1 /*
2  * vhost-user-blk sample application
3  *
4  * Copyright (c) 2017 Intel Corporation. All rights reserved.
5  *
6  * Author:
7  *  Changpeng Liu <changpeng.liu@intel.com>
8  *
9  * This work is based on the "vhost-user-scsi" sample and "virtio-blk" driver
10  * implementation by:
11  *  Felipe Franciosi <felipe@nutanix.com>
12  *  Anthony Liguori <aliguori@us.ibm.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2 only.
15  * See the COPYING file in the top-level directory.
16  */
17 
18 #include "qemu/osdep.h"
19 #include "standard-headers/linux/virtio_blk.h"
20 #include "libvhost-user-glib.h"
21 
22 #if defined(__linux__)
23 #include <linux/fs.h>
24 #include <sys/ioctl.h>
25 #endif
26 
27 enum {
28     VHOST_USER_BLK_MAX_QUEUES = 8,
29 };
30 
31 struct virtio_blk_inhdr {
32     unsigned char status;
33 };
34 
35 /* vhost user block device */
36 typedef struct VubDev {
37     VugDev parent;
38     int blk_fd;
39     struct virtio_blk_config blkcfg;
40     bool enable_ro;
41     char *blk_name;
42     GMainLoop *loop;
43 } VubDev;
44 
45 typedef struct VubReq {
46     VuVirtqElement *elem;
47     int64_t sector_num;
48     size_t size;
49     struct virtio_blk_inhdr *in;
50     struct virtio_blk_outhdr *out;
51     VubDev *vdev_blk;
52     struct VuVirtq *vq;
53 } VubReq;
54 
55 /* refer util/iov.c */
56 static size_t vub_iov_size(const struct iovec *iov,
57                               const unsigned int iov_cnt)
58 {
59     size_t len;
60     unsigned int i;
61 
62     len = 0;
63     for (i = 0; i < iov_cnt; i++) {
64         len += iov[i].iov_len;
65     }
66     return len;
67 }
68 
69 static size_t vub_iov_to_buf(const struct iovec *iov,
70                              const unsigned int iov_cnt, void *buf)
71 {
72     size_t len;
73     unsigned int i;
74 
75     len = 0;
76     for (i = 0; i < iov_cnt; i++) {
77         memcpy(buf + len,  iov[i].iov_base, iov[i].iov_len);
78         len += iov[i].iov_len;
79     }
80     return len;
81 }
82 
83 static void vub_panic_cb(VuDev *vu_dev, const char *buf)
84 {
85     VugDev *gdev;
86     VubDev *vdev_blk;
87 
88     assert(vu_dev);
89 
90     gdev = container_of(vu_dev, VugDev, parent);
91     vdev_blk = container_of(gdev, VubDev, parent);
92     if (buf) {
93         g_warning("vu_panic: %s", buf);
94     }
95 
96     g_main_loop_quit(vdev_blk->loop);
97 }
98 
99 static void vub_req_complete(VubReq *req)
100 {
101     VugDev *gdev = &req->vdev_blk->parent;
102     VuDev *vu_dev = &gdev->parent;
103 
104     /* IO size with 1 extra status byte */
105     vu_queue_push(vu_dev, req->vq, req->elem,
106                   req->size + 1);
107     vu_queue_notify(vu_dev, req->vq);
108 
109     g_free(req->elem);
110     g_free(req);
111 }
112 
113 static int vub_open(const char *file_name, bool wce)
114 {
115     int fd;
116     int flags = O_RDWR;
117 
118     if (!wce) {
119         flags |= O_DIRECT;
120     }
121 
122     fd = open(file_name, flags);
123     if (fd < 0) {
124         fprintf(stderr, "Cannot open file %s, %s\n", file_name,
125                 strerror(errno));
126         return -1;
127     }
128 
129     return fd;
130 }
131 
132 static ssize_t
133 vub_readv(VubReq *req, struct iovec *iov, uint32_t iovcnt)
134 {
135     VubDev *vdev_blk = req->vdev_blk;
136     ssize_t rc;
137 
138     if (!iovcnt) {
139         fprintf(stderr, "Invalid Read IOV count\n");
140         return -1;
141     }
142 
143     req->size = vub_iov_size(iov, iovcnt);
144     rc = preadv(vdev_blk->blk_fd, iov, iovcnt, req->sector_num * 512);
145     if (rc < 0) {
146         fprintf(stderr, "%s, Sector %"PRIu64", Size %zu failed with %s\n",
147                 vdev_blk->blk_name, req->sector_num, req->size,
148                 strerror(errno));
149         return -1;
150     }
151 
152     return rc;
153 }
154 
155 static ssize_t
156 vub_writev(VubReq *req, struct iovec *iov, uint32_t iovcnt)
157 {
158     VubDev *vdev_blk = req->vdev_blk;
159     ssize_t rc;
160 
161     if (!iovcnt) {
162         fprintf(stderr, "Invalid Write IOV count\n");
163         return -1;
164     }
165 
166     req->size = vub_iov_size(iov, iovcnt);
167     rc = pwritev(vdev_blk->blk_fd, iov, iovcnt, req->sector_num * 512);
168     if (rc < 0) {
169         fprintf(stderr, "%s, Sector %"PRIu64", Size %zu failed with %s\n",
170                 vdev_blk->blk_name, req->sector_num, req->size,
171                 strerror(errno));
172         return -1;
173     }
174 
175     return rc;
176 }
177 
178 static int
179 vub_discard_write_zeroes(VubReq *req, struct iovec *iov, uint32_t iovcnt,
180                          uint32_t type)
181 {
182     struct virtio_blk_discard_write_zeroes *desc;
183     ssize_t size;
184     void *buf;
185 
186     size = vub_iov_size(iov, iovcnt);
187     if (size != sizeof(*desc)) {
188         fprintf(stderr, "Invalid size %zd, expect %zd\n", size, sizeof(*desc));
189         return -1;
190     }
191     buf = g_new0(char, size);
192     vub_iov_to_buf(iov, iovcnt, buf);
193 
194     #if defined(__linux__) && defined(BLKDISCARD) && defined(BLKZEROOUT)
195     VubDev *vdev_blk = req->vdev_blk;
196     desc = buf;
197     uint64_t range[2] = { le64toh(desc->sector) << 9,
198                           le32toh(desc->num_sectors) << 9 };
199     if (type == VIRTIO_BLK_T_DISCARD) {
200         if (ioctl(vdev_blk->blk_fd, BLKDISCARD, range) == 0) {
201             g_free(buf);
202             return 0;
203         }
204     } else if (type == VIRTIO_BLK_T_WRITE_ZEROES) {
205         if (ioctl(vdev_blk->blk_fd, BLKZEROOUT, range) == 0) {
206             g_free(buf);
207             return 0;
208         }
209     }
210     #endif
211 
212     g_free(buf);
213     return -1;
214 }
215 
216 static void
217 vub_flush(VubReq *req)
218 {
219     VubDev *vdev_blk = req->vdev_blk;
220 
221     fdatasync(vdev_blk->blk_fd);
222 }
223 
224 static int vub_virtio_process_req(VubDev *vdev_blk,
225                                      VuVirtq *vq)
226 {
227     VugDev *gdev = &vdev_blk->parent;
228     VuDev *vu_dev = &gdev->parent;
229     VuVirtqElement *elem;
230     uint32_t type;
231     unsigned in_num;
232     unsigned out_num;
233     VubReq *req;
234 
235     elem = vu_queue_pop(vu_dev, vq, sizeof(VuVirtqElement) + sizeof(VubReq));
236     if (!elem) {
237         return -1;
238     }
239 
240     /* refer to hw/block/virtio_blk.c */
241     if (elem->out_num < 1 || elem->in_num < 1) {
242         fprintf(stderr, "virtio-blk request missing headers\n");
243         g_free(elem);
244         return -1;
245     }
246 
247     req = g_new0(VubReq, 1);
248     req->vdev_blk = vdev_blk;
249     req->vq = vq;
250     req->elem = elem;
251 
252     in_num = elem->in_num;
253     out_num = elem->out_num;
254 
255     /* don't support VIRTIO_F_ANY_LAYOUT and virtio 1.0 only */
256     if (elem->out_sg[0].iov_len < sizeof(struct virtio_blk_outhdr)) {
257         fprintf(stderr, "Invalid outhdr size\n");
258         goto err;
259     }
260     req->out = (struct virtio_blk_outhdr *)elem->out_sg[0].iov_base;
261     out_num--;
262 
263     if (elem->in_sg[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
264         fprintf(stderr, "Invalid inhdr size\n");
265         goto err;
266     }
267     req->in = (struct virtio_blk_inhdr *)elem->in_sg[in_num - 1].iov_base;
268     in_num--;
269 
270     type = le32toh(req->out->type);
271     switch (type & ~VIRTIO_BLK_T_BARRIER) {
272     case VIRTIO_BLK_T_IN:
273     case VIRTIO_BLK_T_OUT: {
274         ssize_t ret = 0;
275         bool is_write = type & VIRTIO_BLK_T_OUT;
276         req->sector_num = le64toh(req->out->sector);
277         if (is_write) {
278             ret  = vub_writev(req, &elem->out_sg[1], out_num);
279         } else {
280             ret = vub_readv(req, &elem->in_sg[0], in_num);
281         }
282         if (ret >= 0) {
283             req->in->status = VIRTIO_BLK_S_OK;
284         } else {
285             req->in->status = VIRTIO_BLK_S_IOERR;
286         }
287         vub_req_complete(req);
288         break;
289     }
290     case VIRTIO_BLK_T_FLUSH:
291         vub_flush(req);
292         req->in->status = VIRTIO_BLK_S_OK;
293         vub_req_complete(req);
294         break;
295     case VIRTIO_BLK_T_GET_ID: {
296         size_t size = MIN(vub_iov_size(&elem->in_sg[0], in_num),
297                           VIRTIO_BLK_ID_BYTES);
298         snprintf(elem->in_sg[0].iov_base, size, "%s", "vhost_user_blk");
299         req->in->status = VIRTIO_BLK_S_OK;
300         req->size = elem->in_sg[0].iov_len;
301         vub_req_complete(req);
302         break;
303     }
304     case VIRTIO_BLK_T_DISCARD:
305     case VIRTIO_BLK_T_WRITE_ZEROES: {
306         int rc;
307         rc = vub_discard_write_zeroes(req, &elem->out_sg[1], out_num, type);
308         if (rc == 0) {
309             req->in->status = VIRTIO_BLK_S_OK;
310         } else {
311             req->in->status = VIRTIO_BLK_S_IOERR;
312         }
313         vub_req_complete(req);
314         break;
315     }
316     default:
317         req->in->status = VIRTIO_BLK_S_UNSUPP;
318         vub_req_complete(req);
319         break;
320     }
321 
322     return 0;
323 
324 err:
325     g_free(elem);
326     g_free(req);
327     return -1;
328 }
329 
330 static void vub_process_vq(VuDev *vu_dev, int idx)
331 {
332     VugDev *gdev;
333     VubDev *vdev_blk;
334     VuVirtq *vq;
335     int ret;
336 
337     gdev = container_of(vu_dev, VugDev, parent);
338     vdev_blk = container_of(gdev, VubDev, parent);
339     assert(vdev_blk);
340 
341     vq = vu_get_queue(vu_dev, idx);
342     assert(vq);
343 
344     while (1) {
345         ret = vub_virtio_process_req(vdev_blk, vq);
346         if (ret) {
347             break;
348         }
349     }
350 }
351 
352 static void vub_queue_set_started(VuDev *vu_dev, int idx, bool started)
353 {
354     VuVirtq *vq;
355 
356     assert(vu_dev);
357 
358     vq = vu_get_queue(vu_dev, idx);
359     vu_set_queue_handler(vu_dev, vq, started ? vub_process_vq : NULL);
360 }
361 
362 static uint64_t
363 vub_get_features(VuDev *dev)
364 {
365     uint64_t features;
366     VugDev *gdev;
367     VubDev *vdev_blk;
368 
369     gdev = container_of(dev, VugDev, parent);
370     vdev_blk = container_of(gdev, VubDev, parent);
371 
372     features = 1ull << VIRTIO_BLK_F_SIZE_MAX |
373                1ull << VIRTIO_BLK_F_SEG_MAX |
374                1ull << VIRTIO_BLK_F_TOPOLOGY |
375                1ull << VIRTIO_BLK_F_BLK_SIZE |
376                1ull << VIRTIO_BLK_F_FLUSH |
377                #if defined(__linux__) && defined(BLKDISCARD) && defined(BLKZEROOUT)
378                1ull << VIRTIO_BLK_F_DISCARD |
379                1ull << VIRTIO_BLK_F_WRITE_ZEROES |
380                #endif
381                1ull << VIRTIO_BLK_F_CONFIG_WCE;
382 
383     if (vdev_blk->enable_ro) {
384         features |= 1ull << VIRTIO_BLK_F_RO;
385     }
386 
387     return features;
388 }
389 
390 static uint64_t
391 vub_get_protocol_features(VuDev *dev)
392 {
393     return 1ull << VHOST_USER_PROTOCOL_F_CONFIG |
394            1ull << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD;
395 }
396 
397 static int
398 vub_get_config(VuDev *vu_dev, uint8_t *config, uint32_t len)
399 {
400     VugDev *gdev;
401     VubDev *vdev_blk;
402 
403     if (len > sizeof(struct virtio_blk_config)) {
404         return -1;
405     }
406 
407     gdev = container_of(vu_dev, VugDev, parent);
408     vdev_blk = container_of(gdev, VubDev, parent);
409     memcpy(config, &vdev_blk->blkcfg, len);
410 
411     return 0;
412 }
413 
414 static int
415 vub_set_config(VuDev *vu_dev, const uint8_t *data,
416                uint32_t offset, uint32_t size, uint32_t flags)
417 {
418     VugDev *gdev;
419     VubDev *vdev_blk;
420     uint8_t wce;
421     int fd;
422 
423     /* don't support live migration */
424     if (flags != VHOST_SET_CONFIG_TYPE_MASTER) {
425         return -1;
426     }
427 
428     gdev = container_of(vu_dev, VugDev, parent);
429     vdev_blk = container_of(gdev, VubDev, parent);
430 
431     if (offset != offsetof(struct virtio_blk_config, wce) ||
432         size != 1) {
433         return -1;
434     }
435 
436     wce = *data;
437     if (wce == vdev_blk->blkcfg.wce) {
438         /* Do nothing as same with old configuration */
439         return 0;
440     }
441 
442     vdev_blk->blkcfg.wce = wce;
443     fprintf(stdout, "Write Cache Policy Changed\n");
444     if (vdev_blk->blk_fd >= 0) {
445         close(vdev_blk->blk_fd);
446         vdev_blk->blk_fd = -1;
447     }
448 
449     fd = vub_open(vdev_blk->blk_name, wce);
450     if (fd < 0) {
451         fprintf(stderr, "Error to open block device %s\n", vdev_blk->blk_name);
452         vdev_blk->blk_fd = -1;
453         return -1;
454     }
455     vdev_blk->blk_fd = fd;
456 
457     return 0;
458 }
459 
460 static const VuDevIface vub_iface = {
461     .get_features = vub_get_features,
462     .queue_set_started = vub_queue_set_started,
463     .get_protocol_features = vub_get_protocol_features,
464     .get_config = vub_get_config,
465     .set_config = vub_set_config,
466 };
467 
468 static int unix_sock_new(char *unix_fn)
469 {
470     int sock;
471     struct sockaddr_un un;
472     size_t len;
473 
474     assert(unix_fn);
475 
476     sock = socket(AF_UNIX, SOCK_STREAM, 0);
477     if (sock < 0) {
478         perror("socket");
479         return -1;
480     }
481 
482     un.sun_family = AF_UNIX;
483     (void)snprintf(un.sun_path, sizeof(un.sun_path), "%s", unix_fn);
484     len = sizeof(un.sun_family) + strlen(un.sun_path);
485 
486     (void)unlink(unix_fn);
487     if (bind(sock, (struct sockaddr *)&un, len) < 0) {
488         perror("bind");
489         goto fail;
490     }
491 
492     if (listen(sock, 1) < 0) {
493         perror("listen");
494         goto fail;
495     }
496 
497     return sock;
498 
499 fail:
500     (void)close(sock);
501 
502     return -1;
503 }
504 
505 static void vub_free(struct VubDev *vdev_blk)
506 {
507     if (!vdev_blk) {
508         return;
509     }
510 
511     g_main_loop_unref(vdev_blk->loop);
512     if (vdev_blk->blk_fd >= 0) {
513         close(vdev_blk->blk_fd);
514     }
515     g_free(vdev_blk);
516 }
517 
518 static uint32_t
519 vub_get_blocksize(int fd)
520 {
521     uint32_t blocksize = 512;
522 
523 #if defined(__linux__) && defined(BLKSSZGET)
524     if (ioctl(fd, BLKSSZGET, &blocksize) == 0) {
525         return blocksize;
526     }
527 #endif
528 
529     return blocksize;
530 }
531 
532 static void
533 vub_initialize_config(int fd, struct virtio_blk_config *config)
534 {
535     off_t capacity;
536 
537     capacity = lseek(fd, 0, SEEK_END);
538     config->capacity = capacity >> 9;
539     config->blk_size = vub_get_blocksize(fd);
540     config->size_max = 65536;
541     config->seg_max = 128 - 2;
542     config->min_io_size = 1;
543     config->opt_io_size = 1;
544     config->num_queues = 1;
545     #if defined(__linux__) && defined(BLKDISCARD) && defined(BLKZEROOUT)
546     config->max_discard_sectors = 32768;
547     config->max_discard_seg = 1;
548     config->discard_sector_alignment = config->blk_size >> 9;
549     config->max_write_zeroes_sectors = 32768;
550     config->max_write_zeroes_seg = 1;
551     #endif
552 }
553 
554 static VubDev *
555 vub_new(char *blk_file)
556 {
557     VubDev *vdev_blk;
558 
559     vdev_blk = g_new0(VubDev, 1);
560     vdev_blk->loop = g_main_loop_new(NULL, FALSE);
561     vdev_blk->blk_fd = vub_open(blk_file, 0);
562     if (vdev_blk->blk_fd  < 0) {
563         fprintf(stderr, "Error to open block device %s\n", blk_file);
564         vub_free(vdev_blk);
565         return NULL;
566     }
567     vdev_blk->enable_ro = false;
568     vdev_blk->blkcfg.wce = 0;
569     vdev_blk->blk_name = blk_file;
570 
571     /* fill virtio_blk_config with block parameters */
572     vub_initialize_config(vdev_blk->blk_fd, &vdev_blk->blkcfg);
573 
574     return vdev_blk;
575 }
576 
577 static int opt_fdnum = -1;
578 static char *opt_socket_path;
579 static char *opt_blk_file;
580 static gboolean opt_print_caps;
581 static gboolean opt_read_only;
582 
583 static GOptionEntry entries[] = {
584     { "print-capabilities", 'c', 0, G_OPTION_ARG_NONE, &opt_print_caps,
585       "Print capabilities", NULL },
586     { "fd", 'f', 0, G_OPTION_ARG_INT, &opt_fdnum,
587       "Use inherited fd socket", "FDNUM" },
588     { "socket-path", 's', 0, G_OPTION_ARG_FILENAME, &opt_socket_path,
589       "Use UNIX socket path", "PATH" },
590     {"blk-file", 'b', 0, G_OPTION_ARG_FILENAME, &opt_blk_file,
591      "block device or file path", "PATH"},
592     { "read-only", 'r', 0, G_OPTION_ARG_NONE, &opt_read_only,
593       "Enable read-only", NULL },
594     { NULL, },
595 };
596 
597 int main(int argc, char **argv)
598 {
599     int lsock = -1, csock = -1;
600     VubDev *vdev_blk = NULL;
601     GError *error = NULL;
602     GOptionContext *context;
603 
604     context = g_option_context_new(NULL);
605     g_option_context_add_main_entries(context, entries, NULL);
606     if (!g_option_context_parse(context, &argc, &argv, &error)) {
607         g_printerr("Option parsing failed: %s\n", error->message);
608         exit(EXIT_FAILURE);
609     }
610     if (opt_print_caps) {
611         g_print("{\n");
612         g_print("  \"type\": \"block\",\n");
613         g_print("  \"features\": [\n");
614         g_print("    \"read-only\",\n");
615         g_print("    \"blk-file\"\n");
616         g_print("  ]\n");
617         g_print("}\n");
618         exit(EXIT_SUCCESS);
619     }
620 
621     if (!opt_blk_file) {
622         g_print("%s\n", g_option_context_get_help(context, true, NULL));
623         exit(EXIT_FAILURE);
624     }
625 
626     if (opt_socket_path) {
627         lsock = unix_sock_new(opt_socket_path);
628         if (lsock < 0) {
629             exit(EXIT_FAILURE);
630         }
631     } else if (opt_fdnum < 0) {
632         g_print("%s\n", g_option_context_get_help(context, true, NULL));
633         exit(EXIT_FAILURE);
634     } else {
635         lsock = opt_fdnum;
636     }
637 
638     csock = accept(lsock, NULL, NULL);
639     if (csock < 0) {
640         g_printerr("Accept error %s\n", strerror(errno));
641         exit(EXIT_FAILURE);
642     }
643 
644     vdev_blk = vub_new(opt_blk_file);
645     if (!vdev_blk) {
646         exit(EXIT_FAILURE);
647     }
648     if (opt_read_only) {
649         vdev_blk->enable_ro = true;
650     }
651 
652     if (!vug_init(&vdev_blk->parent, VHOST_USER_BLK_MAX_QUEUES, csock,
653                   vub_panic_cb, &vub_iface)) {
654         g_printerr("Failed to initialize libvhost-user-glib\n");
655         exit(EXIT_FAILURE);
656     }
657 
658     g_main_loop_run(vdev_blk->loop);
659     g_main_loop_unref(vdev_blk->loop);
660     g_option_context_free(context);
661     vug_deinit(&vdev_blk->parent);
662     vub_free(vdev_blk);
663     if (csock >= 0) {
664         close(csock);
665     }
666     if (lsock >= 0) {
667         close(lsock);
668     }
669     g_free(opt_socket_path);
670     g_free(opt_blk_file);
671 
672     return 0;
673 }
674