1 /*
2  * Sharing QEMU block devices via vhost-user protocal
3  *
4  * Parts of the code based on nbd/server.c.
5  *
6  * Copyright (c) Coiby Xu <coiby.xu@gmail.com>.
7  * Copyright (c) 2020 Red Hat, Inc.
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2 or
10  * later.  See the COPYING file in the top-level directory.
11  */
12 #include "qemu/osdep.h"
13 #include "block/block.h"
14 #include "subprojects/libvhost-user/libvhost-user.h" /* only for the type definitions */
15 #include "standard-headers/linux/virtio_blk.h"
16 #include "qemu/vhost-user-server.h"
17 #include "vhost-user-blk-server.h"
18 #include "qapi/error.h"
19 #include "qom/object_interfaces.h"
20 #include "util/block-helpers.h"
21 #include "virtio-blk-handler.h"
22 
23 enum {
24     VHOST_USER_BLK_NUM_QUEUES_DEFAULT = 1,
25 };
26 
27 typedef struct VuBlkReq {
28     VuVirtqElement elem;
29     VuServer *server;
30     struct VuVirtq *vq;
31 } VuBlkReq;
32 
33 /* vhost user block device */
34 typedef struct {
35     BlockExport export;
36     VuServer vu_server;
37     VirtioBlkHandler handler;
38     QIOChannelSocket *sioc;
39     struct virtio_blk_config blkcfg;
40 } VuBlkExport;
41 
42 static void vu_blk_req_complete(VuBlkReq *req, size_t in_len)
43 {
44     VuDev *vu_dev = &req->server->vu_dev;
45 
46     vu_queue_push(vu_dev, req->vq, &req->elem, in_len);
47     vu_queue_notify(vu_dev, req->vq);
48 
49     free(req);
50 }
51 
52 /* Called with server refcount increased, must decrease before returning */
53 static void coroutine_fn vu_blk_virtio_process_req(void *opaque)
54 {
55     VuBlkReq *req = opaque;
56     VuServer *server = req->server;
57     VuVirtqElement *elem = &req->elem;
58     VuBlkExport *vexp = container_of(server, VuBlkExport, vu_server);
59     VirtioBlkHandler *handler = &vexp->handler;
60     struct iovec *in_iov = elem->in_sg;
61     struct iovec *out_iov = elem->out_sg;
62     unsigned in_num = elem->in_num;
63     unsigned out_num = elem->out_num;
64     int in_len;
65 
66     in_len = virtio_blk_process_req(handler, in_iov, out_iov,
67                                     in_num, out_num);
68     if (in_len < 0) {
69         free(req);
70         vhost_user_server_unref(server);
71         return;
72     }
73 
74     vu_blk_req_complete(req, in_len);
75     vhost_user_server_unref(server);
76 }
77 
78 static void vu_blk_process_vq(VuDev *vu_dev, int idx)
79 {
80     VuServer *server = container_of(vu_dev, VuServer, vu_dev);
81     VuVirtq *vq = vu_get_queue(vu_dev, idx);
82 
83     while (1) {
84         VuBlkReq *req;
85 
86         req = vu_queue_pop(vu_dev, vq, sizeof(VuBlkReq));
87         if (!req) {
88             break;
89         }
90 
91         req->server = server;
92         req->vq = vq;
93 
94         Coroutine *co =
95             qemu_coroutine_create(vu_blk_virtio_process_req, req);
96 
97         vhost_user_server_ref(server);
98         qemu_coroutine_enter(co);
99     }
100 }
101 
102 static void vu_blk_queue_set_started(VuDev *vu_dev, int idx, bool started)
103 {
104     VuVirtq *vq;
105 
106     assert(vu_dev);
107 
108     vq = vu_get_queue(vu_dev, idx);
109     vu_set_queue_handler(vu_dev, vq, started ? vu_blk_process_vq : NULL);
110 }
111 
112 static uint64_t vu_blk_get_features(VuDev *dev)
113 {
114     uint64_t features;
115     VuServer *server = container_of(dev, VuServer, vu_dev);
116     VuBlkExport *vexp = container_of(server, VuBlkExport, vu_server);
117     features = 1ull << VIRTIO_BLK_F_SIZE_MAX |
118                1ull << VIRTIO_BLK_F_SEG_MAX |
119                1ull << VIRTIO_BLK_F_TOPOLOGY |
120                1ull << VIRTIO_BLK_F_BLK_SIZE |
121                1ull << VIRTIO_BLK_F_FLUSH |
122                1ull << VIRTIO_BLK_F_DISCARD |
123                1ull << VIRTIO_BLK_F_WRITE_ZEROES |
124                1ull << VIRTIO_BLK_F_CONFIG_WCE |
125                1ull << VIRTIO_BLK_F_MQ |
126                1ull << VIRTIO_F_VERSION_1 |
127                1ull << VIRTIO_RING_F_INDIRECT_DESC |
128                1ull << VIRTIO_RING_F_EVENT_IDX |
129                1ull << VHOST_USER_F_PROTOCOL_FEATURES;
130 
131     if (!vexp->handler.writable) {
132         features |= 1ull << VIRTIO_BLK_F_RO;
133     }
134 
135     return features;
136 }
137 
138 static uint64_t vu_blk_get_protocol_features(VuDev *dev)
139 {
140     return 1ull << VHOST_USER_PROTOCOL_F_CONFIG;
141 }
142 
143 static int
144 vu_blk_get_config(VuDev *vu_dev, uint8_t *config, uint32_t len)
145 {
146     VuServer *server = container_of(vu_dev, VuServer, vu_dev);
147     VuBlkExport *vexp = container_of(server, VuBlkExport, vu_server);
148 
149     if (len > sizeof(struct virtio_blk_config)) {
150         return -1;
151     }
152 
153     memcpy(config, &vexp->blkcfg, len);
154     return 0;
155 }
156 
157 static int
158 vu_blk_set_config(VuDev *vu_dev, const uint8_t *data,
159                     uint32_t offset, uint32_t size, uint32_t flags)
160 {
161     VuServer *server = container_of(vu_dev, VuServer, vu_dev);
162     VuBlkExport *vexp = container_of(server, VuBlkExport, vu_server);
163     uint8_t wce;
164 
165     /* don't support live migration */
166     if (flags != VHOST_SET_CONFIG_TYPE_MASTER) {
167         return -EINVAL;
168     }
169 
170     if (offset != offsetof(struct virtio_blk_config, wce) ||
171         size != 1) {
172         return -EINVAL;
173     }
174 
175     wce = *data;
176     vexp->blkcfg.wce = wce;
177     blk_set_enable_write_cache(vexp->export.blk, wce);
178     return 0;
179 }
180 
181 /*
182  * When the client disconnects, it sends a VHOST_USER_NONE request
183  * and vu_process_message will simple call exit which cause the VM
184  * to exit abruptly.
185  * To avoid this issue,  process VHOST_USER_NONE request ahead
186  * of vu_process_message.
187  *
188  */
189 static int vu_blk_process_msg(VuDev *dev, VhostUserMsg *vmsg, int *do_reply)
190 {
191     if (vmsg->request == VHOST_USER_NONE) {
192         dev->panic(dev, "disconnect");
193         return true;
194     }
195     return false;
196 }
197 
198 static const VuDevIface vu_blk_iface = {
199     .get_features          = vu_blk_get_features,
200     .queue_set_started     = vu_blk_queue_set_started,
201     .get_protocol_features = vu_blk_get_protocol_features,
202     .get_config            = vu_blk_get_config,
203     .set_config            = vu_blk_set_config,
204     .process_msg           = vu_blk_process_msg,
205 };
206 
207 static void blk_aio_attached(AioContext *ctx, void *opaque)
208 {
209     VuBlkExport *vexp = opaque;
210 
211     vexp->export.ctx = ctx;
212     vhost_user_server_attach_aio_context(&vexp->vu_server, ctx);
213 }
214 
215 static void blk_aio_detach(void *opaque)
216 {
217     VuBlkExport *vexp = opaque;
218 
219     vhost_user_server_detach_aio_context(&vexp->vu_server);
220     vexp->export.ctx = NULL;
221 }
222 
223 static void
224 vu_blk_initialize_config(BlockDriverState *bs,
225                          struct virtio_blk_config *config,
226                          uint32_t blk_size,
227                          uint16_t num_queues)
228 {
229     config->capacity =
230         cpu_to_le64(bdrv_getlength(bs) >> VIRTIO_BLK_SECTOR_BITS);
231     config->blk_size = cpu_to_le32(blk_size);
232     config->size_max = cpu_to_le32(0);
233     config->seg_max = cpu_to_le32(128 - 2);
234     config->min_io_size = cpu_to_le16(1);
235     config->opt_io_size = cpu_to_le32(1);
236     config->num_queues = cpu_to_le16(num_queues);
237     config->max_discard_sectors =
238         cpu_to_le32(VIRTIO_BLK_MAX_DISCARD_SECTORS);
239     config->max_discard_seg = cpu_to_le32(1);
240     config->discard_sector_alignment =
241         cpu_to_le32(blk_size >> VIRTIO_BLK_SECTOR_BITS);
242     config->max_write_zeroes_sectors
243         = cpu_to_le32(VIRTIO_BLK_MAX_WRITE_ZEROES_SECTORS);
244     config->max_write_zeroes_seg = cpu_to_le32(1);
245 }
246 
247 static void vu_blk_exp_request_shutdown(BlockExport *exp)
248 {
249     VuBlkExport *vexp = container_of(exp, VuBlkExport, export);
250 
251     vhost_user_server_stop(&vexp->vu_server);
252 }
253 
254 static int vu_blk_exp_create(BlockExport *exp, BlockExportOptions *opts,
255                              Error **errp)
256 {
257     VuBlkExport *vexp = container_of(exp, VuBlkExport, export);
258     BlockExportOptionsVhostUserBlk *vu_opts = &opts->u.vhost_user_blk;
259     Error *local_err = NULL;
260     uint64_t logical_block_size;
261     uint16_t num_queues = VHOST_USER_BLK_NUM_QUEUES_DEFAULT;
262 
263     vexp->blkcfg.wce = 0;
264 
265     if (vu_opts->has_logical_block_size) {
266         logical_block_size = vu_opts->logical_block_size;
267     } else {
268         logical_block_size = VIRTIO_BLK_SECTOR_SIZE;
269     }
270     check_block_size(exp->id, "logical-block-size", logical_block_size,
271                      &local_err);
272     if (local_err) {
273         error_propagate(errp, local_err);
274         return -EINVAL;
275     }
276 
277     if (vu_opts->has_num_queues) {
278         num_queues = vu_opts->num_queues;
279     }
280     if (num_queues == 0) {
281         error_setg(errp, "num-queues must be greater than 0");
282         return -EINVAL;
283     }
284     vexp->handler.blk = exp->blk;
285     vexp->handler.serial = g_strdup("vhost_user_blk");
286     vexp->handler.logical_block_size = logical_block_size;
287     vexp->handler.writable = opts->writable;
288 
289     vu_blk_initialize_config(blk_bs(exp->blk), &vexp->blkcfg,
290                              logical_block_size, num_queues);
291 
292     blk_add_aio_context_notifier(exp->blk, blk_aio_attached, blk_aio_detach,
293                                  vexp);
294 
295     if (!vhost_user_server_start(&vexp->vu_server, vu_opts->addr, exp->ctx,
296                                  num_queues, &vu_blk_iface, errp)) {
297         blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
298                                         blk_aio_detach, vexp);
299         g_free(vexp->handler.serial);
300         return -EADDRNOTAVAIL;
301     }
302 
303     return 0;
304 }
305 
306 static void vu_blk_exp_delete(BlockExport *exp)
307 {
308     VuBlkExport *vexp = container_of(exp, VuBlkExport, export);
309 
310     blk_remove_aio_context_notifier(exp->blk, blk_aio_attached, blk_aio_detach,
311                                     vexp);
312     g_free(vexp->handler.serial);
313 }
314 
315 const BlockExportDriver blk_exp_vhost_user_blk = {
316     .type               = BLOCK_EXPORT_TYPE_VHOST_USER_BLK,
317     .instance_size      = sizeof(VuBlkExport),
318     .create             = vu_blk_exp_create,
319     .delete             = vu_blk_exp_delete,
320     .request_shutdown   = vu_blk_exp_request_shutdown,
321 };
322