jcs.org
qemu with hax to log dma reads & writes
jcs.org/2018/11/12/vfio
1/*
2 * virtio-fs glue for FUSE
3 * Copyright (C) 2018 Red Hat, Inc. and/or its affiliates
4 *
5 * Authors:
6 * Dave Gilbert <dgilbert@redhat.com>
7 *
8 * Implements the glue between libfuse and libvhost-user
9 *
10 * This program can be distributed under the terms of the GNU LGPLv2.
11 * See the file COPYING.LIB
12 */
13
14#include "qemu/osdep.h"
15#include "qemu/iov.h"
16#include "qapi/error.h"
17#include "fuse_i.h"
18#include "standard-headers/linux/fuse.h"
19#include "fuse_misc.h"
20#include "fuse_opt.h"
21#include "fuse_virtio.h"
22
23#include <assert.h>
24#include <errno.h>
25#include <glib.h>
26#include <stdint.h>
27#include <stdio.h>
28#include <stdlib.h>
29#include <string.h>
30#include <sys/eventfd.h>
31#include <sys/socket.h>
32#include <sys/types.h>
33#include <sys/un.h>
34#include <unistd.h>
35
36#include "contrib/libvhost-user/libvhost-user.h"
37
38struct fv_VuDev;
39struct fv_QueueInfo {
40 pthread_t thread;
41 /*
42 * This lock protects the VuVirtq preventing races between
43 * fv_queue_thread() and fv_queue_worker().
44 */
45 pthread_mutex_t vq_lock;
46
47 struct fv_VuDev *virtio_dev;
48
49 /* Our queue index, corresponds to array position */
50 int qidx;
51 int kick_fd;
52 int kill_fd; /* For killing the thread */
53};
54
55/* A FUSE request */
56typedef struct {
57 VuVirtqElement elem;
58 struct fuse_chan ch;
59
60 /* Used to complete requests that involve no reply */
61 bool reply_sent;
62} FVRequest;
63
64/*
65 * We pass the dev element into libvhost-user
66 * and then use it to get back to the outer
67 * container for other data.
68 */
69struct fv_VuDev {
70 VuDev dev;
71 struct fuse_session *se;
72
73 /*
74 * Either handle virtqueues or vhost-user protocol messages. Don't do
75 * both at the same time since that could lead to race conditions if
76 * virtqueues or memory tables change while another thread is accessing
77 * them.
78 *
79 * The assumptions are:
80 * 1. fv_queue_thread() reads/writes to virtqueues and only reads VuDev.
81 * 2. virtio_loop() reads/writes virtqueues and VuDev.
82 */
83 pthread_rwlock_t vu_dispatch_rwlock;
84
85 /*
86 * The following pair of fields are only accessed in the main
87 * virtio_loop
88 */
89 size_t nqueues;
90 struct fv_QueueInfo **qi;
91};
92
93/* From spec */
94struct virtio_fs_config {
95 char tag[36];
96 uint32_t num_queues;
97};
98
99/* Callback from libvhost-user */
100static uint64_t fv_get_features(VuDev *dev)
101{
102 return 1ULL << VIRTIO_F_VERSION_1;
103}
104
105/* Callback from libvhost-user */
106static void fv_set_features(VuDev *dev, uint64_t features)
107{
108}
109
110/*
111 * Callback from libvhost-user if there's a new fd we're supposed to listen
112 * to, typically a queue kick?
113 */
114static void fv_set_watch(VuDev *dev, int fd, int condition, vu_watch_cb cb,
115 void *data)
116{
117 fuse_log(FUSE_LOG_WARNING, "%s: TODO! fd=%d\n", __func__, fd);
118}
119
120/*
121 * Callback from libvhost-user if we're no longer supposed to listen on an fd
122 */
123static void fv_remove_watch(VuDev *dev, int fd)
124{
125 fuse_log(FUSE_LOG_WARNING, "%s: TODO! fd=%d\n", __func__, fd);
126}
127
128/* Callback from libvhost-user to panic */
129static void fv_panic(VuDev *dev, const char *err)
130{
131 fuse_log(FUSE_LOG_ERR, "%s: libvhost-user: %s\n", __func__, err);
132 /* TODO: Allow reconnects?? */
133 exit(EXIT_FAILURE);
134}
135
136/*
137 * Copy from an iovec into a fuse_buf (memory only)
138 * Caller must ensure there is space
139 */
140static void copy_from_iov(struct fuse_buf *buf, size_t out_num,
141 const struct iovec *out_sg)
142{
143 void *dest = buf->mem;
144
145 while (out_num) {
146 size_t onelen = out_sg->iov_len;
147 memcpy(dest, out_sg->iov_base, onelen);
148 dest += onelen;
149 out_sg++;
150 out_num--;
151 }
152}
153
154/*
155 * Copy from one iov to another, the given number of bytes
156 * The caller must have checked sizes.
157 */
158static void copy_iov(struct iovec *src_iov, int src_count,
159 struct iovec *dst_iov, int dst_count, size_t to_copy)
160{
161 size_t dst_offset = 0;
162 /* Outer loop copies 'src' elements */
163 while (to_copy) {
164 assert(src_count);
165 size_t src_len = src_iov[0].iov_len;
166 size_t src_offset = 0;
167
168 if (src_len > to_copy) {
169 src_len = to_copy;
170 }
171 /* Inner loop copies contents of one 'src' to maybe multiple dst. */
172 while (src_len) {
173 assert(dst_count);
174 size_t dst_len = dst_iov[0].iov_len - dst_offset;
175 if (dst_len > src_len) {
176 dst_len = src_len;
177 }
178
179 memcpy(dst_iov[0].iov_base + dst_offset,
180 src_iov[0].iov_base + src_offset, dst_len);
181 src_len -= dst_len;
182 to_copy -= dst_len;
183 src_offset += dst_len;
184 dst_offset += dst_len;
185
186 assert(dst_offset <= dst_iov[0].iov_len);
187 if (dst_offset == dst_iov[0].iov_len) {
188 dst_offset = 0;
189 dst_iov++;
190 dst_count--;
191 }
192 }
193 src_iov++;
194 src_count--;
195 }
196}
197
198/*
199 * Called back by ll whenever it wants to send a reply/message back
200 * The 1st element of the iov starts with the fuse_out_header
201 * 'unique'==0 means it's a notify message.
202 */
203int virtio_send_msg(struct fuse_session *se, struct fuse_chan *ch,
204 struct iovec *iov, int count)
205{
206 FVRequest *req = container_of(ch, FVRequest, ch);
207 struct fv_QueueInfo *qi = ch->qi;
208 VuDev *dev = &se->virtio_dev->dev;
209 VuVirtq *q = vu_get_queue(dev, qi->qidx);
210 VuVirtqElement *elem = &req->elem;
211 int ret = 0;
212
213 assert(count >= 1);
214 assert(iov[0].iov_len >= sizeof(struct fuse_out_header));
215
216 struct fuse_out_header *out = iov[0].iov_base;
217 /* TODO: Endianness! */
218
219 size_t tosend_len = iov_size(iov, count);
220
221 /* unique == 0 is notification, which we don't support */
222 assert(out->unique);
223 assert(!req->reply_sent);
224
225 /* The 'in' part of the elem is to qemu */
226 unsigned int in_num = elem->in_num;
227 struct iovec *in_sg = elem->in_sg;
228 size_t in_len = iov_size(in_sg, in_num);
229 fuse_log(FUSE_LOG_DEBUG, "%s: elem %d: with %d in desc of length %zd\n",
230 __func__, elem->index, in_num, in_len);
231
232 /*
233 * The elem should have room for a 'fuse_out_header' (out from fuse)
234 * plus the data based on the len in the header.
235 */
236 if (in_len < sizeof(struct fuse_out_header)) {
237 fuse_log(FUSE_LOG_ERR, "%s: elem %d too short for out_header\n",
238 __func__, elem->index);
239 ret = -E2BIG;
240 goto err;
241 }
242 if (in_len < tosend_len) {
243 fuse_log(FUSE_LOG_ERR, "%s: elem %d too small for data len %zd\n",
244 __func__, elem->index, tosend_len);
245 ret = -E2BIG;
246 goto err;
247 }
248
249 copy_iov(iov, count, in_sg, in_num, tosend_len);
250
251 pthread_rwlock_rdlock(&qi->virtio_dev->vu_dispatch_rwlock);
252 pthread_mutex_lock(&qi->vq_lock);
253 vu_queue_push(dev, q, elem, tosend_len);
254 vu_queue_notify(dev, q);
255 pthread_mutex_unlock(&qi->vq_lock);
256 pthread_rwlock_unlock(&qi->virtio_dev->vu_dispatch_rwlock);
257
258 req->reply_sent = true;
259
260err:
261 return ret;
262}
263
264/*
265 * Callback from fuse_send_data_iov_* when it's virtio and the buffer
266 * is a single FD with FUSE_BUF_IS_FD | FUSE_BUF_FD_SEEK
267 * We need send the iov and then the buffer.
268 * Return 0 on success
269 */
270int virtio_send_data_iov(struct fuse_session *se, struct fuse_chan *ch,
271 struct iovec *iov, int count, struct fuse_bufvec *buf,
272 size_t len)
273{
274 FVRequest *req = container_of(ch, FVRequest, ch);
275 struct fv_QueueInfo *qi = ch->qi;
276 VuDev *dev = &se->virtio_dev->dev;
277 VuVirtq *q = vu_get_queue(dev, qi->qidx);
278 VuVirtqElement *elem = &req->elem;
279 int ret = 0;
280
281 assert(count >= 1);
282 assert(iov[0].iov_len >= sizeof(struct fuse_out_header));
283
284 struct fuse_out_header *out = iov[0].iov_base;
285 /* TODO: Endianness! */
286
287 size_t iov_len = iov_size(iov, count);
288 size_t tosend_len = iov_len + len;
289
290 out->len = tosend_len;
291
292 fuse_log(FUSE_LOG_DEBUG, "%s: count=%d len=%zd iov_len=%zd\n", __func__,
293 count, len, iov_len);
294
295 /* unique == 0 is notification which we don't support */
296 assert(out->unique);
297
298 assert(!req->reply_sent);
299
300 /* The 'in' part of the elem is to qemu */
301 unsigned int in_num = elem->in_num;
302 struct iovec *in_sg = elem->in_sg;
303 size_t in_len = iov_size(in_sg, in_num);
304 fuse_log(FUSE_LOG_DEBUG, "%s: elem %d: with %d in desc of length %zd\n",
305 __func__, elem->index, in_num, in_len);
306
307 /*
308 * The elem should have room for a 'fuse_out_header' (out from fuse)
309 * plus the data based on the len in the header.
310 */
311 if (in_len < sizeof(struct fuse_out_header)) {
312 fuse_log(FUSE_LOG_ERR, "%s: elem %d too short for out_header\n",
313 __func__, elem->index);
314 ret = E2BIG;
315 goto err;
316 }
317 if (in_len < tosend_len) {
318 fuse_log(FUSE_LOG_ERR, "%s: elem %d too small for data len %zd\n",
319 __func__, elem->index, tosend_len);
320 ret = E2BIG;
321 goto err;
322 }
323
324 /* TODO: Limit to 'len' */
325
326 /* First copy the header data from iov->in_sg */
327 copy_iov(iov, count, in_sg, in_num, iov_len);
328
329 /*
330 * Build a copy of the the in_sg iov so we can skip bits in it,
331 * including changing the offsets
332 */
333 struct iovec *in_sg_cpy = calloc(sizeof(struct iovec), in_num);
334 assert(in_sg_cpy);
335 memcpy(in_sg_cpy, in_sg, sizeof(struct iovec) * in_num);
336 /* These get updated as we skip */
337 struct iovec *in_sg_ptr = in_sg_cpy;
338 int in_sg_cpy_count = in_num;
339
340 /* skip over parts of in_sg that contained the header iov */
341 size_t skip_size = iov_len;
342
343 size_t in_sg_left = 0;
344 do {
345 while (skip_size != 0 && in_sg_cpy_count) {
346 if (skip_size >= in_sg_ptr[0].iov_len) {
347 skip_size -= in_sg_ptr[0].iov_len;
348 in_sg_ptr++;
349 in_sg_cpy_count--;
350 } else {
351 in_sg_ptr[0].iov_len -= skip_size;
352 in_sg_ptr[0].iov_base += skip_size;
353 break;
354 }
355 }
356
357 int i;
358 for (i = 0, in_sg_left = 0; i < in_sg_cpy_count; i++) {
359 in_sg_left += in_sg_ptr[i].iov_len;
360 }
361 fuse_log(FUSE_LOG_DEBUG,
362 "%s: after skip skip_size=%zd in_sg_cpy_count=%d "
363 "in_sg_left=%zd\n",
364 __func__, skip_size, in_sg_cpy_count, in_sg_left);
365 ret = preadv(buf->buf[0].fd, in_sg_ptr, in_sg_cpy_count,
366 buf->buf[0].pos);
367
368 if (ret == -1) {
369 ret = errno;
370 fuse_log(FUSE_LOG_DEBUG, "%s: preadv failed (%m) len=%zd\n",
371 __func__, len);
372 free(in_sg_cpy);
373 goto err;
374 }
375 fuse_log(FUSE_LOG_DEBUG, "%s: preadv ret=%d len=%zd\n", __func__,
376 ret, len);
377 if (ret < len && ret) {
378 fuse_log(FUSE_LOG_DEBUG, "%s: ret < len\n", __func__);
379 /* Skip over this much next time around */
380 skip_size = ret;
381 buf->buf[0].pos += ret;
382 len -= ret;
383
384 /* Lets do another read */
385 continue;
386 }
387 if (!ret) {
388 /* EOF case? */
389 fuse_log(FUSE_LOG_DEBUG, "%s: !ret in_sg_left=%zd\n", __func__,
390 in_sg_left);
391 break;
392 }
393 if (ret != len) {
394 fuse_log(FUSE_LOG_DEBUG, "%s: ret!=len\n", __func__);
395 ret = EIO;
396 free(in_sg_cpy);
397 goto err;
398 }
399 in_sg_left -= ret;
400 len -= ret;
401 } while (in_sg_left);
402 free(in_sg_cpy);
403
404 /* Need to fix out->len on EOF */
405 if (len) {
406 struct fuse_out_header *out_sg = in_sg[0].iov_base;
407
408 tosend_len -= len;
409 out_sg->len = tosend_len;
410 }
411
412 ret = 0;
413
414 pthread_rwlock_rdlock(&qi->virtio_dev->vu_dispatch_rwlock);
415 pthread_mutex_lock(&qi->vq_lock);
416 vu_queue_push(dev, q, elem, tosend_len);
417 vu_queue_notify(dev, q);
418 pthread_mutex_unlock(&qi->vq_lock);
419 pthread_rwlock_unlock(&qi->virtio_dev->vu_dispatch_rwlock);
420
421err:
422 if (ret == 0) {
423 req->reply_sent = true;
424 }
425
426 return ret;
427}
428
429static __thread bool clone_fs_called;
430
431/* Process one FVRequest in a thread pool */
432static void fv_queue_worker(gpointer data, gpointer user_data)
433{
434 struct fv_QueueInfo *qi = user_data;
435 struct fuse_session *se = qi->virtio_dev->se;
436 struct VuDev *dev = &qi->virtio_dev->dev;
437 FVRequest *req = data;
438 VuVirtqElement *elem = &req->elem;
439 struct fuse_buf fbuf = {};
440 bool allocated_bufv = false;
441 struct fuse_bufvec bufv;
442 struct fuse_bufvec *pbufv;
443
444 assert(se->bufsize > sizeof(struct fuse_in_header));
445
446 if (!clone_fs_called) {
447 int ret;
448
449 /* unshare FS for xattr operation */
450 ret = unshare(CLONE_FS);
451 /* should not fail */
452 assert(ret == 0);
453
454 clone_fs_called = true;
455 }
456
457 /*
458 * An element contains one request and the space to send our response
459 * They're spread over multiple descriptors in a scatter/gather set
460 * and we can't trust the guest to keep them still; so copy in/out.
461 */
462 fbuf.mem = malloc(se->bufsize);
463 assert(fbuf.mem);
464
465 fuse_mutex_init(&req->ch.lock);
466 req->ch.fd = -1;
467 req->ch.qi = qi;
468
469 /* The 'out' part of the elem is from qemu */
470 unsigned int out_num = elem->out_num;
471 struct iovec *out_sg = elem->out_sg;
472 size_t out_len = iov_size(out_sg, out_num);
473 fuse_log(FUSE_LOG_DEBUG,
474 "%s: elem %d: with %d out desc of length %zd\n",
475 __func__, elem->index, out_num, out_len);
476
477 /*
478 * The elem should contain a 'fuse_in_header' (in to fuse)
479 * plus the data based on the len in the header.
480 */
481 if (out_len < sizeof(struct fuse_in_header)) {
482 fuse_log(FUSE_LOG_ERR, "%s: elem %d too short for in_header\n",
483 __func__, elem->index);
484 assert(0); /* TODO */
485 }
486 if (out_len > se->bufsize) {
487 fuse_log(FUSE_LOG_ERR, "%s: elem %d too large for buffer\n", __func__,
488 elem->index);
489 assert(0); /* TODO */
490 }
491 /* Copy just the first element and look at it */
492 copy_from_iov(&fbuf, 1, out_sg);
493
494 pbufv = NULL; /* Compiler thinks an unitialised path */
495 if (out_num > 2 &&
496 out_sg[0].iov_len == sizeof(struct fuse_in_header) &&
497 ((struct fuse_in_header *)fbuf.mem)->opcode == FUSE_WRITE &&
498 out_sg[1].iov_len == sizeof(struct fuse_write_in)) {
499 /*
500 * For a write we don't actually need to copy the
501 * data, we can just do it straight out of guest memory
502 * but we must still copy the headers in case the guest
503 * was nasty and changed them while we were using them.
504 */
505 fuse_log(FUSE_LOG_DEBUG, "%s: Write special case\n", __func__);
506
507 /* copy the fuse_write_in header afte rthe fuse_in_header */
508 fbuf.mem += out_sg->iov_len;
509 copy_from_iov(&fbuf, 1, out_sg + 1);
510 fbuf.mem -= out_sg->iov_len;
511 fbuf.size = out_sg[0].iov_len + out_sg[1].iov_len;
512
513 /* Allocate the bufv, with space for the rest of the iov */
514 pbufv = malloc(sizeof(struct fuse_bufvec) +
515 sizeof(struct fuse_buf) * (out_num - 2));
516 if (!pbufv) {
517 fuse_log(FUSE_LOG_ERR, "%s: pbufv malloc failed\n",
518 __func__);
519 goto out;
520 }
521
522 allocated_bufv = true;
523 pbufv->count = 1;
524 pbufv->buf[0] = fbuf;
525
526 size_t iovindex, pbufvindex;
527 iovindex = 2; /* 2 headers, separate iovs */
528 pbufvindex = 1; /* 2 headers, 1 fusebuf */
529
530 for (; iovindex < out_num; iovindex++, pbufvindex++) {
531 pbufv->count++;
532 pbufv->buf[pbufvindex].pos = ~0; /* Dummy */
533 pbufv->buf[pbufvindex].flags = 0;
534 pbufv->buf[pbufvindex].mem = out_sg[iovindex].iov_base;
535 pbufv->buf[pbufvindex].size = out_sg[iovindex].iov_len;
536 }
537 } else {
538 /* Normal (non fast write) path */
539
540 /* Copy the rest of the buffer */
541 fbuf.mem += out_sg->iov_len;
542 copy_from_iov(&fbuf, out_num - 1, out_sg + 1);
543 fbuf.mem -= out_sg->iov_len;
544 fbuf.size = out_len;
545
546 /* TODO! Endianness of header */
547
548 /* TODO: Add checks for fuse_session_exited */
549 bufv.buf[0] = fbuf;
550 bufv.count = 1;
551 pbufv = &bufv;
552 }
553 pbufv->idx = 0;
554 pbufv->off = 0;
555 fuse_session_process_buf_int(se, pbufv, &req->ch);
556
557out:
558 if (allocated_bufv) {
559 free(pbufv);
560 }
561
562 /* If the request has no reply, still recycle the virtqueue element */
563 if (!req->reply_sent) {
564 struct VuVirtq *q = vu_get_queue(dev, qi->qidx);
565
566 fuse_log(FUSE_LOG_DEBUG, "%s: elem %d no reply sent\n", __func__,
567 elem->index);
568
569 pthread_rwlock_rdlock(&qi->virtio_dev->vu_dispatch_rwlock);
570 pthread_mutex_lock(&qi->vq_lock);
571 vu_queue_push(dev, q, elem, 0);
572 vu_queue_notify(dev, q);
573 pthread_mutex_unlock(&qi->vq_lock);
574 pthread_rwlock_unlock(&qi->virtio_dev->vu_dispatch_rwlock);
575 }
576
577 pthread_mutex_destroy(&req->ch.lock);
578 free(fbuf.mem);
579 free(req);
580}
581
582/* Thread function for individual queues, created when a queue is 'started' */
583static void *fv_queue_thread(void *opaque)
584{
585 struct fv_QueueInfo *qi = opaque;
586 struct VuDev *dev = &qi->virtio_dev->dev;
587 struct VuVirtq *q = vu_get_queue(dev, qi->qidx);
588 struct fuse_session *se = qi->virtio_dev->se;
589 GThreadPool *pool;
590
591 pool = g_thread_pool_new(fv_queue_worker, qi, se->thread_pool_size, TRUE,
592 NULL);
593 if (!pool) {
594 fuse_log(FUSE_LOG_ERR, "%s: g_thread_pool_new failed\n", __func__);
595 return NULL;
596 }
597
598 fuse_log(FUSE_LOG_INFO, "%s: Start for queue %d kick_fd %d\n", __func__,
599 qi->qidx, qi->kick_fd);
600 while (1) {
601 struct pollfd pf[2];
602 int ret;
603
604 pf[0].fd = qi->kick_fd;
605 pf[0].events = POLLIN;
606 pf[0].revents = 0;
607 pf[1].fd = qi->kill_fd;
608 pf[1].events = POLLIN;
609 pf[1].revents = 0;
610
611 fuse_log(FUSE_LOG_DEBUG, "%s: Waiting for Queue %d event\n", __func__,
612 qi->qidx);
613 int poll_res = ppoll(pf, 2, NULL, NULL);
614
615 if (poll_res == -1) {
616 if (errno == EINTR) {
617 fuse_log(FUSE_LOG_INFO, "%s: ppoll interrupted, going around\n",
618 __func__);
619 continue;
620 }
621 fuse_log(FUSE_LOG_ERR, "fv_queue_thread ppoll: %m\n");
622 break;
623 }
624 assert(poll_res >= 1);
625 if (pf[0].revents & (POLLERR | POLLHUP | POLLNVAL)) {
626 fuse_log(FUSE_LOG_ERR, "%s: Unexpected poll revents %x Queue %d\n",
627 __func__, pf[0].revents, qi->qidx);
628 break;
629 }
630 if (pf[1].revents & (POLLERR | POLLHUP | POLLNVAL)) {
631 fuse_log(FUSE_LOG_ERR,
632 "%s: Unexpected poll revents %x Queue %d killfd\n",
633 __func__, pf[1].revents, qi->qidx);
634 break;
635 }
636 if (pf[1].revents) {
637 fuse_log(FUSE_LOG_INFO, "%s: kill event on queue %d - quitting\n",
638 __func__, qi->qidx);
639 break;
640 }
641 assert(pf[0].revents & POLLIN);
642 fuse_log(FUSE_LOG_DEBUG, "%s: Got queue event on Queue %d\n", __func__,
643 qi->qidx);
644
645 eventfd_t evalue;
646 if (eventfd_read(qi->kick_fd, &evalue)) {
647 fuse_log(FUSE_LOG_ERR, "Eventfd_read for queue: %m\n");
648 break;
649 }
650 /* Mutual exclusion with virtio_loop() */
651 ret = pthread_rwlock_rdlock(&qi->virtio_dev->vu_dispatch_rwlock);
652 assert(ret == 0); /* there is no possible error case */
653 pthread_mutex_lock(&qi->vq_lock);
654 /* out is from guest, in is too guest */
655 unsigned int in_bytes, out_bytes;
656 vu_queue_get_avail_bytes(dev, q, &in_bytes, &out_bytes, ~0, ~0);
657
658 fuse_log(FUSE_LOG_DEBUG,
659 "%s: Queue %d gave evalue: %zx available: in: %u out: %u\n",
660 __func__, qi->qidx, (size_t)evalue, in_bytes, out_bytes);
661
662 while (1) {
663 FVRequest *req = vu_queue_pop(dev, q, sizeof(FVRequest));
664 if (!req) {
665 break;
666 }
667
668 req->reply_sent = false;
669
670 g_thread_pool_push(pool, req, NULL);
671 }
672
673 pthread_mutex_unlock(&qi->vq_lock);
674 pthread_rwlock_unlock(&qi->virtio_dev->vu_dispatch_rwlock);
675 }
676
677 g_thread_pool_free(pool, FALSE, TRUE);
678
679 return NULL;
680}
681
682static void fv_queue_cleanup_thread(struct fv_VuDev *vud, int qidx)
683{
684 int ret;
685 struct fv_QueueInfo *ourqi;
686
687 assert(qidx < vud->nqueues);
688 ourqi = vud->qi[qidx];
689
690 /* Kill the thread */
691 if (eventfd_write(ourqi->kill_fd, 1)) {
692 fuse_log(FUSE_LOG_ERR, "Eventfd_write for queue %d: %s\n",
693 qidx, strerror(errno));
694 }
695 ret = pthread_join(ourqi->thread, NULL);
696 if (ret) {
697 fuse_log(FUSE_LOG_ERR, "%s: Failed to join thread idx %d err %d\n",
698 __func__, qidx, ret);
699 }
700 pthread_mutex_destroy(&ourqi->vq_lock);
701 close(ourqi->kill_fd);
702 ourqi->kick_fd = -1;
703 free(vud->qi[qidx]);
704 vud->qi[qidx] = NULL;
705}
706
707/* Callback from libvhost-user on start or stop of a queue */
708static void fv_queue_set_started(VuDev *dev, int qidx, bool started)
709{
710 struct fv_VuDev *vud = container_of(dev, struct fv_VuDev, dev);
711 struct fv_QueueInfo *ourqi;
712
713 fuse_log(FUSE_LOG_INFO, "%s: qidx=%d started=%d\n", __func__, qidx,
714 started);
715 assert(qidx >= 0);
716
717 /*
718 * Ignore additional request queues for now. passthrough_ll.c must be
719 * audited for thread-safety issues first. It was written with a
720 * well-behaved client in mind and may not protect against all types of
721 * races yet.
722 */
723 if (qidx > 1) {
724 fuse_log(FUSE_LOG_ERR,
725 "%s: multiple request queues not yet implemented, please only "
726 "configure 1 request queue\n",
727 __func__);
728 exit(EXIT_FAILURE);
729 }
730
731 if (started) {
732 /* Fire up a thread to watch this queue */
733 if (qidx >= vud->nqueues) {
734 vud->qi = realloc(vud->qi, (qidx + 1) * sizeof(vud->qi[0]));
735 assert(vud->qi);
736 memset(vud->qi + vud->nqueues, 0,
737 sizeof(vud->qi[0]) * (1 + (qidx - vud->nqueues)));
738 vud->nqueues = qidx + 1;
739 }
740 if (!vud->qi[qidx]) {
741 vud->qi[qidx] = calloc(sizeof(struct fv_QueueInfo), 1);
742 assert(vud->qi[qidx]);
743 vud->qi[qidx]->virtio_dev = vud;
744 vud->qi[qidx]->qidx = qidx;
745 } else {
746 /* Shouldn't have been started */
747 assert(vud->qi[qidx]->kick_fd == -1);
748 }
749 ourqi = vud->qi[qidx];
750 ourqi->kick_fd = dev->vq[qidx].kick_fd;
751
752 ourqi->kill_fd = eventfd(0, EFD_CLOEXEC | EFD_SEMAPHORE);
753 assert(ourqi->kill_fd != -1);
754 pthread_mutex_init(&ourqi->vq_lock, NULL);
755
756 if (pthread_create(&ourqi->thread, NULL, fv_queue_thread, ourqi)) {
757 fuse_log(FUSE_LOG_ERR, "%s: Failed to create thread for queue %d\n",
758 __func__, qidx);
759 assert(0);
760 }
761 } else {
762 fv_queue_cleanup_thread(vud, qidx);
763 }
764}
765
766static bool fv_queue_order(VuDev *dev, int qidx)
767{
768 return false;
769}
770
771static const VuDevIface fv_iface = {
772 .get_features = fv_get_features,
773 .set_features = fv_set_features,
774
775 /* Don't need process message, we've not got any at vhost-user level */
776 .queue_set_started = fv_queue_set_started,
777
778 .queue_is_processed_in_order = fv_queue_order,
779};
780
781/*
782 * Main loop; this mostly deals with events on the vhost-user
783 * socket itself, and not actual fuse data.
784 */
785int virtio_loop(struct fuse_session *se)
786{
787 fuse_log(FUSE_LOG_INFO, "%s: Entry\n", __func__);
788
789 while (!fuse_session_exited(se)) {
790 struct pollfd pf[1];
791 bool ok;
792 int ret;
793 pf[0].fd = se->vu_socketfd;
794 pf[0].events = POLLIN;
795 pf[0].revents = 0;
796
797 fuse_log(FUSE_LOG_DEBUG, "%s: Waiting for VU event\n", __func__);
798 int poll_res = ppoll(pf, 1, NULL, NULL);
799
800 if (poll_res == -1) {
801 if (errno == EINTR) {
802 fuse_log(FUSE_LOG_INFO, "%s: ppoll interrupted, going around\n",
803 __func__);
804 continue;
805 }
806 fuse_log(FUSE_LOG_ERR, "virtio_loop ppoll: %m\n");
807 break;
808 }
809 assert(poll_res == 1);
810 if (pf[0].revents & (POLLERR | POLLHUP | POLLNVAL)) {
811 fuse_log(FUSE_LOG_ERR, "%s: Unexpected poll revents %x\n", __func__,
812 pf[0].revents);
813 break;
814 }
815 assert(pf[0].revents & POLLIN);
816 fuse_log(FUSE_LOG_DEBUG, "%s: Got VU event\n", __func__);
817 /* Mutual exclusion with fv_queue_thread() */
818 ret = pthread_rwlock_wrlock(&se->virtio_dev->vu_dispatch_rwlock);
819 assert(ret == 0); /* there is no possible error case */
820
821 ok = vu_dispatch(&se->virtio_dev->dev);
822
823 pthread_rwlock_unlock(&se->virtio_dev->vu_dispatch_rwlock);
824
825 if (!ok) {
826 fuse_log(FUSE_LOG_ERR, "%s: vu_dispatch failed\n", __func__);
827 break;
828 }
829 }
830
831 /*
832 * Make sure all fv_queue_thread()s quit on exit, as we're about to
833 * free virtio dev and fuse session, no one should access them anymore.
834 */
835 for (int i = 0; i < se->virtio_dev->nqueues; i++) {
836 if (!se->virtio_dev->qi[i]) {
837 continue;
838 }
839
840 fuse_log(FUSE_LOG_INFO, "%s: Stopping queue %d thread\n", __func__, i);
841 fv_queue_cleanup_thread(se->virtio_dev, i);
842 }
843
844 fuse_log(FUSE_LOG_INFO, "%s: Exit\n", __func__);
845
846 return 0;
847}
848
849static void strreplace(char *s, char old, char new)
850{
851 for (; *s; ++s) {
852 if (*s == old) {
853 *s = new;
854 }
855 }
856}
857
858static bool fv_socket_lock(struct fuse_session *se)
859{
860 g_autofree gchar *sk_name = NULL;
861 g_autofree gchar *pidfile = NULL;
862 g_autofree gchar *dir = NULL;
863 Error *local_err = NULL;
864
865 dir = qemu_get_local_state_pathname("run/virtiofsd");
866
867 if (g_mkdir_with_parents(dir, S_IRWXU) < 0) {
868 fuse_log(FUSE_LOG_ERR, "%s: Failed to create directory %s: %s",
869 __func__, dir, strerror(errno));
870 return false;
871 }
872
873 sk_name = g_strdup(se->vu_socket_path);
874 strreplace(sk_name, '/', '.');
875 pidfile = g_strdup_printf("%s/%s.pid", dir, sk_name);
876
877 if (!qemu_write_pidfile(pidfile, &local_err)) {
878 error_report_err(local_err);
879 return false;
880 }
881
882 return true;
883}
884
885static int fv_create_listen_socket(struct fuse_session *se)
886{
887 struct sockaddr_un un;
888 mode_t old_umask;
889
890 /* Nothing to do if fd is already initialized */
891 if (se->vu_listen_fd >= 0) {
892 return 0;
893 }
894
895 if (strlen(se->vu_socket_path) >= sizeof(un.sun_path)) {
896 fuse_log(FUSE_LOG_ERR, "Socket path too long\n");
897 return -1;
898 }
899
900 if (!strlen(se->vu_socket_path)) {
901 fuse_log(FUSE_LOG_ERR, "Socket path is empty\n");
902 return -1;
903 }
904
905 /* Check the vu_socket_path is already used */
906 if (!fv_socket_lock(se)) {
907 return -1;
908 }
909
910 /*
911 * Create the Unix socket to communicate with qemu
912 * based on QEMU's vhost-user-bridge
913 */
914 unlink(se->vu_socket_path);
915 strcpy(un.sun_path, se->vu_socket_path);
916 size_t addr_len = sizeof(un);
917
918 int listen_sock = socket(AF_UNIX, SOCK_STREAM, 0);
919 if (listen_sock == -1) {
920 fuse_log(FUSE_LOG_ERR, "vhost socket creation: %m\n");
921 return -1;
922 }
923 un.sun_family = AF_UNIX;
924
925 /*
926 * Unfortunately bind doesn't let you set the mask on the socket,
927 * so set umask to 077 and restore it later.
928 */
929 old_umask = umask(0077);
930 if (bind(listen_sock, (struct sockaddr *)&un, addr_len) == -1) {
931 fuse_log(FUSE_LOG_ERR, "vhost socket bind: %m\n");
932 close(listen_sock);
933 umask(old_umask);
934 return -1;
935 }
936 umask(old_umask);
937
938 if (listen(listen_sock, 1) == -1) {
939 fuse_log(FUSE_LOG_ERR, "vhost socket listen: %m\n");
940 close(listen_sock);
941 return -1;
942 }
943
944 se->vu_listen_fd = listen_sock;
945 return 0;
946}
947
948int virtio_session_mount(struct fuse_session *se)
949{
950 int ret;
951
952 ret = fv_create_listen_socket(se);
953 if (ret < 0) {
954 return ret;
955 }
956
957 se->fd = -1;
958
959 fuse_log(FUSE_LOG_INFO, "%s: Waiting for vhost-user socket connection...\n",
960 __func__);
961 int data_sock = accept(se->vu_listen_fd, NULL, NULL);
962 if (data_sock == -1) {
963 fuse_log(FUSE_LOG_ERR, "vhost socket accept: %m\n");
964 close(se->vu_listen_fd);
965 return -1;
966 }
967 close(se->vu_listen_fd);
968 se->vu_listen_fd = -1;
969 fuse_log(FUSE_LOG_INFO, "%s: Received vhost-user socket connection\n",
970 __func__);
971
972 /* TODO: Some cleanup/deallocation! */
973 se->virtio_dev = calloc(sizeof(struct fv_VuDev), 1);
974 if (!se->virtio_dev) {
975 fuse_log(FUSE_LOG_ERR, "%s: virtio_dev calloc failed\n", __func__);
976 close(data_sock);
977 return -1;
978 }
979
980 se->vu_socketfd = data_sock;
981 se->virtio_dev->se = se;
982 pthread_rwlock_init(&se->virtio_dev->vu_dispatch_rwlock, NULL);
983 vu_init(&se->virtio_dev->dev, 2, se->vu_socketfd, fv_panic, fv_set_watch,
984 fv_remove_watch, &fv_iface);
985
986 return 0;
987}
988
989void virtio_session_close(struct fuse_session *se)
990{
991 close(se->vu_socketfd);
992
993 if (!se->virtio_dev) {
994 return;
995 }
996
997 free(se->virtio_dev->qi);
998 pthread_rwlock_destroy(&se->virtio_dev->vu_dispatch_rwlock);
999 free(se->virtio_dev);
1000 se->virtio_dev = NULL;
1001}