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 "qapi/error.h" 18 #include "qemu/config-file.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 XenDevice 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 XenDevice *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_malloc0(sizeof(*ring->sg) * 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_malloc0(sizeof(*ring->sg) * 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 < size) { 204 xen_pv_printf(&xen_9pfs->xendev, 0, "Xen 9pfs request type %d" 205 "needs %zu bytes, buffer has %zu\n", pdu->id, size, 206 buf_size); 207 xen_be_set_state(&xen_9pfs->xendev, XenbusStateClosing); 208 xen_9pfs_disconnect(&xen_9pfs->xendev); 209 } 210 211 *piov = ring->sg; 212 *pniov = num; 213 } 214 215 static void xen_9pfs_push_and_notify(V9fsPDU *pdu) 216 { 217 RING_IDX prod; 218 Xen9pfsDev *priv = container_of(pdu->s, Xen9pfsDev, state); 219 Xen9pfsRing *ring = &priv->rings[pdu->tag % priv->num_rings]; 220 221 g_free(ring->sg); 222 ring->sg = NULL; 223 224 ring->intf->out_cons = ring->out_cons; 225 xen_wmb(); 226 227 prod = ring->intf->in_prod; 228 xen_rmb(); 229 ring->intf->in_prod = prod + pdu->size; 230 xen_wmb(); 231 232 ring->inprogress = false; 233 xenevtchn_notify(ring->evtchndev, ring->local_port); 234 235 qemu_bh_schedule(ring->bh); 236 } 237 238 static const V9fsTransport xen_9p_transport = { 239 .pdu_vmarshal = xen_9pfs_pdu_vmarshal, 240 .pdu_vunmarshal = xen_9pfs_pdu_vunmarshal, 241 .init_in_iov_from_pdu = xen_9pfs_init_in_iov_from_pdu, 242 .init_out_iov_from_pdu = xen_9pfs_init_out_iov_from_pdu, 243 .push_and_notify = xen_9pfs_push_and_notify, 244 }; 245 246 static int xen_9pfs_init(struct XenDevice *xendev) 247 { 248 return 0; 249 } 250 251 static int xen_9pfs_receive(Xen9pfsRing *ring) 252 { 253 P9MsgHeader h; 254 RING_IDX cons, prod, masked_prod, masked_cons, queued; 255 V9fsPDU *pdu; 256 257 if (ring->inprogress) { 258 return 0; 259 } 260 261 cons = ring->intf->out_cons; 262 prod = ring->intf->out_prod; 263 xen_rmb(); 264 265 queued = xen_9pfs_queued(prod, cons, XEN_FLEX_RING_SIZE(ring->ring_order)); 266 if (queued < sizeof(h)) { 267 return 0; 268 } 269 ring->inprogress = true; 270 271 masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order)); 272 masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order)); 273 274 xen_9pfs_read_packet((uint8_t *) &h, ring->ring.out, sizeof(h), 275 masked_prod, &masked_cons, 276 XEN_FLEX_RING_SIZE(ring->ring_order)); 277 if (queued < le32_to_cpu(h.size_le)) { 278 return 0; 279 } 280 281 /* cannot fail, because we only handle one request per ring at a time */ 282 pdu = pdu_alloc(&ring->priv->state); 283 ring->out_size = le32_to_cpu(h.size_le); 284 ring->out_cons = cons + le32_to_cpu(h.size_le); 285 286 pdu_submit(pdu, &h); 287 288 return 0; 289 } 290 291 static void xen_9pfs_bh(void *opaque) 292 { 293 Xen9pfsRing *ring = opaque; 294 xen_9pfs_receive(ring); 295 } 296 297 static void xen_9pfs_evtchn_event(void *opaque) 298 { 299 Xen9pfsRing *ring = opaque; 300 evtchn_port_t port; 301 302 port = xenevtchn_pending(ring->evtchndev); 303 xenevtchn_unmask(ring->evtchndev, port); 304 305 qemu_bh_schedule(ring->bh); 306 } 307 308 static void xen_9pfs_disconnect(struct XenDevice *xendev) 309 { 310 Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); 311 int i; 312 313 for (i = 0; i < xen_9pdev->num_rings; i++) { 314 if (xen_9pdev->rings[i].evtchndev != NULL) { 315 qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev), 316 NULL, NULL, NULL); 317 xenevtchn_unbind(xen_9pdev->rings[i].evtchndev, 318 xen_9pdev->rings[i].local_port); 319 xen_9pdev->rings[i].evtchndev = NULL; 320 } 321 } 322 } 323 324 static int xen_9pfs_free(struct XenDevice *xendev) 325 { 326 Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); 327 int i; 328 329 if (xen_9pdev->rings[0].evtchndev != NULL) { 330 xen_9pfs_disconnect(xendev); 331 } 332 333 for (i = 0; i < xen_9pdev->num_rings; i++) { 334 if (xen_9pdev->rings[i].data != NULL) { 335 xen_be_unmap_grant_refs(&xen_9pdev->xendev, 336 xen_9pdev->rings[i].data, 337 (1 << xen_9pdev->rings[i].ring_order)); 338 } 339 if (xen_9pdev->rings[i].intf != NULL) { 340 xen_be_unmap_grant_refs(&xen_9pdev->xendev, 341 xen_9pdev->rings[i].intf, 342 1); 343 } 344 if (xen_9pdev->rings[i].bh != NULL) { 345 qemu_bh_delete(xen_9pdev->rings[i].bh); 346 } 347 } 348 349 g_free(xen_9pdev->id); 350 g_free(xen_9pdev->tag); 351 g_free(xen_9pdev->path); 352 g_free(xen_9pdev->security_model); 353 g_free(xen_9pdev->rings); 354 return 0; 355 } 356 357 static int xen_9pfs_connect(struct XenDevice *xendev) 358 { 359 Error *err = NULL; 360 int i; 361 Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); 362 V9fsState *s = &xen_9pdev->state; 363 QemuOpts *fsdev; 364 365 if (xenstore_read_fe_int(&xen_9pdev->xendev, "num-rings", 366 &xen_9pdev->num_rings) == -1 || 367 xen_9pdev->num_rings > MAX_RINGS || xen_9pdev->num_rings < 1) { 368 return -1; 369 } 370 371 xen_9pdev->rings = g_malloc0(xen_9pdev->num_rings * sizeof(Xen9pfsRing)); 372 for (i = 0; i < xen_9pdev->num_rings; i++) { 373 char *str; 374 int ring_order; 375 376 xen_9pdev->rings[i].priv = xen_9pdev; 377 xen_9pdev->rings[i].evtchn = -1; 378 xen_9pdev->rings[i].local_port = -1; 379 380 str = g_strdup_printf("ring-ref%u", i); 381 if (xenstore_read_fe_int(&xen_9pdev->xendev, str, 382 &xen_9pdev->rings[i].ref) == -1) { 383 g_free(str); 384 goto out; 385 } 386 g_free(str); 387 str = g_strdup_printf("event-channel-%u", i); 388 if (xenstore_read_fe_int(&xen_9pdev->xendev, str, 389 &xen_9pdev->rings[i].evtchn) == -1) { 390 g_free(str); 391 goto out; 392 } 393 g_free(str); 394 395 xen_9pdev->rings[i].intf = 396 xen_be_map_grant_ref(&xen_9pdev->xendev, 397 xen_9pdev->rings[i].ref, 398 PROT_READ | PROT_WRITE); 399 if (!xen_9pdev->rings[i].intf) { 400 goto out; 401 } 402 ring_order = xen_9pdev->rings[i].intf->ring_order; 403 if (ring_order > MAX_RING_ORDER) { 404 goto out; 405 } 406 xen_9pdev->rings[i].ring_order = ring_order; 407 xen_9pdev->rings[i].data = 408 xen_be_map_grant_refs(&xen_9pdev->xendev, 409 xen_9pdev->rings[i].intf->ref, 410 (1 << ring_order), 411 PROT_READ | PROT_WRITE); 412 if (!xen_9pdev->rings[i].data) { 413 goto out; 414 } 415 xen_9pdev->rings[i].ring.in = xen_9pdev->rings[i].data; 416 xen_9pdev->rings[i].ring.out = xen_9pdev->rings[i].data + 417 XEN_FLEX_RING_SIZE(ring_order); 418 419 xen_9pdev->rings[i].bh = qemu_bh_new(xen_9pfs_bh, &xen_9pdev->rings[i]); 420 xen_9pdev->rings[i].out_cons = 0; 421 xen_9pdev->rings[i].out_size = 0; 422 xen_9pdev->rings[i].inprogress = false; 423 424 425 xen_9pdev->rings[i].evtchndev = xenevtchn_open(NULL, 0); 426 if (xen_9pdev->rings[i].evtchndev == NULL) { 427 goto out; 428 } 429 qemu_set_cloexec(xenevtchn_fd(xen_9pdev->rings[i].evtchndev)); 430 xen_9pdev->rings[i].local_port = xenevtchn_bind_interdomain 431 (xen_9pdev->rings[i].evtchndev, 432 xendev->dom, 433 xen_9pdev->rings[i].evtchn); 434 if (xen_9pdev->rings[i].local_port == -1) { 435 xen_pv_printf(xendev, 0, 436 "xenevtchn_bind_interdomain failed port=%d\n", 437 xen_9pdev->rings[i].evtchn); 438 goto out; 439 } 440 xen_pv_printf(xendev, 2, "bind evtchn port %d\n", xendev->local_port); 441 qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev), 442 xen_9pfs_evtchn_event, NULL, &xen_9pdev->rings[i]); 443 } 444 445 xen_9pdev->security_model = xenstore_read_be_str(xendev, "security_model"); 446 xen_9pdev->path = xenstore_read_be_str(xendev, "path"); 447 xen_9pdev->id = s->fsconf.fsdev_id = 448 g_strdup_printf("xen9p%d", xendev->dev); 449 xen_9pdev->tag = s->fsconf.tag = xenstore_read_fe_str(xendev, "tag"); 450 fsdev = qemu_opts_create(qemu_find_opts("fsdev"), 451 s->fsconf.tag, 452 1, NULL); 453 qemu_opt_set(fsdev, "fsdriver", "local", NULL); 454 qemu_opt_set(fsdev, "path", xen_9pdev->path, NULL); 455 qemu_opt_set(fsdev, "security_model", xen_9pdev->security_model, NULL); 456 qemu_opts_set_id(fsdev, s->fsconf.fsdev_id); 457 qemu_fsdev_add(fsdev, &err); 458 if (err) { 459 error_report_err(err); 460 } 461 v9fs_device_realize_common(s, &xen_9p_transport, NULL); 462 463 return 0; 464 465 out: 466 xen_9pfs_free(xendev); 467 return -1; 468 } 469 470 static void xen_9pfs_alloc(struct XenDevice *xendev) 471 { 472 xenstore_write_be_str(xendev, "versions", VERSIONS); 473 xenstore_write_be_int(xendev, "max-rings", MAX_RINGS); 474 xenstore_write_be_int(xendev, "max-ring-page-order", MAX_RING_ORDER); 475 } 476 477 struct XenDevOps xen_9pfs_ops = { 478 .size = sizeof(Xen9pfsDev), 479 .flags = DEVOPS_FLAG_NEED_GNTDEV, 480 .alloc = xen_9pfs_alloc, 481 .init = xen_9pfs_init, 482 .initialise = xen_9pfs_connect, 483 .disconnect = xen_9pfs_disconnect, 484 .free = xen_9pfs_free, 485 }; 486