]>
Commit | Line | Data |
---|---|---|
b37eeb02 SS |
1 | /* |
2 | * Xen 9p backend | |
3 | * | |
4 | * Copyright Aporeto 2017 | |
5 | * | |
6 | * Authors: | |
7 | * Stefano Stabellini <[email protected]> | |
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 | ||
f23ef34a SS |
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 | ||
b37eeb02 SS |
45 | typedef struct Xen9pfsDev { |
46 | struct XenDevice xendev; /* must be first */ | |
f23ef34a SS |
47 | V9fsState state; |
48 | char *path; | |
49 | char *security_model; | |
50 | char *tag; | |
51 | char *id; | |
52 | ||
53 | int num_rings; | |
54 | Xen9pfsRing *rings; | |
b37eeb02 SS |
55 | } Xen9pfsDev; |
56 | ||
40a23892 SS |
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 | ||
b37eeb02 SS |
120 | static ssize_t xen_9pfs_pdu_vmarshal(V9fsPDU *pdu, |
121 | size_t offset, | |
122 | const char *fmt, | |
123 | va_list ap) | |
124 | { | |
40a23892 SS |
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); | |
b37eeb02 SS |
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 | { | |
40a23892 SS |
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); | |
b37eeb02 SS |
146 | } |
147 | ||
148 | static void xen_9pfs_init_out_iov_from_pdu(V9fsPDU *pdu, | |
149 | struct iovec **piov, | |
150 | unsigned int *pniov) | |
151 | { | |
40a23892 SS |
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; | |
b37eeb02 SS |
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 | { | |
40a23892 SS |
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; | |
b37eeb02 SS |
179 | } |
180 | ||
181 | static void xen_9pfs_push_and_notify(V9fsPDU *pdu) | |
182 | { | |
4476e09e SS |
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); | |
b37eeb02 SS |
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 | ||
47b70fb1 SS |
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); | |
47b70fb1 SS |
246 | ring->out_size = le32_to_cpu(h.size_le); |
247 | ring->out_cons = cons + le32_to_cpu(h.size_le); | |
248 | ||
506f3275 | 249 | pdu_submit(pdu, &h); |
47b70fb1 SS |
250 | |
251 | return 0; | |
252 | } | |
253 | ||
f23ef34a SS |
254 | static void xen_9pfs_bh(void *opaque) |
255 | { | |
47b70fb1 SS |
256 | Xen9pfsRing *ring = opaque; |
257 | xen_9pfs_receive(ring); | |
f23ef34a SS |
258 | } |
259 | ||
260 | static void xen_9pfs_evtchn_event(void *opaque) | |
261 | { | |
47b70fb1 SS |
262 | Xen9pfsRing *ring = opaque; |
263 | evtchn_port_t port; | |
264 | ||
265 | port = xenevtchn_pending(ring->evtchndev); | |
266 | xenevtchn_unmask(ring->evtchndev, port); | |
267 | ||
268 | qemu_bh_schedule(ring->bh); | |
f23ef34a SS |
269 | } |
270 | ||
b37eeb02 SS |
271 | static int xen_9pfs_free(struct XenDevice *xendev) |
272 | { | |
f23ef34a SS |
273 | int i; |
274 | Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); | |
275 | ||
276 | g_free(xen_9pdev->id); | |
277 | g_free(xen_9pdev->tag); | |
278 | g_free(xen_9pdev->path); | |
279 | g_free(xen_9pdev->security_model); | |
280 | ||
281 | for (i = 0; i < xen_9pdev->num_rings; i++) { | |
282 | if (xen_9pdev->rings[i].data != NULL) { | |
283 | xengnttab_unmap(xen_9pdev->xendev.gnttabdev, | |
284 | xen_9pdev->rings[i].data, | |
285 | (1 << xen_9pdev->rings[i].ring_order)); | |
286 | } | |
287 | if (xen_9pdev->rings[i].intf != NULL) { | |
288 | xengnttab_unmap(xen_9pdev->xendev.gnttabdev, | |
289 | xen_9pdev->rings[i].intf, | |
290 | 1); | |
291 | } | |
292 | if (xen_9pdev->rings[i].evtchndev > 0) { | |
293 | qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev), | |
294 | NULL, NULL, NULL); | |
295 | xenevtchn_unbind(xen_9pdev->rings[i].evtchndev, | |
296 | xen_9pdev->rings[i].local_port); | |
297 | } | |
298 | if (xen_9pdev->rings[i].bh != NULL) { | |
299 | qemu_bh_delete(xen_9pdev->rings[i].bh); | |
300 | } | |
301 | } | |
302 | g_free(xen_9pdev->rings); | |
303 | return 0; | |
b37eeb02 SS |
304 | } |
305 | ||
306 | static int xen_9pfs_connect(struct XenDevice *xendev) | |
307 | { | |
f23ef34a SS |
308 | int i; |
309 | Xen9pfsDev *xen_9pdev = container_of(xendev, Xen9pfsDev, xendev); | |
310 | V9fsState *s = &xen_9pdev->state; | |
311 | QemuOpts *fsdev; | |
312 | ||
313 | if (xenstore_read_fe_int(&xen_9pdev->xendev, "num-rings", | |
314 | &xen_9pdev->num_rings) == -1 || | |
315 | xen_9pdev->num_rings > MAX_RINGS || xen_9pdev->num_rings < 1) { | |
316 | return -1; | |
317 | } | |
318 | ||
319 | xen_9pdev->rings = g_malloc0(xen_9pdev->num_rings * sizeof(Xen9pfsRing)); | |
320 | for (i = 0; i < xen_9pdev->num_rings; i++) { | |
321 | char *str; | |
322 | int ring_order; | |
323 | ||
324 | xen_9pdev->rings[i].priv = xen_9pdev; | |
325 | xen_9pdev->rings[i].evtchn = -1; | |
326 | xen_9pdev->rings[i].local_port = -1; | |
327 | ||
328 | str = g_strdup_printf("ring-ref%u", i); | |
329 | if (xenstore_read_fe_int(&xen_9pdev->xendev, str, | |
330 | &xen_9pdev->rings[i].ref) == -1) { | |
c0c24b95 | 331 | g_free(str); |
f23ef34a SS |
332 | goto out; |
333 | } | |
334 | g_free(str); | |
335 | str = g_strdup_printf("event-channel-%u", i); | |
336 | if (xenstore_read_fe_int(&xen_9pdev->xendev, str, | |
337 | &xen_9pdev->rings[i].evtchn) == -1) { | |
c0c24b95 | 338 | g_free(str); |
f23ef34a SS |
339 | goto out; |
340 | } | |
341 | g_free(str); | |
342 | ||
343 | xen_9pdev->rings[i].intf = xengnttab_map_grant_ref( | |
344 | xen_9pdev->xendev.gnttabdev, | |
345 | xen_9pdev->xendev.dom, | |
346 | xen_9pdev->rings[i].ref, | |
347 | PROT_READ | PROT_WRITE); | |
348 | if (!xen_9pdev->rings[i].intf) { | |
349 | goto out; | |
350 | } | |
351 | ring_order = xen_9pdev->rings[i].intf->ring_order; | |
352 | if (ring_order > MAX_RING_ORDER) { | |
353 | goto out; | |
354 | } | |
355 | xen_9pdev->rings[i].ring_order = ring_order; | |
356 | xen_9pdev->rings[i].data = xengnttab_map_domain_grant_refs( | |
357 | xen_9pdev->xendev.gnttabdev, | |
358 | (1 << ring_order), | |
359 | xen_9pdev->xendev.dom, | |
360 | xen_9pdev->rings[i].intf->ref, | |
361 | PROT_READ | PROT_WRITE); | |
362 | if (!xen_9pdev->rings[i].data) { | |
363 | goto out; | |
364 | } | |
365 | xen_9pdev->rings[i].ring.in = xen_9pdev->rings[i].data; | |
366 | xen_9pdev->rings[i].ring.out = xen_9pdev->rings[i].data + | |
367 | XEN_FLEX_RING_SIZE(ring_order); | |
368 | ||
369 | xen_9pdev->rings[i].bh = qemu_bh_new(xen_9pfs_bh, &xen_9pdev->rings[i]); | |
370 | xen_9pdev->rings[i].out_cons = 0; | |
371 | xen_9pdev->rings[i].out_size = 0; | |
372 | xen_9pdev->rings[i].inprogress = false; | |
373 | ||
374 | ||
375 | xen_9pdev->rings[i].evtchndev = xenevtchn_open(NULL, 0); | |
376 | if (xen_9pdev->rings[i].evtchndev == NULL) { | |
377 | goto out; | |
378 | } | |
01cd90b6 | 379 | qemu_set_cloexec(xenevtchn_fd(xen_9pdev->rings[i].evtchndev)); |
f23ef34a SS |
380 | xen_9pdev->rings[i].local_port = xenevtchn_bind_interdomain |
381 | (xen_9pdev->rings[i].evtchndev, | |
382 | xendev->dom, | |
383 | xen_9pdev->rings[i].evtchn); | |
384 | if (xen_9pdev->rings[i].local_port == -1) { | |
385 | xen_pv_printf(xendev, 0, | |
386 | "xenevtchn_bind_interdomain failed port=%d\n", | |
387 | xen_9pdev->rings[i].evtchn); | |
388 | goto out; | |
389 | } | |
390 | xen_pv_printf(xendev, 2, "bind evtchn port %d\n", xendev->local_port); | |
391 | qemu_set_fd_handler(xenevtchn_fd(xen_9pdev->rings[i].evtchndev), | |
392 | xen_9pfs_evtchn_event, NULL, &xen_9pdev->rings[i]); | |
393 | } | |
394 | ||
395 | xen_9pdev->security_model = xenstore_read_be_str(xendev, "security_model"); | |
396 | xen_9pdev->path = xenstore_read_be_str(xendev, "path"); | |
397 | xen_9pdev->id = s->fsconf.fsdev_id = | |
398 | g_strdup_printf("xen9p%d", xendev->dev); | |
399 | xen_9pdev->tag = s->fsconf.tag = xenstore_read_fe_str(xendev, "tag"); | |
400 | v9fs_register_transport(s, &xen_9p_transport); | |
401 | fsdev = qemu_opts_create(qemu_find_opts("fsdev"), | |
402 | s->fsconf.tag, | |
403 | 1, NULL); | |
404 | qemu_opt_set(fsdev, "fsdriver", "local", NULL); | |
405 | qemu_opt_set(fsdev, "path", xen_9pdev->path, NULL); | |
406 | qemu_opt_set(fsdev, "security_model", xen_9pdev->security_model, NULL); | |
407 | qemu_opts_set_id(fsdev, s->fsconf.fsdev_id); | |
408 | qemu_fsdev_add(fsdev); | |
409 | v9fs_device_realize_common(s, NULL); | |
410 | ||
b37eeb02 | 411 | return 0; |
f23ef34a SS |
412 | |
413 | out: | |
414 | xen_9pfs_free(xendev); | |
415 | return -1; | |
b37eeb02 SS |
416 | } |
417 | ||
418 | static void xen_9pfs_alloc(struct XenDevice *xendev) | |
419 | { | |
f23ef34a SS |
420 | xenstore_write_be_str(xendev, "versions", VERSIONS); |
421 | xenstore_write_be_int(xendev, "max-rings", MAX_RINGS); | |
422 | xenstore_write_be_int(xendev, "max-ring-page-order", MAX_RING_ORDER); | |
b37eeb02 SS |
423 | } |
424 | ||
425 | static void xen_9pfs_disconnect(struct XenDevice *xendev) | |
426 | { | |
f23ef34a | 427 | /* Dynamic hotplug of PV filesystems at runtime is not supported. */ |
b37eeb02 SS |
428 | } |
429 | ||
430 | struct XenDevOps xen_9pfs_ops = { | |
431 | .size = sizeof(Xen9pfsDev), | |
432 | .flags = DEVOPS_FLAG_NEED_GNTDEV, | |
433 | .alloc = xen_9pfs_alloc, | |
434 | .init = xen_9pfs_init, | |
435 | .initialise = xen_9pfs_connect, | |
436 | .disconnect = xen_9pfs_disconnect, | |
437 | .free = xen_9pfs_free, | |
438 | }; |