb();
 229
 230     prod = ring->intf->in_prod;
 231     xen_rmb();
 232     ring->intf->in_prod = prod + pdu->size;
 233     xen_wmb();
 234
 235     ring->inprogress = false;
 236     xenevtchn_notify(ring->evtchndev, ring->local_port);
 237
 238     qemu_bh_schedule(ring->bh);
 239 }
 240
 241 static const V9fsTransport xen_9p_transport = {
 242     .pdu_vmarshal = xen_9pfs_pdu_vmarshal,
 243     .pdu_vunmarshal = xen_9pfs_pdu_vunmarshal,
 244     .init_in_iov_from_pdu = xen_9pfs_init_in_iov_from_pdu,
 245     .init_out_iov_from_pdu = xen_9pfs_init_out_iov_from_pdu,
 246     .push_and_notify = xen_9pfs_push_and_notify,
 247 };
 248
 249 static int xen_9pfs_init(struct XenLegacyDevice *xendev)
 250 {
 251     return 0;
 252 }
 253
 254 static int xen_9pfs_receive(Xen9pfsRing *ring)
 255 {
 256     P9MsgHeader h;
 257     RING_IDX cons, prod, masked_prod, masked_cons, queued;
 258     V9fsPDU *pdu;
 259
 260     if (ring->inprogress) {
 261         return 0;
 262     }
 263
 264     cons = ring->intf->out_cons;
 265     prod = ring->intf->out_prod;
 266     xen_rmb();
 267
 268     queued = xen_9pfs_queued(prod, cons, XEN_FLEX_RING_SIZE(ring->ring_order));
 269     if (queued < sizeof(h)) {
 270         return 0;
 271     }
 272     ring->inprogress = true;
 273
 274     masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order));
 275     masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order));
 276
 277     xen_9pfs_read_packet((uint8_t *) &h, ring->ring.out, sizeof(h),
 278                          masked_prod, &masked_cons,
 279                          XEN_FLEX_RING_SIZE(ring->ring_order));
 280     if (queued < le32_to_cpu(h.size_le)) {
 281         return 0;
 282     }
 283
 284     /* cannot fail, because we only handle one request per ring at a time */
 285     pdu = pdu_alloc(&ring->priv->state);
 286     ring->out_size = le32_to_cpu(h.size_le);
 287     ring->out_cons = cons + le32_to_cpu(h.size_le);
 288
 289     pdu_submit(pdu, &h);
 290
 291     return 0;
 292 }
 293
 294 static void xen_9pfs_bh(void *opaque)
 295 {
 296     Xen9pfsRing *ring = opaque;
 297     xen_9pfs_receive(ring);
 298 }
 299
 300 static void xen_9pfs_evtchn_event(void *opaque)
 301 {
 302     Xen9pfsRing *ring = opaque;
 303     evtchn_port_t port;
 304
 305     port = xenevtchn_pending(ring->evtchndev);
 306     xenevtchn_unmask(ring->evtchndev, port);
 307
 308     qemu_bh_schedule(ring->bh);
 309 }
 310
 311 static void xen_9pfs_disconnect(struct XenLegacyDevice *xendev)
 312 {
 313     Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev);
 314     int i;
 315
 316     for (i = 0; i < xen_9pdev->num_rings; i++) {
 317         if (xen_9pdev->rings[i].evtchndev != NULL) {
 318             qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev),
 319                     NULL, NULL, NULL);
 320             xenevtchn_unbind(xen_9pdev->rings[i].evtchndev,
 321                              xen_9pdev->rings[i].local_port);
 322             xen_9pdev->rings[i].evtchndev = NULL;
 323         }
 324     }
 325 }
 326
 327 static int xen_9pfs_free(struct XenLegacyDevice *xendev)
 328 {
 329     Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev);
 330     int i;
 331
 332     if (xen_9pdev->rings[0].evtchndev != NULL) {
 333         xen_9pfs_disconnect(xendev);
 334     }
 335
 336     for (i = 0; i < xen_9pdev->num_rings; i++) {
 337         if (xen_9pdev->rings[i].data != NULL) {
 338             xen_be_unmap_grant_refs(&xen_9pdev->xendev,
 339                                     xen_9pdev->rings[i].data,
 340                                     (1 << xen_9pdev->rings[i].ring_order));
 341         }
 342         if (xen_9pdev->rings[i].intf != NULL) {
 343             xen_be_unmap_grant_refs(&xen_9pdev->xendev,
 344                                     xen_9pdev->rings[i].intf,
 345                                     1);
 346         }
 347         if (xen_9pdev->rings[i].bh != NULL) {
 348             qemu_bh_delete(xen_9pdev->rings[i].bh);
 349         }
 350     }
 351
 352     g_free(xen_9pdev->id);
 353     g_free(xen_9pdev->tag);
 354     g_free(xen_9pdev->path);
 355     g_free(xen_9pdev->security_model);
 356     g_free(xen_9pdev->rings);
 357     return 0;
 358 }
 359
 360 static int xen_9pfs_connect(struct XenLegacyDevice *xendev)
 361 {
 362     Error *err = NULL;
 363     int i;
 364     Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev);
 365     V9fsState *s = &xen_9pdev->state;
 366     QemuOpts *fsdev;
 367
 368     if (xenstore_read_fe_int(&xen_9pdev->xendev, "num-rings",
 369                              &xen_9pdev->num_rings) == -1 ||
 370         xen_9pdev->num_rings > MAX_RINGS || xen_9pdev->num_rings < 1) {
 371         return -1;
 372     }
 373
 374     xen_9pdev->rings = g_new0(Xen9pfsRing, xen_9pdev->num_rings);
 375     for (i = 0; i < xen_9pdev->num_rings; i++) {
 376         char *str;
 377         int ring_order;
 378
 379         xen_9pdev->rings[i].priv = xen_9pdev;
 380         xen_9pdev->rings[i].evtchn = -1;
 381         xen_9pdev->rings[i].local_port = -1;
 382
 383         str = g_strdup_printf("ring-ref%u", i);
 384         if (xenstore_read_fe_int(&xen_9pdev->xendev, str,
 385                                  &xen_9pdev->rings[i].ref) == -1) {
 386             g_free(str);
 387             goto out;
 388         }
 389         g_free(str);
 390         str = g_strdup_printf("event-channel-%u", i);
 391         if (xenstore_read_fe_int(&xen_9pdev->xendev, str,
 392                                  &xen_9pdev->rings[i].evtchn) == -1) {
 393             g_free(str);
 394             goto out;
 395         }
 396         g_free(str);
 397
 398         xen_9pdev->rings[i].intf =
 399             xen_be_map_grant_ref(&xen_9pdev->xendev,
 400                                  xen_9pdev->rings[i].ref,
 401                                  PROT_READ | PROT_WRITE);
 402         if (!xen_9pdev->rings[i].intf) {
 403             goto out;
 404         }
 405         ring_order = xen_9pdev->rings[i].intf->ring_order;
 406         if (ring_order > MAX_RING_ORDER) {
 407             goto out;
 408         }
 409         xen_9pdev->rings[i].ring_order = ring_order;
 410         xen_9pdev->rings[i].data =
 411             xen_be_map_grant_refs(&xen_9pdev->xendev,
 412                                   xen_9pdev->rings[i].intf->ref,
 413                                   (1 << ring_order),
 414                                   PROT_READ | PROT_WRITE);
 415         if (!xen_9pdev->rings[i].data) {
 416             goto out;
 417         }
 418         xen_9pdev->rings[i].ring.in = xen_9pdev->rings[i].data;
 419         xen_9pdev->rings[i].ring.out = xen_9pdev->rings[i].data +
 420                                        XEN_FLEX_RING_SIZE(ring_order);
 421
 422         xen_9pdev->rings[i].bh = qemu_bh_new(xen_9pfs_bh, &xen_9pdev->rings[i]);
 423         xen_9pdev->rings[i].out_cons = 0;
 424         xen_9pdev->rings[i].out_size = 0;
 425         xen_9pdev->rings[i].inprogress = false;
 426
 427
 428         xen_9pdev->rings[i].evtchndev = xenevtchn_open(NULL, 0);
 429         if (xen_9pdev->rings[i].evtchndev == NULL) {
 430             goto out;
 431         }
 432         qemu_set_cloexec(xenevtchn_fd(xen_9pdev->rings[i].evtchndev));
 433         xen_9pdev->rings[i].local_port = xenevtchn_bind_interdomain
 434                                             (xen_9pdev->rings[i].evtchndev,
 435                                              xendev->dom,
 436                                              xen_9pdev->rings[i].evtchn);
 437         if (xen_9pdev->rings[i].local_port == -1) {
 438             xen_pv_printf(xendev, 0,
 439                           "xenevtchn_bind_interdomain failed port=%d\n",
 440                           xen_9pdev->rings[i].evtchn);
 441             goto out;
 442         }
 443         xen_pv_printf(xendev, 2, "bind evtchn port %d\n", xendev->local_port);
 444         qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev),
 445                 xen_9pfs_evtchn_event, NULL, &xen_9pdev->rings[i]);
 446     }
 447
 448     xen_9pdev->security_model = xenstore_read_be_str(xendev, "security_model");
 449     xen_9pdev->path = xenstore_read_be_str(xendev, "path");
 450     xen_9pdev->id = s->fsconf.fsdev_id =
 451         g_strdup_printf("xen9p%d", xendev->dev);
 452     xen_9pdev->tag = s->fsconf.tag = xenstore_read_fe_str(xendev, "tag");
 453     fsdev = qemu_opts_create(qemu_find_opts("fsdev"),
 454             s->fsconf.tag,
 455             1, NULL);
 456     qemu_opt_set(fsdev, "fsdriver", "local", NULL);
 457     qemu_opt_set(fsdev, "path", xen_9pdev->path, NULL);
 458     qemu_opt_set(fsdev, "security_model", xen_9pdev->security_model, NULL);
 459     qemu_opts_set_id(fsdev, s->fsconf.fsdev_id);
 460     qemu_fsdev_add(fsdev, &err);
 461     if (err) {
 462         error_report_err(err);
 463     }
 464     v9fs_device_realize_common(s, &xen_9p_transport, NULL);
 465
 466     return 0;
 467
 468 out:
 469     xen_9pfs_free(xendev);
 470     return -1;
 471 }
 472
 473 static void xen_9pfs_alloc(struct XenLegacyDevice *xendev)
 474 {
 475     xenstore_write_be_str(xendev, "versions", VERSIONS);
 476     xenstore_write_be_int(xendev, "max-rings", MAX_RINGS);
 477     xenstore_write_be_int(xendev, "max-ring-page-order", MAX_RING_ORDER);
 478 }
 479
 480 struct XenDevOps xen_9pfs_ops = {
 481     .size       = sizeof(Xen9pfsDev),
 482     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 483     .alloc      = xen_9pfs_alloc,
 484     .init       = xen_9pfs_init,
 485     .initialise = xen_9pfs_connect,
 486     .disconnect = xen_9pfs_disconnect,
 487     .free       = xen_9pfs_free,
 488 };