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/hw.h"
  14 #include "hw/9pfs/9p.h"
  15 #include "hw/xen/xen_backend.h"
  16 #include "hw/9pfs/xen-9pfs.h"
  17 #include "qemu/config-file.h"
  18 #include "fsdev/qemu-fsdev.h"
  19
  20 #define VERSIONS "1"
  21 #define MAX_RINGS 8
  22 #define MAX_RING_ORDER 8
  23
  24 typedef struct Xen9pfsRing {
  25     struct Xen9pfsDev *priv;
  26
  27     int ref;
  28     xenevtchn_handle   *evtchndev;
  29     int evtchn;
  30     int local_port;
  31     int ring_order;
  32     struct xen_9pfs_data_intf *intf;
  33     unsigned char *data;
  34     struct xen_9pfs_data ring;
  35
  36     struct iovec *sg;
  37     QEMUBH *bh;
  38
  39     /* local copies, so that we can read/write PDU data directly from
  40      * the ring */
  41     RING_IDX out_cons, out_size, in_cons;
  42     bool inprogress;
  43 } Xen9pfsRing;
  44
  45 typedef struct Xen9pfsDev {
  46     struct XenDevice xendev;  /* must be first */
  47     V9fsState state;
  48     char *path;
  49     char *security_model;
  50     char *tag;
  51     char *id;
  52
  53     int num_rings;
  54     Xen9pfsRing *rings;
  55 } Xen9pfsDev;
  56
  57 static void xen_9pfs_in_sg(Xen9pfsRing *ring,
  58                            struct iovec *in_sg,
  59                            int *num,
  60                            uint32_t idx,
  61                            uint32_t size)
  62 {
  63     RING_IDX cons, prod, masked_prod, masked_cons;
  64
  65     cons = ring->intf->in_cons;
  66     prod = ring->intf->in_prod;
  67     xen_rmb();
  68     masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order));
  69     masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order));
  70
  71     if (masked_prod < masked_cons) {
  72         in_sg[0].iov_base = ring->ring.in + masked_prod;
  73         in_sg[0].iov_len = masked_cons - masked_prod;
  74         *num = 1;
  75     } else {
  76         in_sg[0].iov_base = ring->ring.in + masked_prod;
  77         in_sg[0].iov_len = XEN_FLEX_RING_SIZE(ring->ring_order) - masked_prod;
  78         in_sg[1].iov_base = ring->ring.in;
  79         in_sg[1].iov_len = masked_cons;
  80         *num = 2;
  81     }
  82 }
  83
  84 static void xen_9pfs_out_sg(Xen9pfsRing *ring,
  85                             struct iovec *out_sg,
  86                             int *num,
  87                             uint32_t idx)
  88 {
  89     RING_IDX cons, prod, masked_prod, masked_cons;
  90
  91     cons = ring->intf->out_cons;
  92     prod = ring->intf->out_prod;
  93     xen_rmb();
  94     masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order));
  95     masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order));
  96
  97     if (masked_cons < masked_prod) {
  98         out_sg[0].iov_base = ring->ring.out + masked_cons;
  99         out_sg[0].iov_len = ring->out_size;
 100         *num = 1;
 101     } else {
 102         if (ring->out_size >
 103             (XEN_FLEX_RING_SIZE(ring->ring_order) - masked_cons)) {
 104             out_sg[0].iov_base = ring->ring.out + masked_cons;
 105             out_sg[0].iov_len = XEN_FLEX_RING_SIZE(ring->ring_order) -
 106                                 masked_cons;
 107             out_sg[1].iov_base = ring->ring.out;
 108             out_sg[1].iov_len = ring->out_size -
 109                                 (XEN_FLEX_RING_SIZE(ring->ring_order) -
 110                                  masked_cons);
 111             *num = 2;
 112         } else {
 113             out_sg[0].iov_base = ring->ring.out + masked_cons;
 114             out_sg[0].iov_len = ring->out_size;
 115             *num = 1;
 116         }
 117     }
 118 }
 119
 120 static ssize_t xen_9pfs_pdu_vmarshal(V9fsPDU *pdu,
 121                                      size_t offset,
 122                                      const char *fmt,
 123                                      va_list ap)
 124 {
 125     Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state);
 126     struct iovec in_sg[2];
 127     int num;
 128
 129     xen_9pfs_in_sg(&xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings],
 130                    in_sg, &num, pdu->idx, ROUND_UP(offset + 128, 512));
 131     return v9fs_iov_vmarshal(in_sg, num, offset, 0, fmt, ap);
 132 }
 133
 134 static ssize_t xen_9pfs_pdu_vunmarshal(V9fsPDU *pdu,
 135                                        size_t offset,
 136                                        const char *fmt,
 137                                        va_list ap)
 138 {
 139     Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state);
 140     struct iovec out_sg[2];
 141     int num;
 142
 143     xen_9pfs_out_sg(&xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings],
 144                     out_sg, &num, pdu->idx);
 145     return v9fs_iov_vunmarshal(out_sg, num, offset, 0, fmt, ap);
 146 }
 147
 148 static void xen_9pfs_init_out_iov_from_pdu(V9fsPDU *pdu,
 149                                            struct iovec **piov,
 150                                            unsigned int *pniov)
 151 {
 152     Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state);
 153     Xen9pfsRing *ring = &xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings];
 154     int num;
 155
 156     g_free(ring->sg);
 157
 158     ring->sg = g_malloc0(sizeof(*ring->sg) * 2);
 159     xen_9pfs_out_sg(ring, ring->sg, &num, pdu->idx);
 160     *piov = ring->sg;
 161     *pniov = num;
 162 }
 163
 164 static void xen_9pfs_init_in_iov_from_pdu(V9fsPDU *pdu,
 165                                           struct iovec **piov,
 166                                           unsigned int *pniov,
 167                                           size_t size)
 168 {
 169     Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state);
 170     Xen9pfsRing *ring = &xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings];
 171     int num;
 172
 173     g_free(ring->sg);
 174
 175     ring->sg = g_malloc0(sizeof(*ring->sg) * 2);
 176     xen_9pfs_in_sg(ring, ring->sg, &num, pdu->idx, size);
 177     *piov = ring->sg;
 178     *pniov = num;
 179 }
 180
 181 static void xen_9pfs_push_and_notify(V9fsPDU *pdu)
 182 {
 183     RING_IDX prod;
 184     Xen9pfsDev *priv = container_of(pdu->s, Xen9pfsDev, state);
 185     Xen9pfsRing *ring = &priv->rings[pdu->tag % priv->num_rings];
 186
 187     g_free(ring->sg);
 188     ring->sg = NULL;
 189
 190     ring->intf->out_cons = ring->out_cons;
 191     xen_wmb();
 192
 193     prod = ring->intf->in_prod;
 194     xen_rmb();
 195     ring->intf->in_prod = prod + pdu->size;
 196     xen_wmb();
 197
 198     ring->inprogress = false;
 199     xenevtchn_notify(ring->evtchndev, ring->local_port);
 200
 201     qemu_bh_schedule(ring->bh);
 202 }
 203
 204 static const struct V9fsTransport xen_9p_transport = {
 205     .pdu_vmarshal = xen_9pfs_pdu_vmarshal,
 206     .pdu_vunmarshal = xen_9pfs_pdu_vunmarshal,
 207     .init_in_iov_from_pdu = xen_9pfs_init_in_iov_from_pdu,
 208     .init_out_iov_from_pdu = xen_9pfs_init_out_iov_from_pdu,
 209     .push_and_notify = xen_9pfs_push_and_notify,
 210 };
 211
 212 static int xen_9pfs_init(struct XenDevice *xendev)
 213 {
 214     return 0;
 215 }
 216
 217 static int xen_9pfs_receive(Xen9pfsRing *ring)
 218 {
 219     P9MsgHeader h;
 220     RING_IDX cons, prod, masked_prod, masked_cons;
 221     V9fsPDU *pdu;
 222
 223     if (ring->inprogress) {
 224         return 0;
 225     }
 226
 227     cons = ring->intf->out_cons;
 228     prod = ring->intf->out_prod;
 229     xen_rmb();
 230
 231     if (xen_9pfs_queued(prod, cons, XEN_FLEX_RING_SIZE(ring->ring_order)) <
 232         sizeof(h)) {
 233         return 0;
 234     }
 235     ring->inprogress = true;
 236
 237     masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order));
 238     masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order));
 239
 240     xen_9pfs_read_packet((uint8_t *) &h, ring->ring.out, sizeof(h),
 241                          masked_prod, &masked_cons,
 242                          XEN_FLEX_RING_SIZE(ring->ring_order));
 243
 244     /* cannot fail, because we only handle one request per ring at a time */
 245     pdu = pdu_alloc(&ring->priv->state);
 246     pdu->size = le32_to_cpu(h.size_le);
 247     pdu->id = h.id;
 248     pdu->tag = le32_to_cpu(h.tag_le);
 249     ring->out_size = le32_to_cpu(h.size_le);
 250     ring->out_cons = cons + le32_to_cpu(h.size_le);
 251
 252     qemu_co_queue_init(&pdu->complete);
 253     pdu_submit(pdu);
 254
 255     return 0;
 256 }
 257
 258 static void xen_9pfs_bh(void *opaque)
 259 {
 260     Xen9pfsRing *ring = opaque;
 261     xen_9pfs_receive(ring);
 262 }
 263
 264 static void xen_9pfs_evtchn_event(void *opaque)
 265 {
 266     Xen9pfsRing *ring = opaque;
 267     evtchn_port_t port;
 268
 269     port = xenevtchn_pending(ring->evtchndev);
 270     xenevtchn_unmask(ring->evtchndev, port);
 271
 272     qemu_bh_schedule(ring->bh);
 273 }
 274
 275 static int xen_9pfs_free(struct XenDevice *xendev)
 276 {
 277     int i;
 278     Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev);
 279
 280     g_free(xen_9pdev->id);
 281     g_free(xen_9pdev->tag);
 282     g_free(xen_9pdev->path);
 283     g_free(xen_9pdev->security_model);
 284
 285     for (i = 0; i < xen_9pdev->num_rings; i++) {
 286         if (xen_9pdev->rings[i].data != NULL) {
 287             xengnttab_unmap(xen_9pdev->xendev.gnttabdev,
 288                     xen_9pdev->rings[i].data,
 289                     (1 << xen_9pdev->rings[i].ring_order));
 290         }
 291         if (xen_9pdev->rings[i].intf != NULL) {
 292             xengnttab_unmap(xen_9pdev->xendev.gnttabdev,
 293                     xen_9pdev->rings[i].intf,
 294                     1);
 295         }
 296         if (xen_9pdev->rings[i].evtchndev > 0) {
 297             qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev),
 298                     NULL, NULL, NULL);
 299             xenevtchn_unbind(xen_9pdev->rings[i].evtchndev,
 300                              xen_9pdev->rings[i].local_port);
 301         }
 302         if (xen_9pdev->rings[i].bh != NULL) {
 303             qemu_bh_delete(xen_9pdev->rings[i].bh);
 304         }
 305     }
 306     g_free(xen_9pdev->rings);
 307     return 0;
 308 }
 309
 310 static int xen_9pfs_connect(struct XenDevice *xendev)
 311 {
 312     int i;
 313     Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev);
 314     V9fsState *s = &xen_9pdev->state;
 315     QemuOpts *fsdev;
 316
 317     if (xenstore_read_fe_int(&xen_9pdev->xendev, "num-rings",
 318                              &xen_9pdev->num_rings) == -1 ||
 319         xen_9pdev->num_rings > MAX_RINGS || xen_9pdev->num_rings < 1) {
 320         return -1;
 321     }
 322
 323     xen_9pdev->rings = g_malloc0(xen_9pdev->num_rings * sizeof(Xen9pfsRing));
 324     for (i = 0; i < xen_9pdev->num_rings; i++) {
 325         char *str;
 326         int ring_order;
 327
 328         xen_9pdev->rings[i].priv = xen_9pdev;
 329         xen_9pdev->rings[i].evtchn = -1;
 330         xen_9pdev->rings[i].local_port = -1;
 331
 332         str = g_strdup_printf("ring-ref%u", i);
 333         if (xenstore_read_fe_int(&xen_9pdev->xendev, str,
 334                                  &xen_9pdev->rings[i].ref) == -1) {
 335             goto out;
 336         }
 337         g_free(str);
 338         str = g_strdup_printf("event-channel-%u", i);
 339         if (xenstore_read_fe_int(&xen_9pdev->xendev, str,
 340                                  &xen_9pdev->rings[i].evtchn) == -1) {
 341             goto out;
 342         }
 343         g_free(str);
 344
 345         xen_9pdev->rings[i].intf =  xengnttab_map_grant_ref(
 346                 xen_9pdev->xendev.gnttabdev,
 347                 xen_9pdev->xendev.dom,
 348                 xen_9pdev->rings[i].ref,
 349                 PROT_READ | PROT_WRITE);
 350         if (!xen_9pdev->rings[i].intf) {
 351             goto out;
 352         }
 353         ring_order = xen_9pdev->rings[i].intf->ring_order;
 354         if (ring_order > MAX_RING_ORDER) {
 355             goto out;
 356         }
 357         xen_9pdev->rings[i].ring_order = ring_order;
 358         xen_9pdev->rings[i].data = xengnttab_map_domain_grant_refs(
 359                 xen_9pdev->xendev.gnttabdev,
 360                 (1 << ring_order),
 361                 xen_9pdev->xendev.dom,
 362                 xen_9pdev->rings[i].intf->ref,
 363                 PROT_READ | PROT_WRITE);
 364         if (!xen_9pdev->rings[i].data) {
 365             goto out;
 366         }
 367         xen_9pdev->rings[i].ring.in = xen_9pdev->rings[i].data;
 368         xen_9pdev->rings[i].ring.out = xen_9pdev->rings[i].data +
 369                                        XEN_FLEX_RING_SIZE(ring_order);
 370
 371         xen_9pdev->rings[i].bh = qemu_bh_new(xen_9pfs_bh, &xen_9pdev->rings[i]);
 372         xen_9pdev->rings[i].out_cons = 0;
 373         xen_9pdev->rings[i].out_size = 0;
 374         xen_9pdev->rings[i].inprogress = false;
 375
 376
 377         xen_9pdev->rings[i].evtchndev = xenevtchn_open(NULL, 0);
 378         if (xen_9pdev->rings[i].evtchndev == NULL) {
 379             goto out;
 380         }
 381         fcntl(xenevtchn_fd(xen_9pdev->rings[i].evtchndev), F_SETFD, FD_CLOEXEC);
 382         xen_9pdev->rings[i].local_port = xenevtchn_bind_interdomain
 383                                             (xen_9pdev->rings[i].evtchndev,
 384                                              xendev->dom,
 385                                              xen_9pdev->rings[i].evtchn);
 386         if (xen_9pdev->rings[i].local_port == -1) {
 387             xen_pv_printf(xendev, 0,
 388                           "xenevtchn_bind_interdomain failed port=%d\n",
 389                           xen_9pdev->rings[i].evtchn);
 390             goto out;
 391         }
 392         xen_pv_printf(xendev, 2, "bind evtchn port %d\n", xendev->local_port);
 393         qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev),
 394                 xen_9pfs_evtchn_event, NULL, &xen_9pdev->rings[i]);
 395     }
 396
 397     xen_9pdev->security_model = xenstore_read_be_str(xendev, "security_model");
 398     xen_9pdev->path = xenstore_read_be_str(xendev, "path");
 399     xen_9pdev->id = s->fsconf.fsdev_id =
 400         g_strdup_printf("xen9p%d", xendev->dev);
 401     xen_9pdev->tag = s->fsconf.tag = xenstore_read_fe_str(xendev, "tag");
 402     v9fs_register_transport(s, &xen_9p_transport);
 403     fsdev = qemu_opts_create(qemu_find_opts("fsdev"),
 404             s->fsconf.tag,
 405             1, NULL);
 406     qemu_opt_set(fsdev, "fsdriver", "local", NULL);
 407     qemu_opt_set(fsdev, "path", xen_9pdev->path, NULL);
 408     qemu_opt_set(fsdev, "security_model", xen_9pdev->security_model, NULL);
 409     qemu_opts_set_id(fsdev, s->fsconf.fsdev_id);
 410     qemu_fsdev_add(fsdev);
 411     v9fs_device_realize_common(s, NULL);
 412
 413     return 0;
 414
 415 out:
 416     xen_9pfs_free(xendev);
 417     return -1;
 418 }
 419
 420 static void xen_9pfs_alloc(struct XenDevice *xendev)
 421 {
 422     xenstore_write_be_str(xendev, "versions", VERSIONS);
 423     xenstore_write_be_int(xendev, "max-rings", MAX_RINGS);
 424     xenstore_write_be_int(xendev, "max-ring-page-order", MAX_RING_ORDER);
 425 }
 426
 427 static void xen_9pfs_disconnect(struct XenDevice *xendev)
 428 {
 429     /* Dynamic hotplug of PV filesystems at runtime is not supported. */
 430 }
 431
 432 struct XenDevOps xen_9pfs_ops = {
 433     .size       = sizeof(Xen9pfsDev),
 434     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 435     .alloc      = xen_9pfs_alloc,
 436     .init       = xen_9pfs_init,
 437     .initialise = xen_9pfs_connect,
 438     .disconnect = xen_9pfs_disconnect,
 439     .free       = xen_9pfs_free,
 440 };