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 reply type %d\n", 141 pdu->id + 1); 142 xen_be_set_state(&xen_9pfs->xendev, XenbusStateClosing); 143 xen_9pfs_disconnect(&xen_9pfs->xendev); 144 } 145 return ret; 146 } 147 148 static ssize_t xen_9pfs_pdu_vunmarshal(V9fsPDU *pdu, 149 size_t offset, 150 const char *fmt, 151 va_list ap) 152 { 153 Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state); 154 struct iovec out_sg[2]; 155 int num; 156 ssize_t ret; 157 158 xen_9pfs_out_sg(&xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings], 159 out_sg, &num, pdu->idx); 160 161 ret = v9fs_iov_vunmarshal(out_sg, num, offset, 0, fmt, ap); 162 if (ret < 0) { 163 xen_pv_printf(&xen_9pfs->xendev, 0, 164 "Failed to decode VirtFS request type %d\n", pdu->id); 165 xen_be_set_state(&xen_9pfs->xendev, XenbusStateClosing); 166 xen_9pfs_disconnect(&xen_9pfs->xendev); 167 } 168 return ret; 169 } 170 171 static void xen_9pfs_init_out_iov_from_pdu(V9fsPDU *pdu, 172 struct iovec **piov, 173 unsigned int *pniov, 174 size_t size) 175 { 176 Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state); 177 Xen9pfsRing *ring = &xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings]; 178 int num; 179 180 g_free(ring->sg); 181 182 ring->sg = g_new0(struct iovec, 2); 183 xen_9pfs_out_sg(ring, ring->sg, &num, pdu->idx); 184 *piov = ring->sg; 185 *pniov = num; 186 } 187 188 static void xen_9pfs_init_in_iov_from_pdu(V9fsPDU *pdu, 189 struct iovec **piov, 190 unsigned int *pniov, 191 size_t *size) 192 { 193 Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state); 194 Xen9pfsRing *ring = &xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings]; 195 int num; 196 size_t buf_size; 197 198 g_free(ring->sg); 199 200 ring->sg = g_new0(struct iovec, 2); 201 xen_9pfs_in_sg(ring, ring->sg, &num, pdu->idx, *size); 202 203 buf_size = iov_size(ring->sg, num); 204 if (buf_size < P9_IOHDRSZ) { 205 xen_pv_printf(&xen_9pfs->xendev, 0, "Xen 9pfs reply type %d needs " 206 "%zu bytes, buffer has %zu, less than minimum\n", 207 pdu->id + 1, *size, buf_size); 208 xen_be_set_state(&xen_9pfs->xendev, XenbusStateClosing); 209 xen_9pfs_disconnect(&xen_9pfs->xendev); 210 } 211 if (buf_size < *size) { 212 *size = buf_size; 213 } 214 215 *piov = ring->sg; 216 *pniov = num; 217 } 218 219 static void xen_9pfs_push_and_notify(V9fsPDU *pdu) 220 { 221 RING_IDX prod; 222 Xen9pfsDev *priv = container_of(pdu->s, Xen9pfsDev, state); 223 Xen9pfsRing *ring = &priv->rings[pdu->tag % priv->num_rings]; 224 225 g_free(ring->sg); 226 ring->sg = NULL; 227 228 ring->intf->out_cons = ring->out_cons; 229 xen_wmb(); 230 231 prod = ring->intf->in_prod; 232 xen_rmb(); 233 ring->intf->in_prod = prod + pdu->size; 234 xen_wmb(); 235 236 ring->inprogress = false; 237 xenevtchn_notify(ring->evtchndev, ring->local_port); 238 239 qemu_bh_schedule(ring->bh); 240 } 241 242 static const V9fsTransport xen_9p_transport = { 243 .pdu_vmarshal = xen_9pfs_pdu_vmarshal, 244 .pdu_vunmarshal = xen_9pfs_pdu_vunmarshal, 245 .init_in_iov_from_pdu = xen_9pfs_init_in_iov_from_pdu, 246 .init_out_iov_from_pdu = xen_9pfs_init_out_iov_from_pdu, 247 .push_and_notify = xen_9pfs_push_and_notify, 248 }; 249 250 static int xen_9pfs_init(struct XenLegacyDevice *xendev) 251 { 252 return 0; 253 } 254 255 static int xen_9pfs_receive(Xen9pfsRing *ring) 256 { 257 P9MsgHeader h; 258 RING_IDX cons, prod, masked_prod, masked_cons, queued; 259 V9fsPDU *pdu; 260 261 if (ring->inprogress) { 262 return 0; 263 } 264 265 cons = ring->intf->out_cons; 266 prod = ring->intf->out_prod; 267 xen_rmb(); 268 269 queued = xen_9pfs_queued(prod, cons, XEN_FLEX_RING_SIZE(ring->ring_order)); 270 if (queued < sizeof(h)) { 271 return 0; 272 } 273 ring->inprogress = true; 274 275 masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order)); 276 masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order)); 277 278 xen_9pfs_read_packet((uint8_t *) &h, ring->ring.out, sizeof(h), 279 masked_prod, &masked_cons, 280 XEN_FLEX_RING_SIZE(ring->ring_order)); 281 if (queued < le32_to_cpu(h.size_le)) { 282 return 0; 283 } 284 285 /* cannot fail, because we only handle one request per ring at a time */ 286 pdu = pdu_alloc(&ring->priv->state); 287 ring->out_size = le32_to_cpu(h.size_le); 288 ring->out_cons = cons + le32_to_cpu(h.size_le); 289 290 pdu_submit(pdu, &h); 291 292 return 0; 293 } 294 295 static void xen_9pfs_bh(void *opaque) 296 { 297 Xen9pfsRing *ring = opaque; 298 xen_9pfs_receive(ring); 299 } 300 301 static void xen_9pfs_evtchn_event(void *opaque) 302 { 303 Xen9pfsRing *ring = opaque; 304 evtchn_port_t port; 305 306 port = xenevtchn_pending(ring->evtchndev); 307 xenevtchn_unmask(ring->evtchndev, port); 308 309 qemu_bh_schedule(ring->bh); 310 } 311 312 static void xen_9pfs_disconnect(struct XenLegacyDevice *xendev) 313 { 314 Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); 315 int i; 316 317 for (i = 0; i < xen_9pdev->num_rings; i++) { 318 if (xen_9pdev->rings[i].evtchndev != NULL) { 319 qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev), 320 NULL, NULL, NULL); 321 xenevtchn_unbind(xen_9pdev->rings[i].evtchndev, 322 xen_9pdev->rings[i].local_port); 323 xen_9pdev->rings[i].evtchndev = NULL; 324 } 325 } 326 } 327 328 static int xen_9pfs_free(struct XenLegacyDevice *xendev) 329 { 330 Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); 331 int i; 332 333 if (xen_9pdev->rings[0].evtchndev != NULL) { 334 xen_9pfs_disconnect(xendev); 335 } 336 337 for (i = 0; i < xen_9pdev->num_rings; i++) { 338 if (xen_9pdev->rings[i].data != NULL) { 339 xen_be_unmap_grant_refs(&xen_9pdev->xendev, 340 xen_9pdev->rings[i].data, 341 (1 << xen_9pdev->rings[i].ring_order)); 342 } 343 if (xen_9pdev->rings[i].intf != NULL) { 344 xen_be_unmap_grant_refs(&xen_9pdev->xendev, 345 xen_9pdev->rings[i].intf, 346 1); 347 } 348 if (xen_9pdev->rings[i].bh != NULL) { 349 qemu_bh_delete(xen_9pdev->rings[i].bh); 350 } 351 } 352 353 g_free(xen_9pdev->id); 354 g_free(xen_9pdev->tag); 355 g_free(xen_9pdev->path); 356 g_free(xen_9pdev->security_model); 357 g_free(xen_9pdev->rings); 358 return 0; 359 } 360 361 static int xen_9pfs_connect(struct XenLegacyDevice *xendev) 362 { 363 Error *err = NULL; 364 int i; 365 Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); 366 V9fsState *s = &xen_9pdev->state; 367 QemuOpts *fsdev; 368 369 if (xenstore_read_fe_int(&xen_9pdev->xendev, "num-rings", 370 &xen_9pdev->num_rings) == -1 || 371 xen_9pdev->num_rings > MAX_RINGS || xen_9pdev->num_rings < 1) { 372 return -1; 373 } 374 375 xen_9pdev->rings = g_new0(Xen9pfsRing, xen_9pdev->num_rings); 376 for (i = 0; i < xen_9pdev->num_rings; i++) { 377 char *str; 378 int ring_order; 379 380 xen_9pdev->rings[i].priv = xen_9pdev; 381 xen_9pdev->rings[i].evtchn = -1; 382 xen_9pdev->rings[i].local_port = -1; 383 384 str = g_strdup_printf("ring-ref%u", i); 385 if (xenstore_read_fe_int(&xen_9pdev->xendev, str, 386 &xen_9pdev->rings[i].ref) == -1) { 387 g_free(str); 388 goto out; 389 } 390 g_free(str); 391 str = g_strdup_printf("event-channel-%u", i); 392 if (xenstore_read_fe_int(&xen_9pdev->xendev, str, 393 &xen_9pdev->rings[i].evtchn) == -1) { 394 g_free(str); 395 goto out; 396 } 397 g_free(str); 398 399 xen_9pdev->rings[i].intf = 400 xen_be_map_grant_ref(&xen_9pdev->xendev, 401 xen_9pdev->rings[i].ref, 402 PROT_READ | PROT_WRITE); 403 if (!xen_9pdev->rings[i].intf) { 404 goto out; 405 } 406 ring_order = xen_9pdev->rings[i].intf->ring_order; 407 if (ring_order > MAX_RING_ORDER) { 408 goto out; 409 } 410 xen_9pdev->rings[i].ring_order = ring_order; 411 xen_9pdev->rings[i].data = 412 xen_be_map_grant_refs(&xen_9pdev->xendev, 413 xen_9pdev->rings[i].intf->ref, 414 (1 << ring_order), 415 PROT_READ | PROT_WRITE); 416 if (!xen_9pdev->rings[i].data) { 417 goto out; 418 } 419 xen_9pdev->rings[i].ring.in = xen_9pdev->rings[i].data; 420 xen_9pdev->rings[i].ring.out = xen_9pdev->rings[i].data + 421 XEN_FLEX_RING_SIZE(ring_order); 422 423 xen_9pdev->rings[i].bh = qemu_bh_new(xen_9pfs_bh, &xen_9pdev->rings[i]); 424 xen_9pdev->rings[i].out_cons = 0; 425 xen_9pdev->rings[i].out_size = 0; 426 xen_9pdev->rings[i].inprogress = false; 427 428 429 xen_9pdev->rings[i].evtchndev = xenevtchn_open(NULL, 0); 430 if (xen_9pdev->rings[i].evtchndev == NULL) { 431 goto out; 432 } 433 qemu_set_cloexec(xenevtchn_fd(xen_9pdev->rings[i].evtchndev)); 434 xen_9pdev->rings[i].local_port = xenevtchn_bind_interdomain 435 (xen_9pdev->rings[i].evtchndev, 436 xendev->dom, 437 xen_9pdev->rings[i].evtchn); 438 if (xen_9pdev->rings[i].local_port == -1) { 439 xen_pv_printf(xendev, 0, 440 "xenevtchn_bind_interdomain failed port=%d\n", 441 xen_9pdev->rings[i].evtchn); 442 goto out; 443 } 444 xen_pv_printf(xendev, 2, "bind evtchn port %d\n", xendev->local_port); 445 qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev), 446 xen_9pfs_evtchn_event, NULL, &xen_9pdev->rings[i]); 447 } 448 449 xen_9pdev->security_model = xenstore_read_be_str(xendev, "security_model"); 450 xen_9pdev->path = xenstore_read_be_str(xendev, "path"); 451 xen_9pdev->id = s->fsconf.fsdev_id = 452 g_strdup_printf("xen9p%d", xendev->dev); 453 xen_9pdev->tag = s->fsconf.tag = xenstore_read_fe_str(xendev, "tag"); 454 fsdev = qemu_opts_create(qemu_find_opts("fsdev"), 455 s->fsconf.tag, 456 1, NULL); 457 qemu_opt_set(fsdev, "fsdriver", "local", NULL); 458 qemu_opt_set(fsdev, "path", xen_9pdev->path, NULL); 459 qemu_opt_set(fsdev, "security_model", xen_9pdev->security_model, NULL); 460 qemu_opts_set_id(fsdev, s->fsconf.fsdev_id); 461 qemu_fsdev_add(fsdev, &err); 462 if (err) { 463 error_report_err(err); 464 } 465 v9fs_device_realize_common(s, &xen_9p_transport, NULL); 466 467 return 0; 468 469 out: 470 xen_9pfs_free(xendev); 471 return -1; 472 } 473 474 static void xen_9pfs_alloc(struct XenLegacyDevice *xendev) 475 { 476 xenstore_write_be_str(xendev, "versions", VERSIONS); 477 xenstore_write_be_int(xendev, "max-rings", MAX_RINGS); 478 xenstore_write_be_int(xendev, "max-ring-page-order", MAX_RING_ORDER); 479 } 480 481 struct XenDevOps xen_9pfs_ops = { 482 .size = sizeof(Xen9pfsDev), 483 .flags = DEVOPS_FLAG_NEED_GNTDEV, 484 .alloc = xen_9pfs_alloc, 485 .init = xen_9pfs_init, 486 .initialise = xen_9pfs_connect, 487 .disconnect = xen_9pfs_disconnect, 488 .free = xen_9pfs_free, 489 }; 490