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 9 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 Coroutine *co; 41 42 /* local copies, so that we can read/write PDU data directly from 43 * the ring */ 44 RING_IDX out_cons, out_size, in_cons; 45 bool inprogress; 46 } Xen9pfsRing; 47 48 typedef struct Xen9pfsDev { 49 struct XenLegacyDevice xendev; /* must be first */ 50 V9fsState state; 51 char *path; 52 char *security_model; 53 char *tag; 54 char *id; 55 56 int num_rings; 57 Xen9pfsRing *rings; 58 } Xen9pfsDev; 59 60 static void xen_9pfs_disconnect(struct XenLegacyDevice *xendev); 61 62 static void xen_9pfs_in_sg(Xen9pfsRing *ring, 63 struct iovec *in_sg, 64 int *num, 65 uint32_t idx, 66 uint32_t size) 67 { 68 RING_IDX cons, prod, masked_prod, masked_cons; 69 70 cons = ring->intf->in_cons; 71 prod = ring->intf->in_prod; 72 xen_rmb(); 73 masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order)); 74 masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order)); 75 76 if (masked_prod < masked_cons) { 77 in_sg[0].iov_base = ring->ring.in + masked_prod; 78 in_sg[0].iov_len = masked_cons - masked_prod; 79 *num = 1; 80 } else { 81 in_sg[0].iov_base = ring->ring.in + masked_prod; 82 in_sg[0].iov_len = XEN_FLEX_RING_SIZE(ring->ring_order) - masked_prod; 83 in_sg[1].iov_base = ring->ring.in; 84 in_sg[1].iov_len = masked_cons; 85 *num = 2; 86 } 87 } 88 89 static void xen_9pfs_out_sg(Xen9pfsRing *ring, 90 struct iovec *out_sg, 91 int *num, 92 uint32_t idx) 93 { 94 RING_IDX cons, prod, masked_prod, masked_cons; 95 96 cons = ring->intf->out_cons; 97 prod = ring->intf->out_prod; 98 xen_rmb(); 99 masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order)); 100 masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order)); 101 102 if (masked_cons < masked_prod) { 103 out_sg[0].iov_base = ring->ring.out + masked_cons; 104 out_sg[0].iov_len = ring->out_size; 105 *num = 1; 106 } else { 107 if (ring->out_size > 108 (XEN_FLEX_RING_SIZE(ring->ring_order) - masked_cons)) { 109 out_sg[0].iov_base = ring->ring.out + masked_cons; 110 out_sg[0].iov_len = XEN_FLEX_RING_SIZE(ring->ring_order) - 111 masked_cons; 112 out_sg[1].iov_base = ring->ring.out; 113 out_sg[1].iov_len = ring->out_size - 114 (XEN_FLEX_RING_SIZE(ring->ring_order) - 115 masked_cons); 116 *num = 2; 117 } else { 118 out_sg[0].iov_base = ring->ring.out + masked_cons; 119 out_sg[0].iov_len = ring->out_size; 120 *num = 1; 121 } 122 } 123 } 124 125 static ssize_t xen_9pfs_pdu_vmarshal(V9fsPDU *pdu, 126 size_t offset, 127 const char *fmt, 128 va_list ap) 129 { 130 Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state); 131 struct iovec in_sg[2]; 132 int num; 133 ssize_t ret; 134 135 xen_9pfs_in_sg(&xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings], 136 in_sg, &num, pdu->idx, ROUND_UP(offset + 128, 512)); 137 138 ret = v9fs_iov_vmarshal(in_sg, num, offset, 0, fmt, ap); 139 if (ret < 0) { 140 xen_pv_printf(&xen_9pfs->xendev, 0, 141 "Failed to encode VirtFS reply type %d\n", 142 pdu->id + 1); 143 xen_be_set_state(&xen_9pfs->xendev, XenbusStateClosing); 144 xen_9pfs_disconnect(&xen_9pfs->xendev); 145 } 146 return ret; 147 } 148 149 static ssize_t xen_9pfs_pdu_vunmarshal(V9fsPDU *pdu, 150 size_t offset, 151 const char *fmt, 152 va_list ap) 153 { 154 Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state); 155 struct iovec out_sg[2]; 156 int num; 157 ssize_t ret; 158 159 xen_9pfs_out_sg(&xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings], 160 out_sg, &num, pdu->idx); 161 162 ret = v9fs_iov_vunmarshal(out_sg, num, offset, 0, fmt, ap); 163 if (ret < 0) { 164 xen_pv_printf(&xen_9pfs->xendev, 0, 165 "Failed to decode VirtFS request type %d\n", pdu->id); 166 xen_be_set_state(&xen_9pfs->xendev, XenbusStateClosing); 167 xen_9pfs_disconnect(&xen_9pfs->xendev); 168 } 169 return ret; 170 } 171 172 static void xen_9pfs_init_out_iov_from_pdu(V9fsPDU *pdu, 173 struct iovec **piov, 174 unsigned int *pniov, 175 size_t size) 176 { 177 Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state); 178 Xen9pfsRing *ring = &xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings]; 179 int num; 180 181 g_free(ring->sg); 182 183 ring->sg = g_new0(struct iovec, 2); 184 xen_9pfs_out_sg(ring, ring->sg, &num, pdu->idx); 185 *piov = ring->sg; 186 *pniov = num; 187 } 188 189 static void xen_9pfs_init_in_iov_from_pdu(V9fsPDU *pdu, 190 struct iovec **piov, 191 unsigned int *pniov, 192 size_t size) 193 { 194 Xen9pfsDev *xen_9pfs = container_of(pdu->s, Xen9pfsDev, state); 195 Xen9pfsRing *ring = &xen_9pfs->rings[pdu->tag % xen_9pfs->num_rings]; 196 int num; 197 size_t buf_size; 198 199 g_free(ring->sg); 200 201 ring->sg = g_new0(struct iovec, 2); 202 ring->co = qemu_coroutine_self(); 203 /* make sure other threads see ring->co changes before continuing */ 204 smp_wmb(); 205 206 again: 207 xen_9pfs_in_sg(ring, ring->sg, &num, pdu->idx, size); 208 buf_size = iov_size(ring->sg, num); 209 if (buf_size < size) { 210 qemu_coroutine_yield(); 211 goto again; 212 } 213 ring->co = NULL; 214 /* make sure other threads see ring->co changes before continuing */ 215 smp_wmb(); 216 217 *piov = ring->sg; 218 *pniov = num; 219 } 220 221 static void xen_9pfs_push_and_notify(V9fsPDU *pdu) 222 { 223 RING_IDX prod; 224 Xen9pfsDev *priv = container_of(pdu->s, Xen9pfsDev, state); 225 Xen9pfsRing *ring = &priv->rings[pdu->tag % priv->num_rings]; 226 227 g_free(ring->sg); 228 ring->sg = NULL; 229 230 ring->intf->out_cons = ring->out_cons; 231 xen_wmb(); 232 233 prod = ring->intf->in_prod; 234 xen_rmb(); 235 ring->intf->in_prod = prod + pdu->size; 236 xen_wmb(); 237 238 ring->inprogress = false; 239 xenevtchn_notify(ring->evtchndev, ring->local_port); 240 241 qemu_bh_schedule(ring->bh); 242 } 243 244 static const V9fsTransport xen_9p_transport = { 245 .pdu_vmarshal = xen_9pfs_pdu_vmarshal, 246 .pdu_vunmarshal = xen_9pfs_pdu_vunmarshal, 247 .init_in_iov_from_pdu = xen_9pfs_init_in_iov_from_pdu, 248 .init_out_iov_from_pdu = xen_9pfs_init_out_iov_from_pdu, 249 .push_and_notify = xen_9pfs_push_and_notify, 250 }; 251 252 static int xen_9pfs_init(struct XenLegacyDevice *xendev) 253 { 254 return 0; 255 } 256 257 static int xen_9pfs_receive(Xen9pfsRing *ring) 258 { 259 P9MsgHeader h; 260 RING_IDX cons, prod, masked_prod, masked_cons, queued; 261 V9fsPDU *pdu; 262 263 if (ring->inprogress) { 264 return 0; 265 } 266 267 cons = ring->intf->out_cons; 268 prod = ring->intf->out_prod; 269 xen_rmb(); 270 271 queued = xen_9pfs_queued(prod, cons, XEN_FLEX_RING_SIZE(ring->ring_order)); 272 if (queued < sizeof(h)) { 273 return 0; 274 } 275 ring->inprogress = true; 276 277 masked_prod = xen_9pfs_mask(prod, XEN_FLEX_RING_SIZE(ring->ring_order)); 278 masked_cons = xen_9pfs_mask(cons, XEN_FLEX_RING_SIZE(ring->ring_order)); 279 280 xen_9pfs_read_packet((uint8_t *) &h, ring->ring.out, sizeof(h), 281 masked_prod, &masked_cons, 282 XEN_FLEX_RING_SIZE(ring->ring_order)); 283 if (queued < le32_to_cpu(h.size_le)) { 284 return 0; 285 } 286 287 /* cannot fail, because we only handle one request per ring at a time */ 288 pdu = pdu_alloc(&ring->priv->state); 289 ring->out_size = le32_to_cpu(h.size_le); 290 ring->out_cons = cons + le32_to_cpu(h.size_le); 291 292 pdu_submit(pdu, &h); 293 294 return 0; 295 } 296 297 static void xen_9pfs_bh(void *opaque) 298 { 299 Xen9pfsRing *ring = opaque; 300 bool wait; 301 302 again: 303 wait = ring->co != NULL && qemu_coroutine_entered(ring->co); 304 /* paired with the smb_wmb barriers in xen_9pfs_init_in_iov_from_pdu */ 305 smp_rmb(); 306 if (wait) { 307 cpu_relax(); 308 goto again; 309 } 310 311 if (ring->co != NULL) { 312 qemu_coroutine_enter_if_inactive(ring->co); 313 } 314 xen_9pfs_receive(ring); 315 } 316 317 static void xen_9pfs_evtchn_event(void *opaque) 318 { 319 Xen9pfsRing *ring = opaque; 320 evtchn_port_t port; 321 322 port = xenevtchn_pending(ring->evtchndev); 323 xenevtchn_unmask(ring->evtchndev, port); 324 325 qemu_bh_schedule(ring->bh); 326 } 327 328 static void xen_9pfs_disconnect(struct XenLegacyDevice *xendev) 329 { 330 Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); 331 int i; 332 333 for (i = 0; i < xen_9pdev->num_rings; i++) { 334 if (xen_9pdev->rings[i].evtchndev != NULL) { 335 qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev), 336 NULL, NULL, NULL); 337 xenevtchn_unbind(xen_9pdev->rings[i].evtchndev, 338 xen_9pdev->rings[i].local_port); 339 xen_9pdev->rings[i].evtchndev = NULL; 340 } 341 } 342 } 343 344 static int xen_9pfs_free(struct XenLegacyDevice *xendev) 345 { 346 Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); 347 int i; 348 349 if (xen_9pdev->rings[0].evtchndev != NULL) { 350 xen_9pfs_disconnect(xendev); 351 } 352 353 for (i = 0; i < xen_9pdev->num_rings; i++) { 354 if (xen_9pdev->rings[i].data != NULL) { 355 xen_be_unmap_grant_refs(&xen_9pdev->xendev, 356 xen_9pdev->rings[i].data, 357 (1 << xen_9pdev->rings[i].ring_order)); 358 } 359 if (xen_9pdev->rings[i].intf != NULL) { 360 xen_be_unmap_grant_refs(&xen_9pdev->xendev, 361 xen_9pdev->rings[i].intf, 362 1); 363 } 364 if (xen_9pdev->rings[i].bh != NULL) { 365 qemu_bh_delete(xen_9pdev->rings[i].bh); 366 } 367 } 368 369 g_free(xen_9pdev->id); 370 g_free(xen_9pdev->tag); 371 g_free(xen_9pdev->path); 372 g_free(xen_9pdev->security_model); 373 g_free(xen_9pdev->rings); 374 return 0; 375 } 376 377 static int xen_9pfs_connect(struct XenLegacyDevice *xendev) 378 { 379 Error *err = NULL; 380 int i; 381 Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); 382 V9fsState *s = &xen_9pdev->state; 383 QemuOpts *fsdev; 384 385 if (xenstore_read_fe_int(&xen_9pdev->xendev, "num-rings", 386 &xen_9pdev->num_rings) == -1 || 387 xen_9pdev->num_rings > MAX_RINGS || xen_9pdev->num_rings < 1) { 388 return -1; 389 } 390 391 xen_9pdev->rings = g_new0(Xen9pfsRing, xen_9pdev->num_rings); 392 for (i = 0; i < xen_9pdev->num_rings; i++) { 393 char *str; 394 int ring_order; 395 396 xen_9pdev->rings[i].priv = xen_9pdev; 397 xen_9pdev->rings[i].evtchn = -1; 398 xen_9pdev->rings[i].local_port = -1; 399 400 str = g_strdup_printf("ring-ref%u", i); 401 if (xenstore_read_fe_int(&xen_9pdev->xendev, str, 402 &xen_9pdev->rings[i].ref) == -1) { 403 g_free(str); 404 goto out; 405 } 406 g_free(str); 407 str = g_strdup_printf("event-channel-%u", i); 408 if (xenstore_read_fe_int(&xen_9pdev->xendev, str, 409 &xen_9pdev->rings[i].evtchn) == -1) { 410 g_free(str); 411 goto out; 412 } 413 g_free(str); 414 415 xen_9pdev->rings[i].intf = 416 xen_be_map_grant_ref(&xen_9pdev->xendev, 417 xen_9pdev->rings[i].ref, 418 PROT_READ | PROT_WRITE); 419 if (!xen_9pdev->rings[i].intf) { 420 goto out; 421 } 422 ring_order = xen_9pdev->rings[i].intf->ring_order; 423 if (ring_order > MAX_RING_ORDER) { 424 goto out; 425 } 426 xen_9pdev->rings[i].ring_order = ring_order; 427 xen_9pdev->rings[i].data = 428 xen_be_map_grant_refs(&xen_9pdev->xendev, 429 xen_9pdev->rings[i].intf->ref, 430 (1 << ring_order), 431 PROT_READ | PROT_WRITE); 432 if (!xen_9pdev->rings[i].data) { 433 goto out; 434 } 435 xen_9pdev->rings[i].ring.in = xen_9pdev->rings[i].data; 436 xen_9pdev->rings[i].ring.out = xen_9pdev->rings[i].data + 437 XEN_FLEX_RING_SIZE(ring_order); 438 439 xen_9pdev->rings[i].bh = qemu_bh_new(xen_9pfs_bh, &xen_9pdev->rings[i]); 440 xen_9pdev->rings[i].out_cons = 0; 441 xen_9pdev->rings[i].out_size = 0; 442 xen_9pdev->rings[i].inprogress = false; 443 444 445 xen_9pdev->rings[i].evtchndev = xenevtchn_open(NULL, 0); 446 if (xen_9pdev->rings[i].evtchndev == NULL) { 447 goto out; 448 } 449 qemu_set_cloexec(xenevtchn_fd(xen_9pdev->rings[i].evtchndev)); 450 xen_9pdev->rings[i].local_port = xenevtchn_bind_interdomain 451 (xen_9pdev->rings[i].evtchndev, 452 xendev->dom, 453 xen_9pdev->rings[i].evtchn); 454 if (xen_9pdev->rings[i].local_port == -1) { 455 xen_pv_printf(xendev, 0, 456 "xenevtchn_bind_interdomain failed port=%d\n", 457 xen_9pdev->rings[i].evtchn); 458 goto out; 459 } 460 xen_pv_printf(xendev, 2, "bind evtchn port %d\n", xendev->local_port); 461 qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev), 462 xen_9pfs_evtchn_event, NULL, &xen_9pdev->rings[i]); 463 } 464 465 xen_9pdev->security_model = xenstore_read_be_str(xendev, "security_model"); 466 xen_9pdev->path = xenstore_read_be_str(xendev, "path"); 467 xen_9pdev->id = s->fsconf.fsdev_id = 468 g_strdup_printf("xen9p%d", xendev->dev); 469 xen_9pdev->tag = s->fsconf.tag = xenstore_read_fe_str(xendev, "tag"); 470 fsdev = qemu_opts_create(qemu_find_opts("fsdev"), 471 s->fsconf.tag, 472 1, NULL); 473 qemu_opt_set(fsdev, "fsdriver", "local", NULL); 474 qemu_opt_set(fsdev, "path", xen_9pdev->path, NULL); 475 qemu_opt_set(fsdev, "security_model", xen_9pdev->security_model, NULL); 476 qemu_opts_set_id(fsdev, s->fsconf.fsdev_id); 477 qemu_fsdev_add(fsdev, &err); 478 if (err) { 479 error_report_err(err); 480 } 481 v9fs_device_realize_common(s, &xen_9p_transport, NULL); 482 483 return 0; 484 485 out: 486 xen_9pfs_free(xendev); 487 return -1; 488 } 489 490 static void xen_9pfs_alloc(struct XenLegacyDevice *xendev) 491 { 492 xenstore_write_be_str(xendev, "versions", VERSIONS); 493 xenstore_write_be_int(xendev, "max-rings", MAX_RINGS); 494 xenstore_write_be_int(xendev, "max-ring-page-order", MAX_RING_ORDER); 495 } 496 497 struct XenDevOps xen_9pfs_ops = { 498 .size = sizeof(Xen9pfsDev), 499 .flags = DEVOPS_FLAG_NEED_GNTDEV, 500 .alloc = xen_9pfs_alloc, 501 .init = xen_9pfs_init, 502 .initialise = xen_9pfs_connect, 503 .disconnect = xen_9pfs_disconnect, 504 .free = xen_9pfs_free, 505 }; 506