xref: /openbmc/qemu/hw/virtio/vhost-user.c (revision ffe98631)
1 /*
2  * vhost-user
3  *
4  * Copyright (c) 2013 Virtual Open Systems Sarl.
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
7  * See the COPYING file in the top-level directory.
8  *
9  */
10 
11 #include "qemu/osdep.h"
12 #include "qapi/error.h"
13 #include "hw/virtio/vhost.h"
14 #include "hw/virtio/vhost-user.h"
15 #include "hw/virtio/vhost-backend.h"
16 #include "hw/virtio/virtio.h"
17 #include "hw/virtio/virtio-net.h"
18 #include "chardev/char-fe.h"
19 #include "io/channel-socket.h"
20 #include "sysemu/kvm.h"
21 #include "qemu/error-report.h"
22 #include "qemu/main-loop.h"
23 #include "qemu/sockets.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/cryptodev.h"
26 #include "migration/migration.h"
27 #include "migration/postcopy-ram.h"
28 #include "trace.h"
29 #include "exec/ramblock.h"
30 
31 #include <sys/ioctl.h>
32 #include <sys/socket.h>
33 #include <sys/un.h>
34 
35 #include "standard-headers/linux/vhost_types.h"
36 
37 #ifdef CONFIG_LINUX
38 #include <linux/userfaultfd.h>
39 #endif
40 
41 #define VHOST_MEMORY_BASELINE_NREGIONS    8
42 #define VHOST_USER_F_PROTOCOL_FEATURES 30
43 #define VHOST_USER_SLAVE_MAX_FDS     8
44 
45 /*
46  * Set maximum number of RAM slots supported to
47  * the maximum number supported by the target
48  * hardware plaform.
49  */
50 #if defined(TARGET_X86) || defined(TARGET_X86_64) || \
51     defined(TARGET_ARM) || defined(TARGET_ARM_64)
52 #include "hw/acpi/acpi.h"
53 #define VHOST_USER_MAX_RAM_SLOTS ACPI_MAX_RAM_SLOTS
54 
55 #elif defined(TARGET_PPC) || defined(TARGET_PPC64)
56 #include "hw/ppc/spapr.h"
57 #define VHOST_USER_MAX_RAM_SLOTS SPAPR_MAX_RAM_SLOTS
58 
59 #else
60 #define VHOST_USER_MAX_RAM_SLOTS 512
61 #endif
62 
63 /*
64  * Maximum size of virtio device config space
65  */
66 #define VHOST_USER_MAX_CONFIG_SIZE 256
67 
68 enum VhostUserProtocolFeature {
69     VHOST_USER_PROTOCOL_F_MQ = 0,
70     VHOST_USER_PROTOCOL_F_LOG_SHMFD = 1,
71     VHOST_USER_PROTOCOL_F_RARP = 2,
72     VHOST_USER_PROTOCOL_F_REPLY_ACK = 3,
73     VHOST_USER_PROTOCOL_F_NET_MTU = 4,
74     VHOST_USER_PROTOCOL_F_SLAVE_REQ = 5,
75     VHOST_USER_PROTOCOL_F_CROSS_ENDIAN = 6,
76     VHOST_USER_PROTOCOL_F_CRYPTO_SESSION = 7,
77     VHOST_USER_PROTOCOL_F_PAGEFAULT = 8,
78     VHOST_USER_PROTOCOL_F_CONFIG = 9,
79     VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD = 10,
80     VHOST_USER_PROTOCOL_F_HOST_NOTIFIER = 11,
81     VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD = 12,
82     VHOST_USER_PROTOCOL_F_RESET_DEVICE = 13,
83     /* Feature 14 reserved for VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS. */
84     VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS = 15,
85     VHOST_USER_PROTOCOL_F_STATUS = 16,
86     VHOST_USER_PROTOCOL_F_MAX
87 };
88 
89 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
90 
91 typedef enum VhostUserRequest {
92     VHOST_USER_NONE = 0,
93     VHOST_USER_GET_FEATURES = 1,
94     VHOST_USER_SET_FEATURES = 2,
95     VHOST_USER_SET_OWNER = 3,
96     VHOST_USER_RESET_OWNER = 4,
97     VHOST_USER_SET_MEM_TABLE = 5,
98     VHOST_USER_SET_LOG_BASE = 6,
99     VHOST_USER_SET_LOG_FD = 7,
100     VHOST_USER_SET_VRING_NUM = 8,
101     VHOST_USER_SET_VRING_ADDR = 9,
102     VHOST_USER_SET_VRING_BASE = 10,
103     VHOST_USER_GET_VRING_BASE = 11,
104     VHOST_USER_SET_VRING_KICK = 12,
105     VHOST_USER_SET_VRING_CALL = 13,
106     VHOST_USER_SET_VRING_ERR = 14,
107     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
108     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
109     VHOST_USER_GET_QUEUE_NUM = 17,
110     VHOST_USER_SET_VRING_ENABLE = 18,
111     VHOST_USER_SEND_RARP = 19,
112     VHOST_USER_NET_SET_MTU = 20,
113     VHOST_USER_SET_SLAVE_REQ_FD = 21,
114     VHOST_USER_IOTLB_MSG = 22,
115     VHOST_USER_SET_VRING_ENDIAN = 23,
116     VHOST_USER_GET_CONFIG = 24,
117     VHOST_USER_SET_CONFIG = 25,
118     VHOST_USER_CREATE_CRYPTO_SESSION = 26,
119     VHOST_USER_CLOSE_CRYPTO_SESSION = 27,
120     VHOST_USER_POSTCOPY_ADVISE  = 28,
121     VHOST_USER_POSTCOPY_LISTEN  = 29,
122     VHOST_USER_POSTCOPY_END     = 30,
123     VHOST_USER_GET_INFLIGHT_FD = 31,
124     VHOST_USER_SET_INFLIGHT_FD = 32,
125     VHOST_USER_GPU_SET_SOCKET = 33,
126     VHOST_USER_RESET_DEVICE = 34,
127     /* Message number 35 reserved for VHOST_USER_VRING_KICK. */
128     VHOST_USER_GET_MAX_MEM_SLOTS = 36,
129     VHOST_USER_ADD_MEM_REG = 37,
130     VHOST_USER_REM_MEM_REG = 38,
131     VHOST_USER_SET_STATUS = 39,
132     VHOST_USER_GET_STATUS = 40,
133     VHOST_USER_MAX
134 } VhostUserRequest;
135 
136 typedef enum VhostUserSlaveRequest {
137     VHOST_USER_SLAVE_NONE = 0,
138     VHOST_USER_SLAVE_IOTLB_MSG = 1,
139     VHOST_USER_SLAVE_CONFIG_CHANGE_MSG = 2,
140     VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG = 3,
141     VHOST_USER_SLAVE_MAX
142 }  VhostUserSlaveRequest;
143 
144 typedef struct VhostUserMemoryRegion {
145     uint64_t guest_phys_addr;
146     uint64_t memory_size;
147     uint64_t userspace_addr;
148     uint64_t mmap_offset;
149 } VhostUserMemoryRegion;
150 
151 typedef struct VhostUserMemory {
152     uint32_t nregions;
153     uint32_t padding;
154     VhostUserMemoryRegion regions[VHOST_MEMORY_BASELINE_NREGIONS];
155 } VhostUserMemory;
156 
157 typedef struct VhostUserMemRegMsg {
158     uint64_t padding;
159     VhostUserMemoryRegion region;
160 } VhostUserMemRegMsg;
161 
162 typedef struct VhostUserLog {
163     uint64_t mmap_size;
164     uint64_t mmap_offset;
165 } VhostUserLog;
166 
167 typedef struct VhostUserConfig {
168     uint32_t offset;
169     uint32_t size;
170     uint32_t flags;
171     uint8_t region[VHOST_USER_MAX_CONFIG_SIZE];
172 } VhostUserConfig;
173 
174 #define VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN    512
175 #define VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN  64
176 
177 typedef struct VhostUserCryptoSession {
178     /* session id for success, -1 on errors */
179     int64_t session_id;
180     CryptoDevBackendSymSessionInfo session_setup_data;
181     uint8_t key[VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN];
182     uint8_t auth_key[VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN];
183 } VhostUserCryptoSession;
184 
185 static VhostUserConfig c __attribute__ ((unused));
186 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
187                                    + sizeof(c.size) \
188                                    + sizeof(c.flags))
189 
190 typedef struct VhostUserVringArea {
191     uint64_t u64;
192     uint64_t size;
193     uint64_t offset;
194 } VhostUserVringArea;
195 
196 typedef struct VhostUserInflight {
197     uint64_t mmap_size;
198     uint64_t mmap_offset;
199     uint16_t num_queues;
200     uint16_t queue_size;
201 } VhostUserInflight;
202 
203 typedef struct {
204     VhostUserRequest request;
205 
206 #define VHOST_USER_VERSION_MASK     (0x3)
207 #define VHOST_USER_REPLY_MASK       (0x1 << 2)
208 #define VHOST_USER_NEED_REPLY_MASK  (0x1 << 3)
209     uint32_t flags;
210     uint32_t size; /* the following payload size */
211 } QEMU_PACKED VhostUserHeader;
212 
213 typedef union {
214 #define VHOST_USER_VRING_IDX_MASK   (0xff)
215 #define VHOST_USER_VRING_NOFD_MASK  (0x1 << 8)
216         uint64_t u64;
217         struct vhost_vring_state state;
218         struct vhost_vring_addr addr;
219         VhostUserMemory memory;
220         VhostUserMemRegMsg mem_reg;
221         VhostUserLog log;
222         struct vhost_iotlb_msg iotlb;
223         VhostUserConfig config;
224         VhostUserCryptoSession session;
225         VhostUserVringArea area;
226         VhostUserInflight inflight;
227 } VhostUserPayload;
228 
229 typedef struct VhostUserMsg {
230     VhostUserHeader hdr;
231     VhostUserPayload payload;
232 } QEMU_PACKED VhostUserMsg;
233 
234 static VhostUserMsg m __attribute__ ((unused));
235 #define VHOST_USER_HDR_SIZE (sizeof(VhostUserHeader))
236 
237 #define VHOST_USER_PAYLOAD_SIZE (sizeof(VhostUserPayload))
238 
239 /* The version of the protocol we support */
240 #define VHOST_USER_VERSION    (0x1)
241 
242 struct vhost_user {
243     struct vhost_dev *dev;
244     /* Shared between vhost devs of the same virtio device */
245     VhostUserState *user;
246     QIOChannel *slave_ioc;
247     GSource *slave_src;
248     NotifierWithReturn postcopy_notifier;
249     struct PostCopyFD  postcopy_fd;
250     uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
251     /* Length of the region_rb and region_rb_offset arrays */
252     size_t             region_rb_len;
253     /* RAMBlock associated with a given region */
254     RAMBlock         **region_rb;
255     /*
256      * The offset from the start of the RAMBlock to the start of the
257      * vhost region.
258      */
259     ram_addr_t        *region_rb_offset;
260 
261     /* True once we've entered postcopy_listen */
262     bool               postcopy_listen;
263 
264     /* Our current regions */
265     int num_shadow_regions;
266     struct vhost_memory_region shadow_regions[VHOST_USER_MAX_RAM_SLOTS];
267 };
268 
269 struct scrub_regions {
270     struct vhost_memory_region *region;
271     int reg_idx;
272     int fd_idx;
273 };
274 
275 static bool ioeventfd_enabled(void)
276 {
277     return !kvm_enabled() || kvm_eventfds_enabled();
278 }
279 
280 static int vhost_user_read_header(struct vhost_dev *dev, VhostUserMsg *msg)
281 {
282     struct vhost_user *u = dev->opaque;
283     CharBackend *chr = u->user->chr;
284     uint8_t *p = (uint8_t *) msg;
285     int r, size = VHOST_USER_HDR_SIZE;
286 
287     r = qemu_chr_fe_read_all(chr, p, size);
288     if (r != size) {
289         int saved_errno = errno;
290         error_report("Failed to read msg header. Read %d instead of %d."
291                      " Original request %d.", r, size, msg->hdr.request);
292         return r < 0 ? -saved_errno : -EIO;
293     }
294 
295     /* validate received flags */
296     if (msg->hdr.flags != (VHOST_USER_REPLY_MASK | VHOST_USER_VERSION)) {
297         error_report("Failed to read msg header."
298                 " Flags 0x%x instead of 0x%x.", msg->hdr.flags,
299                 VHOST_USER_REPLY_MASK | VHOST_USER_VERSION);
300         return -EPROTO;
301     }
302 
303     trace_vhost_user_read(msg->hdr.request, msg->hdr.flags);
304 
305     return 0;
306 }
307 
308 struct vhost_user_read_cb_data {
309     struct vhost_dev *dev;
310     VhostUserMsg *msg;
311     GMainLoop *loop;
312     int ret;
313 };
314 
315 static gboolean vhost_user_read_cb(void *do_not_use, GIOCondition condition,
316                                    gpointer opaque)
317 {
318     struct vhost_user_read_cb_data *data = opaque;
319     struct vhost_dev *dev = data->dev;
320     VhostUserMsg *msg = data->msg;
321     struct vhost_user *u = dev->opaque;
322     CharBackend *chr = u->user->chr;
323     uint8_t *p = (uint8_t *) msg;
324     int r, size;
325 
326     r = vhost_user_read_header(dev, msg);
327     if (r < 0) {
328         data->ret = r;
329         goto end;
330     }
331 
332     /* validate message size is sane */
333     if (msg->hdr.size > VHOST_USER_PAYLOAD_SIZE) {
334         error_report("Failed to read msg header."
335                 " Size %d exceeds the maximum %zu.", msg->hdr.size,
336                 VHOST_USER_PAYLOAD_SIZE);
337         data->ret = -EPROTO;
338         goto end;
339     }
340 
341     if (msg->hdr.size) {
342         p += VHOST_USER_HDR_SIZE;
343         size = msg->hdr.size;
344         r = qemu_chr_fe_read_all(chr, p, size);
345         if (r != size) {
346             int saved_errno = errno;
347             error_report("Failed to read msg payload."
348                          " Read %d instead of %d.", r, msg->hdr.size);
349             data->ret = r < 0 ? -saved_errno : -EIO;
350             goto end;
351         }
352     }
353 
354 end:
355     g_main_loop_quit(data->loop);
356     return G_SOURCE_REMOVE;
357 }
358 
359 static gboolean slave_read(QIOChannel *ioc, GIOCondition condition,
360                            gpointer opaque);
361 
362 /*
363  * This updates the read handler to use a new event loop context.
364  * Event sources are removed from the previous context : this ensures
365  * that events detected in the previous context are purged. They will
366  * be re-detected and processed in the new context.
367  */
368 static void slave_update_read_handler(struct vhost_dev *dev,
369                                       GMainContext *ctxt)
370 {
371     struct vhost_user *u = dev->opaque;
372 
373     if (!u->slave_ioc) {
374         return;
375     }
376 
377     if (u->slave_src) {
378         g_source_destroy(u->slave_src);
379         g_source_unref(u->slave_src);
380     }
381 
382     u->slave_src = qio_channel_add_watch_source(u->slave_ioc,
383                                                 G_IO_IN | G_IO_HUP,
384                                                 slave_read, dev, NULL,
385                                                 ctxt);
386 }
387 
388 static int vhost_user_read(struct vhost_dev *dev, VhostUserMsg *msg)
389 {
390     struct vhost_user *u = dev->opaque;
391     CharBackend *chr = u->user->chr;
392     GMainContext *prev_ctxt = chr->chr->gcontext;
393     GMainContext *ctxt = g_main_context_new();
394     GMainLoop *loop = g_main_loop_new(ctxt, FALSE);
395     struct vhost_user_read_cb_data data = {
396         .dev = dev,
397         .loop = loop,
398         .msg = msg,
399         .ret = 0
400     };
401 
402     /*
403      * We want to be able to monitor the slave channel fd while waiting
404      * for chr I/O. This requires an event loop, but we can't nest the
405      * one to which chr is currently attached : its fd handlers might not
406      * be prepared for re-entrancy. So we create a new one and switch chr
407      * to use it.
408      */
409     slave_update_read_handler(dev, ctxt);
410     qemu_chr_be_update_read_handlers(chr->chr, ctxt);
411     qemu_chr_fe_add_watch(chr, G_IO_IN | G_IO_HUP, vhost_user_read_cb, &data);
412 
413     g_main_loop_run(loop);
414 
415     /*
416      * Restore the previous event loop context. This also destroys/recreates
417      * event sources : this guarantees that all pending events in the original
418      * context that have been processed by the nested loop are purged.
419      */
420     qemu_chr_be_update_read_handlers(chr->chr, prev_ctxt);
421     slave_update_read_handler(dev, NULL);
422 
423     g_main_loop_unref(loop);
424     g_main_context_unref(ctxt);
425 
426     return data.ret;
427 }
428 
429 static int process_message_reply(struct vhost_dev *dev,
430                                  const VhostUserMsg *msg)
431 {
432     int ret;
433     VhostUserMsg msg_reply;
434 
435     if ((msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) == 0) {
436         return 0;
437     }
438 
439     ret = vhost_user_read(dev, &msg_reply);
440     if (ret < 0) {
441         return ret;
442     }
443 
444     if (msg_reply.hdr.request != msg->hdr.request) {
445         error_report("Received unexpected msg type. "
446                      "Expected %d received %d",
447                      msg->hdr.request, msg_reply.hdr.request);
448         return -EPROTO;
449     }
450 
451     return msg_reply.payload.u64 ? -EIO : 0;
452 }
453 
454 static bool vhost_user_one_time_request(VhostUserRequest request)
455 {
456     switch (request) {
457     case VHOST_USER_SET_OWNER:
458     case VHOST_USER_RESET_OWNER:
459     case VHOST_USER_SET_MEM_TABLE:
460     case VHOST_USER_GET_QUEUE_NUM:
461     case VHOST_USER_NET_SET_MTU:
462         return true;
463     default:
464         return false;
465     }
466 }
467 
468 /* most non-init callers ignore the error */
469 static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
470                             int *fds, int fd_num)
471 {
472     struct vhost_user *u = dev->opaque;
473     CharBackend *chr = u->user->chr;
474     int ret, size = VHOST_USER_HDR_SIZE + msg->hdr.size;
475 
476     /*
477      * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
478      * we just need send it once in the first time. For later such
479      * request, we just ignore it.
480      */
481     if (vhost_user_one_time_request(msg->hdr.request) && dev->vq_index != 0) {
482         msg->hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
483         return 0;
484     }
485 
486     if (qemu_chr_fe_set_msgfds(chr, fds, fd_num) < 0) {
487         error_report("Failed to set msg fds.");
488         return -EINVAL;
489     }
490 
491     ret = qemu_chr_fe_write_all(chr, (const uint8_t *) msg, size);
492     if (ret != size) {
493         int saved_errno = errno;
494         error_report("Failed to write msg."
495                      " Wrote %d instead of %d.", ret, size);
496         return ret < 0 ? -saved_errno : -EIO;
497     }
498 
499     trace_vhost_user_write(msg->hdr.request, msg->hdr.flags);
500 
501     return 0;
502 }
503 
504 int vhost_user_gpu_set_socket(struct vhost_dev *dev, int fd)
505 {
506     VhostUserMsg msg = {
507         .hdr.request = VHOST_USER_GPU_SET_SOCKET,
508         .hdr.flags = VHOST_USER_VERSION,
509     };
510 
511     return vhost_user_write(dev, &msg, &fd, 1);
512 }
513 
514 static int vhost_user_set_log_base(struct vhost_dev *dev, uint64_t base,
515                                    struct vhost_log *log)
516 {
517     int fds[VHOST_USER_MAX_RAM_SLOTS];
518     size_t fd_num = 0;
519     bool shmfd = virtio_has_feature(dev->protocol_features,
520                                     VHOST_USER_PROTOCOL_F_LOG_SHMFD);
521     int ret;
522     VhostUserMsg msg = {
523         .hdr.request = VHOST_USER_SET_LOG_BASE,
524         .hdr.flags = VHOST_USER_VERSION,
525         .payload.log.mmap_size = log->size * sizeof(*(log->log)),
526         .payload.log.mmap_offset = 0,
527         .hdr.size = sizeof(msg.payload.log),
528     };
529 
530     /* Send only once with first queue pair */
531     if (dev->vq_index != 0) {
532         return 0;
533     }
534 
535     if (shmfd && log->fd != -1) {
536         fds[fd_num++] = log->fd;
537     }
538 
539     ret = vhost_user_write(dev, &msg, fds, fd_num);
540     if (ret < 0) {
541         return ret;
542     }
543 
544     if (shmfd) {
545         msg.hdr.size = 0;
546         ret = vhost_user_read(dev, &msg);
547         if (ret < 0) {
548             return ret;
549         }
550 
551         if (msg.hdr.request != VHOST_USER_SET_LOG_BASE) {
552             error_report("Received unexpected msg type. "
553                          "Expected %d received %d",
554                          VHOST_USER_SET_LOG_BASE, msg.hdr.request);
555             return -EPROTO;
556         }
557     }
558 
559     return 0;
560 }
561 
562 static MemoryRegion *vhost_user_get_mr_data(uint64_t addr, ram_addr_t *offset,
563                                             int *fd)
564 {
565     MemoryRegion *mr;
566 
567     assert((uintptr_t)addr == addr);
568     mr = memory_region_from_host((void *)(uintptr_t)addr, offset);
569     *fd = memory_region_get_fd(mr);
570 
571     return mr;
572 }
573 
574 static void vhost_user_fill_msg_region(VhostUserMemoryRegion *dst,
575                                        struct vhost_memory_region *src,
576                                        uint64_t mmap_offset)
577 {
578     assert(src != NULL && dst != NULL);
579     dst->userspace_addr = src->userspace_addr;
580     dst->memory_size = src->memory_size;
581     dst->guest_phys_addr = src->guest_phys_addr;
582     dst->mmap_offset = mmap_offset;
583 }
584 
585 static int vhost_user_fill_set_mem_table_msg(struct vhost_user *u,
586                                              struct vhost_dev *dev,
587                                              VhostUserMsg *msg,
588                                              int *fds, size_t *fd_num,
589                                              bool track_ramblocks)
590 {
591     int i, fd;
592     ram_addr_t offset;
593     MemoryRegion *mr;
594     struct vhost_memory_region *reg;
595     VhostUserMemoryRegion region_buffer;
596 
597     msg->hdr.request = VHOST_USER_SET_MEM_TABLE;
598 
599     for (i = 0; i < dev->mem->nregions; ++i) {
600         reg = dev->mem->regions + i;
601 
602         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
603         if (fd > 0) {
604             if (track_ramblocks) {
605                 assert(*fd_num < VHOST_MEMORY_BASELINE_NREGIONS);
606                 trace_vhost_user_set_mem_table_withfd(*fd_num, mr->name,
607                                                       reg->memory_size,
608                                                       reg->guest_phys_addr,
609                                                       reg->userspace_addr,
610                                                       offset);
611                 u->region_rb_offset[i] = offset;
612                 u->region_rb[i] = mr->ram_block;
613             } else if (*fd_num == VHOST_MEMORY_BASELINE_NREGIONS) {
614                 error_report("Failed preparing vhost-user memory table msg");
615                 return -ENOBUFS;
616             }
617             vhost_user_fill_msg_region(&region_buffer, reg, offset);
618             msg->payload.memory.regions[*fd_num] = region_buffer;
619             fds[(*fd_num)++] = fd;
620         } else if (track_ramblocks) {
621             u->region_rb_offset[i] = 0;
622             u->region_rb[i] = NULL;
623         }
624     }
625 
626     msg->payload.memory.nregions = *fd_num;
627 
628     if (!*fd_num) {
629         error_report("Failed initializing vhost-user memory map, "
630                      "consider using -object memory-backend-file share=on");
631         return -EINVAL;
632     }
633 
634     msg->hdr.size = sizeof(msg->payload.memory.nregions);
635     msg->hdr.size += sizeof(msg->payload.memory.padding);
636     msg->hdr.size += *fd_num * sizeof(VhostUserMemoryRegion);
637 
638     return 0;
639 }
640 
641 static inline bool reg_equal(struct vhost_memory_region *shadow_reg,
642                              struct vhost_memory_region *vdev_reg)
643 {
644     return shadow_reg->guest_phys_addr == vdev_reg->guest_phys_addr &&
645         shadow_reg->userspace_addr == vdev_reg->userspace_addr &&
646         shadow_reg->memory_size == vdev_reg->memory_size;
647 }
648 
649 static void scrub_shadow_regions(struct vhost_dev *dev,
650                                  struct scrub_regions *add_reg,
651                                  int *nr_add_reg,
652                                  struct scrub_regions *rem_reg,
653                                  int *nr_rem_reg, uint64_t *shadow_pcb,
654                                  bool track_ramblocks)
655 {
656     struct vhost_user *u = dev->opaque;
657     bool found[VHOST_USER_MAX_RAM_SLOTS] = {};
658     struct vhost_memory_region *reg, *shadow_reg;
659     int i, j, fd, add_idx = 0, rm_idx = 0, fd_num = 0;
660     ram_addr_t offset;
661     MemoryRegion *mr;
662     bool matching;
663 
664     /*
665      * Find memory regions present in our shadow state which are not in
666      * the device's current memory state.
667      *
668      * Mark regions in both the shadow and device state as "found".
669      */
670     for (i = 0; i < u->num_shadow_regions; i++) {
671         shadow_reg = &u->shadow_regions[i];
672         matching = false;
673 
674         for (j = 0; j < dev->mem->nregions; j++) {
675             reg = &dev->mem->regions[j];
676 
677             mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
678 
679             if (reg_equal(shadow_reg, reg)) {
680                 matching = true;
681                 found[j] = true;
682                 if (track_ramblocks) {
683                     /*
684                      * Reset postcopy client bases, region_rb, and
685                      * region_rb_offset in case regions are removed.
686                      */
687                     if (fd > 0) {
688                         u->region_rb_offset[j] = offset;
689                         u->region_rb[j] = mr->ram_block;
690                         shadow_pcb[j] = u->postcopy_client_bases[i];
691                     } else {
692                         u->region_rb_offset[j] = 0;
693                         u->region_rb[j] = NULL;
694                     }
695                 }
696                 break;
697             }
698         }
699 
700         /*
701          * If the region was not found in the current device memory state
702          * create an entry for it in the removed list.
703          */
704         if (!matching) {
705             rem_reg[rm_idx].region = shadow_reg;
706             rem_reg[rm_idx++].reg_idx = i;
707         }
708     }
709 
710     /*
711      * For regions not marked "found", create entries in the added list.
712      *
713      * Note their indexes in the device memory state and the indexes of their
714      * file descriptors.
715      */
716     for (i = 0; i < dev->mem->nregions; i++) {
717         reg = &dev->mem->regions[i];
718         vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
719         if (fd > 0) {
720             ++fd_num;
721         }
722 
723         /*
724          * If the region was in both the shadow and device state we don't
725          * need to send a VHOST_USER_ADD_MEM_REG message for it.
726          */
727         if (found[i]) {
728             continue;
729         }
730 
731         add_reg[add_idx].region = reg;
732         add_reg[add_idx].reg_idx = i;
733         add_reg[add_idx++].fd_idx = fd_num;
734     }
735     *nr_rem_reg = rm_idx;
736     *nr_add_reg = add_idx;
737 
738     return;
739 }
740 
741 static int send_remove_regions(struct vhost_dev *dev,
742                                struct scrub_regions *remove_reg,
743                                int nr_rem_reg, VhostUserMsg *msg,
744                                bool reply_supported)
745 {
746     struct vhost_user *u = dev->opaque;
747     struct vhost_memory_region *shadow_reg;
748     int i, fd, shadow_reg_idx, ret;
749     ram_addr_t offset;
750     VhostUserMemoryRegion region_buffer;
751 
752     /*
753      * The regions in remove_reg appear in the same order they do in the
754      * shadow table. Therefore we can minimize memory copies by iterating
755      * through remove_reg backwards.
756      */
757     for (i = nr_rem_reg - 1; i >= 0; i--) {
758         shadow_reg = remove_reg[i].region;
759         shadow_reg_idx = remove_reg[i].reg_idx;
760 
761         vhost_user_get_mr_data(shadow_reg->userspace_addr, &offset, &fd);
762 
763         if (fd > 0) {
764             msg->hdr.request = VHOST_USER_REM_MEM_REG;
765             vhost_user_fill_msg_region(&region_buffer, shadow_reg, 0);
766             msg->payload.mem_reg.region = region_buffer;
767 
768             ret = vhost_user_write(dev, msg, NULL, 0);
769             if (ret < 0) {
770                 return ret;
771             }
772 
773             if (reply_supported) {
774                 ret = process_message_reply(dev, msg);
775                 if (ret) {
776                     return ret;
777                 }
778             }
779         }
780 
781         /*
782          * At this point we know the backend has unmapped the region. It is now
783          * safe to remove it from the shadow table.
784          */
785         memmove(&u->shadow_regions[shadow_reg_idx],
786                 &u->shadow_regions[shadow_reg_idx + 1],
787                 sizeof(struct vhost_memory_region) *
788                 (u->num_shadow_regions - shadow_reg_idx - 1));
789         u->num_shadow_regions--;
790     }
791 
792     return 0;
793 }
794 
795 static int send_add_regions(struct vhost_dev *dev,
796                             struct scrub_regions *add_reg, int nr_add_reg,
797                             VhostUserMsg *msg, uint64_t *shadow_pcb,
798                             bool reply_supported, bool track_ramblocks)
799 {
800     struct vhost_user *u = dev->opaque;
801     int i, fd, ret, reg_idx, reg_fd_idx;
802     struct vhost_memory_region *reg;
803     MemoryRegion *mr;
804     ram_addr_t offset;
805     VhostUserMsg msg_reply;
806     VhostUserMemoryRegion region_buffer;
807 
808     for (i = 0; i < nr_add_reg; i++) {
809         reg = add_reg[i].region;
810         reg_idx = add_reg[i].reg_idx;
811         reg_fd_idx = add_reg[i].fd_idx;
812 
813         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
814 
815         if (fd > 0) {
816             if (track_ramblocks) {
817                 trace_vhost_user_set_mem_table_withfd(reg_fd_idx, mr->name,
818                                                       reg->memory_size,
819                                                       reg->guest_phys_addr,
820                                                       reg->userspace_addr,
821                                                       offset);
822                 u->region_rb_offset[reg_idx] = offset;
823                 u->region_rb[reg_idx] = mr->ram_block;
824             }
825             msg->hdr.request = VHOST_USER_ADD_MEM_REG;
826             vhost_user_fill_msg_region(&region_buffer, reg, offset);
827             msg->payload.mem_reg.region = region_buffer;
828 
829             ret = vhost_user_write(dev, msg, &fd, 1);
830             if (ret < 0) {
831                 return ret;
832             }
833 
834             if (track_ramblocks) {
835                 uint64_t reply_gpa;
836 
837                 ret = vhost_user_read(dev, &msg_reply);
838                 if (ret < 0) {
839                     return ret;
840                 }
841 
842                 reply_gpa = msg_reply.payload.mem_reg.region.guest_phys_addr;
843 
844                 if (msg_reply.hdr.request != VHOST_USER_ADD_MEM_REG) {
845                     error_report("%s: Received unexpected msg type."
846                                  "Expected %d received %d", __func__,
847                                  VHOST_USER_ADD_MEM_REG,
848                                  msg_reply.hdr.request);
849                     return -EPROTO;
850                 }
851 
852                 /*
853                  * We're using the same structure, just reusing one of the
854                  * fields, so it should be the same size.
855                  */
856                 if (msg_reply.hdr.size != msg->hdr.size) {
857                     error_report("%s: Unexpected size for postcopy reply "
858                                  "%d vs %d", __func__, msg_reply.hdr.size,
859                                  msg->hdr.size);
860                     return -EPROTO;
861                 }
862 
863                 /* Get the postcopy client base from the backend's reply. */
864                 if (reply_gpa == dev->mem->regions[reg_idx].guest_phys_addr) {
865                     shadow_pcb[reg_idx] =
866                         msg_reply.payload.mem_reg.region.userspace_addr;
867                     trace_vhost_user_set_mem_table_postcopy(
868                         msg_reply.payload.mem_reg.region.userspace_addr,
869                         msg->payload.mem_reg.region.userspace_addr,
870                         reg_fd_idx, reg_idx);
871                 } else {
872                     error_report("%s: invalid postcopy reply for region. "
873                                  "Got guest physical address %" PRIX64 ", expected "
874                                  "%" PRIX64, __func__, reply_gpa,
875                                  dev->mem->regions[reg_idx].guest_phys_addr);
876                     return -EPROTO;
877                 }
878             } else if (reply_supported) {
879                 ret = process_message_reply(dev, msg);
880                 if (ret) {
881                     return ret;
882                 }
883             }
884         } else if (track_ramblocks) {
885             u->region_rb_offset[reg_idx] = 0;
886             u->region_rb[reg_idx] = NULL;
887         }
888 
889         /*
890          * At this point, we know the backend has mapped in the new
891          * region, if the region has a valid file descriptor.
892          *
893          * The region should now be added to the shadow table.
894          */
895         u->shadow_regions[u->num_shadow_regions].guest_phys_addr =
896             reg->guest_phys_addr;
897         u->shadow_regions[u->num_shadow_regions].userspace_addr =
898             reg->userspace_addr;
899         u->shadow_regions[u->num_shadow_regions].memory_size =
900             reg->memory_size;
901         u->num_shadow_regions++;
902     }
903 
904     return 0;
905 }
906 
907 static int vhost_user_add_remove_regions(struct vhost_dev *dev,
908                                          VhostUserMsg *msg,
909                                          bool reply_supported,
910                                          bool track_ramblocks)
911 {
912     struct vhost_user *u = dev->opaque;
913     struct scrub_regions add_reg[VHOST_USER_MAX_RAM_SLOTS];
914     struct scrub_regions rem_reg[VHOST_USER_MAX_RAM_SLOTS];
915     uint64_t shadow_pcb[VHOST_USER_MAX_RAM_SLOTS] = {};
916     int nr_add_reg, nr_rem_reg;
917     int ret;
918 
919     msg->hdr.size = sizeof(msg->payload.mem_reg);
920 
921     /* Find the regions which need to be removed or added. */
922     scrub_shadow_regions(dev, add_reg, &nr_add_reg, rem_reg, &nr_rem_reg,
923                          shadow_pcb, track_ramblocks);
924 
925     if (nr_rem_reg) {
926         ret = send_remove_regions(dev, rem_reg, nr_rem_reg, msg,
927                                   reply_supported);
928         if (ret < 0) {
929             goto err;
930         }
931     }
932 
933     if (nr_add_reg) {
934         ret = send_add_regions(dev, add_reg, nr_add_reg, msg, shadow_pcb,
935                                reply_supported, track_ramblocks);
936         if (ret < 0) {
937             goto err;
938         }
939     }
940 
941     if (track_ramblocks) {
942         memcpy(u->postcopy_client_bases, shadow_pcb,
943                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
944         /*
945          * Now we've registered this with the postcopy code, we ack to the
946          * client, because now we're in the position to be able to deal with
947          * any faults it generates.
948          */
949         /* TODO: Use this for failure cases as well with a bad value. */
950         msg->hdr.size = sizeof(msg->payload.u64);
951         msg->payload.u64 = 0; /* OK */
952 
953         ret = vhost_user_write(dev, msg, NULL, 0);
954         if (ret < 0) {
955             return ret;
956         }
957     }
958 
959     return 0;
960 
961 err:
962     if (track_ramblocks) {
963         memcpy(u->postcopy_client_bases, shadow_pcb,
964                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
965     }
966 
967     return ret;
968 }
969 
970 static int vhost_user_set_mem_table_postcopy(struct vhost_dev *dev,
971                                              struct vhost_memory *mem,
972                                              bool reply_supported,
973                                              bool config_mem_slots)
974 {
975     struct vhost_user *u = dev->opaque;
976     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
977     size_t fd_num = 0;
978     VhostUserMsg msg_reply;
979     int region_i, msg_i;
980     int ret;
981 
982     VhostUserMsg msg = {
983         .hdr.flags = VHOST_USER_VERSION,
984     };
985 
986     if (u->region_rb_len < dev->mem->nregions) {
987         u->region_rb = g_renew(RAMBlock*, u->region_rb, dev->mem->nregions);
988         u->region_rb_offset = g_renew(ram_addr_t, u->region_rb_offset,
989                                       dev->mem->nregions);
990         memset(&(u->region_rb[u->region_rb_len]), '\0',
991                sizeof(RAMBlock *) * (dev->mem->nregions - u->region_rb_len));
992         memset(&(u->region_rb_offset[u->region_rb_len]), '\0',
993                sizeof(ram_addr_t) * (dev->mem->nregions - u->region_rb_len));
994         u->region_rb_len = dev->mem->nregions;
995     }
996 
997     if (config_mem_slots) {
998         ret = vhost_user_add_remove_regions(dev, &msg, reply_supported, true);
999         if (ret < 0) {
1000             return ret;
1001         }
1002     } else {
1003         ret = vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
1004                                                 true);
1005         if (ret < 0) {
1006             return ret;
1007         }
1008 
1009         ret = vhost_user_write(dev, &msg, fds, fd_num);
1010         if (ret < 0) {
1011             return ret;
1012         }
1013 
1014         ret = vhost_user_read(dev, &msg_reply);
1015         if (ret < 0) {
1016             return ret;
1017         }
1018 
1019         if (msg_reply.hdr.request != VHOST_USER_SET_MEM_TABLE) {
1020             error_report("%s: Received unexpected msg type."
1021                          "Expected %d received %d", __func__,
1022                          VHOST_USER_SET_MEM_TABLE, msg_reply.hdr.request);
1023             return -EPROTO;
1024         }
1025 
1026         /*
1027          * We're using the same structure, just reusing one of the
1028          * fields, so it should be the same size.
1029          */
1030         if (msg_reply.hdr.size != msg.hdr.size) {
1031             error_report("%s: Unexpected size for postcopy reply "
1032                          "%d vs %d", __func__, msg_reply.hdr.size,
1033                          msg.hdr.size);
1034             return -EPROTO;
1035         }
1036 
1037         memset(u->postcopy_client_bases, 0,
1038                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
1039 
1040         /*
1041          * They're in the same order as the regions that were sent
1042          * but some of the regions were skipped (above) if they
1043          * didn't have fd's
1044          */
1045         for (msg_i = 0, region_i = 0;
1046              region_i < dev->mem->nregions;
1047              region_i++) {
1048             if (msg_i < fd_num &&
1049                 msg_reply.payload.memory.regions[msg_i].guest_phys_addr ==
1050                 dev->mem->regions[region_i].guest_phys_addr) {
1051                 u->postcopy_client_bases[region_i] =
1052                     msg_reply.payload.memory.regions[msg_i].userspace_addr;
1053                 trace_vhost_user_set_mem_table_postcopy(
1054                     msg_reply.payload.memory.regions[msg_i].userspace_addr,
1055                     msg.payload.memory.regions[msg_i].userspace_addr,
1056                     msg_i, region_i);
1057                 msg_i++;
1058             }
1059         }
1060         if (msg_i != fd_num) {
1061             error_report("%s: postcopy reply not fully consumed "
1062                          "%d vs %zd",
1063                          __func__, msg_i, fd_num);
1064             return -EIO;
1065         }
1066 
1067         /*
1068          * Now we've registered this with the postcopy code, we ack to the
1069          * client, because now we're in the position to be able to deal
1070          * with any faults it generates.
1071          */
1072         /* TODO: Use this for failure cases as well with a bad value. */
1073         msg.hdr.size = sizeof(msg.payload.u64);
1074         msg.payload.u64 = 0; /* OK */
1075         ret = vhost_user_write(dev, &msg, NULL, 0);
1076         if (ret < 0) {
1077             return ret;
1078         }
1079     }
1080 
1081     return 0;
1082 }
1083 
1084 static int vhost_user_set_mem_table(struct vhost_dev *dev,
1085                                     struct vhost_memory *mem)
1086 {
1087     struct vhost_user *u = dev->opaque;
1088     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
1089     size_t fd_num = 0;
1090     bool do_postcopy = u->postcopy_listen && u->postcopy_fd.handler;
1091     bool reply_supported = virtio_has_feature(dev->protocol_features,
1092                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1093     bool config_mem_slots =
1094         virtio_has_feature(dev->protocol_features,
1095                            VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS);
1096     int ret;
1097 
1098     if (do_postcopy) {
1099         /*
1100          * Postcopy has enough differences that it's best done in it's own
1101          * version
1102          */
1103         return vhost_user_set_mem_table_postcopy(dev, mem, reply_supported,
1104                                                  config_mem_slots);
1105     }
1106 
1107     VhostUserMsg msg = {
1108         .hdr.flags = VHOST_USER_VERSION,
1109     };
1110 
1111     if (reply_supported) {
1112         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1113     }
1114 
1115     if (config_mem_slots) {
1116         ret = vhost_user_add_remove_regions(dev, &msg, reply_supported, false);
1117         if (ret < 0) {
1118             return ret;
1119         }
1120     } else {
1121         ret = vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
1122                                                 false);
1123         if (ret < 0) {
1124             return ret;
1125         }
1126 
1127         ret = vhost_user_write(dev, &msg, fds, fd_num);
1128         if (ret < 0) {
1129             return ret;
1130         }
1131 
1132         if (reply_supported) {
1133             return process_message_reply(dev, &msg);
1134         }
1135     }
1136 
1137     return 0;
1138 }
1139 
1140 static int vhost_user_set_vring_endian(struct vhost_dev *dev,
1141                                        struct vhost_vring_state *ring)
1142 {
1143     bool cross_endian = virtio_has_feature(dev->protocol_features,
1144                                            VHOST_USER_PROTOCOL_F_CROSS_ENDIAN);
1145     VhostUserMsg msg = {
1146         .hdr.request = VHOST_USER_SET_VRING_ENDIAN,
1147         .hdr.flags = VHOST_USER_VERSION,
1148         .payload.state = *ring,
1149         .hdr.size = sizeof(msg.payload.state),
1150     };
1151 
1152     if (!cross_endian) {
1153         error_report("vhost-user trying to send unhandled ioctl");
1154         return -ENOTSUP;
1155     }
1156 
1157     return vhost_user_write(dev, &msg, NULL, 0);
1158 }
1159 
1160 static int vhost_set_vring(struct vhost_dev *dev,
1161                            unsigned long int request,
1162                            struct vhost_vring_state *ring)
1163 {
1164     VhostUserMsg msg = {
1165         .hdr.request = request,
1166         .hdr.flags = VHOST_USER_VERSION,
1167         .payload.state = *ring,
1168         .hdr.size = sizeof(msg.payload.state),
1169     };
1170 
1171     return vhost_user_write(dev, &msg, NULL, 0);
1172 }
1173 
1174 static int vhost_user_set_vring_num(struct vhost_dev *dev,
1175                                     struct vhost_vring_state *ring)
1176 {
1177     return vhost_set_vring(dev, VHOST_USER_SET_VRING_NUM, ring);
1178 }
1179 
1180 static void vhost_user_host_notifier_free(VhostUserHostNotifier *n)
1181 {
1182     assert(n && n->unmap_addr);
1183     munmap(n->unmap_addr, qemu_real_host_page_size());
1184     n->unmap_addr = NULL;
1185 }
1186 
1187 /*
1188  * clean-up function for notifier, will finally free the structure
1189  * under rcu.
1190  */
1191 static void vhost_user_host_notifier_remove(VhostUserHostNotifier *n,
1192                                             VirtIODevice *vdev)
1193 {
1194     if (n->addr) {
1195         if (vdev) {
1196             virtio_queue_set_host_notifier_mr(vdev, n->idx, &n->mr, false);
1197         }
1198         assert(!n->unmap_addr);
1199         n->unmap_addr = n->addr;
1200         n->addr = NULL;
1201         call_rcu(n, vhost_user_host_notifier_free, rcu);
1202     }
1203 }
1204 
1205 static int vhost_user_set_vring_base(struct vhost_dev *dev,
1206                                      struct vhost_vring_state *ring)
1207 {
1208     return vhost_set_vring(dev, VHOST_USER_SET_VRING_BASE, ring);
1209 }
1210 
1211 static int vhost_user_set_vring_enable(struct vhost_dev *dev, int enable)
1212 {
1213     int i;
1214 
1215     if (!virtio_has_feature(dev->features, VHOST_USER_F_PROTOCOL_FEATURES)) {
1216         return -EINVAL;
1217     }
1218 
1219     for (i = 0; i < dev->nvqs; ++i) {
1220         int ret;
1221         struct vhost_vring_state state = {
1222             .index = dev->vq_index + i,
1223             .num   = enable,
1224         };
1225 
1226         ret = vhost_set_vring(dev, VHOST_USER_SET_VRING_ENABLE, &state);
1227         if (ret < 0) {
1228             /*
1229              * Restoring the previous state is likely infeasible, as well as
1230              * proceeding regardless the error, so just bail out and hope for
1231              * the device-level recovery.
1232              */
1233             return ret;
1234         }
1235     }
1236 
1237     return 0;
1238 }
1239 
1240 static VhostUserHostNotifier *fetch_notifier(VhostUserState *u,
1241                                              int idx)
1242 {
1243     if (idx >= u->notifiers->len) {
1244         return NULL;
1245     }
1246     return g_ptr_array_index(u->notifiers, idx);
1247 }
1248 
1249 static int vhost_user_get_vring_base(struct vhost_dev *dev,
1250                                      struct vhost_vring_state *ring)
1251 {
1252     int ret;
1253     VhostUserMsg msg = {
1254         .hdr.request = VHOST_USER_GET_VRING_BASE,
1255         .hdr.flags = VHOST_USER_VERSION,
1256         .payload.state = *ring,
1257         .hdr.size = sizeof(msg.payload.state),
1258     };
1259     struct vhost_user *u = dev->opaque;
1260 
1261     VhostUserHostNotifier *n = fetch_notifier(u->user, ring->index);
1262     if (n) {
1263         vhost_user_host_notifier_remove(n, dev->vdev);
1264     }
1265 
1266     ret = vhost_user_write(dev, &msg, NULL, 0);
1267     if (ret < 0) {
1268         return ret;
1269     }
1270 
1271     ret = vhost_user_read(dev, &msg);
1272     if (ret < 0) {
1273         return ret;
1274     }
1275 
1276     if (msg.hdr.request != VHOST_USER_GET_VRING_BASE) {
1277         error_report("Received unexpected msg type. Expected %d received %d",
1278                      VHOST_USER_GET_VRING_BASE, msg.hdr.request);
1279         return -EPROTO;
1280     }
1281 
1282     if (msg.hdr.size != sizeof(msg.payload.state)) {
1283         error_report("Received bad msg size.");
1284         return -EPROTO;
1285     }
1286 
1287     *ring = msg.payload.state;
1288 
1289     return 0;
1290 }
1291 
1292 static int vhost_set_vring_file(struct vhost_dev *dev,
1293                                 VhostUserRequest request,
1294                                 struct vhost_vring_file *file)
1295 {
1296     int fds[VHOST_USER_MAX_RAM_SLOTS];
1297     size_t fd_num = 0;
1298     VhostUserMsg msg = {
1299         .hdr.request = request,
1300         .hdr.flags = VHOST_USER_VERSION,
1301         .payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK,
1302         .hdr.size = sizeof(msg.payload.u64),
1303     };
1304 
1305     if (ioeventfd_enabled() && file->fd > 0) {
1306         fds[fd_num++] = file->fd;
1307     } else {
1308         msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
1309     }
1310 
1311     return vhost_user_write(dev, &msg, fds, fd_num);
1312 }
1313 
1314 static int vhost_user_set_vring_kick(struct vhost_dev *dev,
1315                                      struct vhost_vring_file *file)
1316 {
1317     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_KICK, file);
1318 }
1319 
1320 static int vhost_user_set_vring_call(struct vhost_dev *dev,
1321                                      struct vhost_vring_file *file)
1322 {
1323     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_CALL, file);
1324 }
1325 
1326 static int vhost_user_set_vring_err(struct vhost_dev *dev,
1327                                     struct vhost_vring_file *file)
1328 {
1329     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_ERR, file);
1330 }
1331 
1332 static int vhost_user_get_u64(struct vhost_dev *dev, int request, uint64_t *u64)
1333 {
1334     int ret;
1335     VhostUserMsg msg = {
1336         .hdr.request = request,
1337         .hdr.flags = VHOST_USER_VERSION,
1338     };
1339 
1340     if (vhost_user_one_time_request(request) && dev->vq_index != 0) {
1341         return 0;
1342     }
1343 
1344     ret = vhost_user_write(dev, &msg, NULL, 0);
1345     if (ret < 0) {
1346         return ret;
1347     }
1348 
1349     ret = vhost_user_read(dev, &msg);
1350     if (ret < 0) {
1351         return ret;
1352     }
1353 
1354     if (msg.hdr.request != request) {
1355         error_report("Received unexpected msg type. Expected %d received %d",
1356                      request, msg.hdr.request);
1357         return -EPROTO;
1358     }
1359 
1360     if (msg.hdr.size != sizeof(msg.payload.u64)) {
1361         error_report("Received bad msg size.");
1362         return -EPROTO;
1363     }
1364 
1365     *u64 = msg.payload.u64;
1366 
1367     return 0;
1368 }
1369 
1370 static int vhost_user_get_features(struct vhost_dev *dev, uint64_t *features)
1371 {
1372     if (vhost_user_get_u64(dev, VHOST_USER_GET_FEATURES, features) < 0) {
1373         return -EPROTO;
1374     }
1375 
1376     return 0;
1377 }
1378 
1379 static int enforce_reply(struct vhost_dev *dev,
1380                          const VhostUserMsg *msg)
1381 {
1382     uint64_t dummy;
1383 
1384     if (msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1385         return process_message_reply(dev, msg);
1386     }
1387 
1388    /*
1389     * We need to wait for a reply but the backend does not
1390     * support replies for the command we just sent.
1391     * Send VHOST_USER_GET_FEATURES which makes all backends
1392     * send a reply.
1393     */
1394     return vhost_user_get_features(dev, &dummy);
1395 }
1396 
1397 static int vhost_user_set_vring_addr(struct vhost_dev *dev,
1398                                      struct vhost_vring_addr *addr)
1399 {
1400     int ret;
1401     VhostUserMsg msg = {
1402         .hdr.request = VHOST_USER_SET_VRING_ADDR,
1403         .hdr.flags = VHOST_USER_VERSION,
1404         .payload.addr = *addr,
1405         .hdr.size = sizeof(msg.payload.addr),
1406     };
1407 
1408     bool reply_supported = virtio_has_feature(dev->protocol_features,
1409                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1410 
1411     /*
1412      * wait for a reply if logging is enabled to make sure
1413      * backend is actually logging changes
1414      */
1415     bool wait_for_reply = addr->flags & (1 << VHOST_VRING_F_LOG);
1416 
1417     if (reply_supported && wait_for_reply) {
1418         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1419     }
1420 
1421     ret = vhost_user_write(dev, &msg, NULL, 0);
1422     if (ret < 0) {
1423         return ret;
1424     }
1425 
1426     if (wait_for_reply) {
1427         return enforce_reply(dev, &msg);
1428     }
1429 
1430     return 0;
1431 }
1432 
1433 static int vhost_user_set_u64(struct vhost_dev *dev, int request, uint64_t u64,
1434                               bool wait_for_reply)
1435 {
1436     VhostUserMsg msg = {
1437         .hdr.request = request,
1438         .hdr.flags = VHOST_USER_VERSION,
1439         .payload.u64 = u64,
1440         .hdr.size = sizeof(msg.payload.u64),
1441     };
1442     int ret;
1443 
1444     if (wait_for_reply) {
1445         bool reply_supported = virtio_has_feature(dev->protocol_features,
1446                                           VHOST_USER_PROTOCOL_F_REPLY_ACK);
1447         if (reply_supported) {
1448             msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1449         }
1450     }
1451 
1452     ret = vhost_user_write(dev, &msg, NULL, 0);
1453     if (ret < 0) {
1454         return ret;
1455     }
1456 
1457     if (wait_for_reply) {
1458         return enforce_reply(dev, &msg);
1459     }
1460 
1461     return 0;
1462 }
1463 
1464 static int vhost_user_set_status(struct vhost_dev *dev, uint8_t status)
1465 {
1466     return vhost_user_set_u64(dev, VHOST_USER_SET_STATUS, status, false);
1467 }
1468 
1469 static int vhost_user_get_status(struct vhost_dev *dev, uint8_t *status)
1470 {
1471     uint64_t value;
1472     int ret;
1473 
1474     ret = vhost_user_get_u64(dev, VHOST_USER_GET_STATUS, &value);
1475     if (ret < 0) {
1476         return ret;
1477     }
1478     *status = value;
1479 
1480     return 0;
1481 }
1482 
1483 static int vhost_user_add_status(struct vhost_dev *dev, uint8_t status)
1484 {
1485     uint8_t s;
1486     int ret;
1487 
1488     ret = vhost_user_get_status(dev, &s);
1489     if (ret < 0) {
1490         return ret;
1491     }
1492 
1493     if ((s & status) == status) {
1494         return 0;
1495     }
1496     s |= status;
1497 
1498     return vhost_user_set_status(dev, s);
1499 }
1500 
1501 static int vhost_user_set_features(struct vhost_dev *dev,
1502                                    uint64_t features)
1503 {
1504     /*
1505      * wait for a reply if logging is enabled to make sure
1506      * backend is actually logging changes
1507      */
1508     bool log_enabled = features & (0x1ULL << VHOST_F_LOG_ALL);
1509     int ret;
1510 
1511     /*
1512      * We need to include any extra backend only feature bits that
1513      * might be needed by our device. Currently this includes the
1514      * VHOST_USER_F_PROTOCOL_FEATURES bit for enabling protocol
1515      * features.
1516      */
1517     ret = vhost_user_set_u64(dev, VHOST_USER_SET_FEATURES,
1518                               features | dev->backend_features,
1519                               log_enabled);
1520 
1521     if (virtio_has_feature(dev->protocol_features,
1522                            VHOST_USER_PROTOCOL_F_STATUS)) {
1523         if (!ret) {
1524             return vhost_user_add_status(dev, VIRTIO_CONFIG_S_FEATURES_OK);
1525         }
1526     }
1527 
1528     return ret;
1529 }
1530 
1531 static int vhost_user_set_protocol_features(struct vhost_dev *dev,
1532                                             uint64_t features)
1533 {
1534     return vhost_user_set_u64(dev, VHOST_USER_SET_PROTOCOL_FEATURES, features,
1535                               false);
1536 }
1537 
1538 static int vhost_user_set_owner(struct vhost_dev *dev)
1539 {
1540     VhostUserMsg msg = {
1541         .hdr.request = VHOST_USER_SET_OWNER,
1542         .hdr.flags = VHOST_USER_VERSION,
1543     };
1544 
1545     return vhost_user_write(dev, &msg, NULL, 0);
1546 }
1547 
1548 static int vhost_user_get_max_memslots(struct vhost_dev *dev,
1549                                        uint64_t *max_memslots)
1550 {
1551     uint64_t backend_max_memslots;
1552     int err;
1553 
1554     err = vhost_user_get_u64(dev, VHOST_USER_GET_MAX_MEM_SLOTS,
1555                              &backend_max_memslots);
1556     if (err < 0) {
1557         return err;
1558     }
1559 
1560     *max_memslots = backend_max_memslots;
1561 
1562     return 0;
1563 }
1564 
1565 static int vhost_user_reset_device(struct vhost_dev *dev)
1566 {
1567     VhostUserMsg msg = {
1568         .hdr.flags = VHOST_USER_VERSION,
1569     };
1570 
1571     msg.hdr.request = virtio_has_feature(dev->protocol_features,
1572                                          VHOST_USER_PROTOCOL_F_RESET_DEVICE)
1573         ? VHOST_USER_RESET_DEVICE
1574         : VHOST_USER_RESET_OWNER;
1575 
1576     return vhost_user_write(dev, &msg, NULL, 0);
1577 }
1578 
1579 static int vhost_user_slave_handle_config_change(struct vhost_dev *dev)
1580 {
1581     if (!dev->config_ops || !dev->config_ops->vhost_dev_config_notifier) {
1582         return -ENOSYS;
1583     }
1584 
1585     return dev->config_ops->vhost_dev_config_notifier(dev);
1586 }
1587 
1588 /*
1589  * Fetch or create the notifier for a given idx. Newly created
1590  * notifiers are added to the pointer array that tracks them.
1591  */
1592 static VhostUserHostNotifier *fetch_or_create_notifier(VhostUserState *u,
1593                                                        int idx)
1594 {
1595     VhostUserHostNotifier *n = NULL;
1596     if (idx >= u->notifiers->len) {
1597         g_ptr_array_set_size(u->notifiers, idx + 1);
1598     }
1599 
1600     n = g_ptr_array_index(u->notifiers, idx);
1601     if (!n) {
1602         /*
1603          * In case notification arrive out-of-order,
1604          * make room for current index.
1605          */
1606         g_ptr_array_remove_index(u->notifiers, idx);
1607         n = g_new0(VhostUserHostNotifier, 1);
1608         n->idx = idx;
1609         g_ptr_array_insert(u->notifiers, idx, n);
1610         trace_vhost_user_create_notifier(idx, n);
1611     }
1612 
1613     return n;
1614 }
1615 
1616 static int vhost_user_slave_handle_vring_host_notifier(struct vhost_dev *dev,
1617                                                        VhostUserVringArea *area,
1618                                                        int fd)
1619 {
1620     int queue_idx = area->u64 & VHOST_USER_VRING_IDX_MASK;
1621     size_t page_size = qemu_real_host_page_size();
1622     struct vhost_user *u = dev->opaque;
1623     VhostUserState *user = u->user;
1624     VirtIODevice *vdev = dev->vdev;
1625     VhostUserHostNotifier *n;
1626     void *addr;
1627     char *name;
1628 
1629     if (!virtio_has_feature(dev->protocol_features,
1630                             VHOST_USER_PROTOCOL_F_HOST_NOTIFIER) ||
1631         vdev == NULL || queue_idx >= virtio_get_num_queues(vdev)) {
1632         return -EINVAL;
1633     }
1634 
1635     /*
1636      * Fetch notifier and invalidate any old data before setting up
1637      * new mapped address.
1638      */
1639     n = fetch_or_create_notifier(user, queue_idx);
1640     vhost_user_host_notifier_remove(n, vdev);
1641 
1642     if (area->u64 & VHOST_USER_VRING_NOFD_MASK) {
1643         return 0;
1644     }
1645 
1646     /* Sanity check. */
1647     if (area->size != page_size) {
1648         return -EINVAL;
1649     }
1650 
1651     addr = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1652                 fd, area->offset);
1653     if (addr == MAP_FAILED) {
1654         return -EFAULT;
1655     }
1656 
1657     name = g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
1658                            user, queue_idx);
1659     if (!n->mr.ram) { /* Don't init again after suspend. */
1660         memory_region_init_ram_device_ptr(&n->mr, OBJECT(vdev), name,
1661                                           page_size, addr);
1662     } else {
1663         n->mr.ram_block->host = addr;
1664     }
1665     g_free(name);
1666 
1667     if (virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, true)) {
1668         object_unparent(OBJECT(&n->mr));
1669         munmap(addr, page_size);
1670         return -ENXIO;
1671     }
1672 
1673     n->addr = addr;
1674 
1675     return 0;
1676 }
1677 
1678 static void close_slave_channel(struct vhost_user *u)
1679 {
1680     g_source_destroy(u->slave_src);
1681     g_source_unref(u->slave_src);
1682     u->slave_src = NULL;
1683     object_unref(OBJECT(u->slave_ioc));
1684     u->slave_ioc = NULL;
1685 }
1686 
1687 static gboolean slave_read(QIOChannel *ioc, GIOCondition condition,
1688                            gpointer opaque)
1689 {
1690     struct vhost_dev *dev = opaque;
1691     struct vhost_user *u = dev->opaque;
1692     VhostUserHeader hdr = { 0, };
1693     VhostUserPayload payload = { 0, };
1694     Error *local_err = NULL;
1695     gboolean rc = G_SOURCE_CONTINUE;
1696     int ret = 0;
1697     struct iovec iov;
1698     g_autofree int *fd = NULL;
1699     size_t fdsize = 0;
1700     int i;
1701 
1702     /* Read header */
1703     iov.iov_base = &hdr;
1704     iov.iov_len = VHOST_USER_HDR_SIZE;
1705 
1706     if (qio_channel_readv_full_all(ioc, &iov, 1, &fd, &fdsize, &local_err)) {
1707         error_report_err(local_err);
1708         goto err;
1709     }
1710 
1711     if (hdr.size > VHOST_USER_PAYLOAD_SIZE) {
1712         error_report("Failed to read msg header."
1713                 " Size %d exceeds the maximum %zu.", hdr.size,
1714                 VHOST_USER_PAYLOAD_SIZE);
1715         goto err;
1716     }
1717 
1718     /* Read payload */
1719     if (qio_channel_read_all(ioc, (char *) &payload, hdr.size, &local_err)) {
1720         error_report_err(local_err);
1721         goto err;
1722     }
1723 
1724     switch (hdr.request) {
1725     case VHOST_USER_SLAVE_IOTLB_MSG:
1726         ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
1727         break;
1728     case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
1729         ret = vhost_user_slave_handle_config_change(dev);
1730         break;
1731     case VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG:
1732         ret = vhost_user_slave_handle_vring_host_notifier(dev, &payload.area,
1733                                                           fd ? fd[0] : -1);
1734         break;
1735     default:
1736         error_report("Received unexpected msg type: %d.", hdr.request);
1737         ret = -EINVAL;
1738     }
1739 
1740     /*
1741      * REPLY_ACK feature handling. Other reply types has to be managed
1742      * directly in their request handlers.
1743      */
1744     if (hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1745         struct iovec iovec[2];
1746 
1747 
1748         hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
1749         hdr.flags |= VHOST_USER_REPLY_MASK;
1750 
1751         payload.u64 = !!ret;
1752         hdr.size = sizeof(payload.u64);
1753 
1754         iovec[0].iov_base = &hdr;
1755         iovec[0].iov_len = VHOST_USER_HDR_SIZE;
1756         iovec[1].iov_base = &payload;
1757         iovec[1].iov_len = hdr.size;
1758 
1759         if (qio_channel_writev_all(ioc, iovec, ARRAY_SIZE(iovec), &local_err)) {
1760             error_report_err(local_err);
1761             goto err;
1762         }
1763     }
1764 
1765     goto fdcleanup;
1766 
1767 err:
1768     close_slave_channel(u);
1769     rc = G_SOURCE_REMOVE;
1770 
1771 fdcleanup:
1772     if (fd) {
1773         for (i = 0; i < fdsize; i++) {
1774             close(fd[i]);
1775         }
1776     }
1777     return rc;
1778 }
1779 
1780 static int vhost_setup_slave_channel(struct vhost_dev *dev)
1781 {
1782     VhostUserMsg msg = {
1783         .hdr.request = VHOST_USER_SET_SLAVE_REQ_FD,
1784         .hdr.flags = VHOST_USER_VERSION,
1785     };
1786     struct vhost_user *u = dev->opaque;
1787     int sv[2], ret = 0;
1788     bool reply_supported = virtio_has_feature(dev->protocol_features,
1789                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1790     Error *local_err = NULL;
1791     QIOChannel *ioc;
1792 
1793     if (!virtio_has_feature(dev->protocol_features,
1794                             VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
1795         return 0;
1796     }
1797 
1798     if (qemu_socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
1799         int saved_errno = errno;
1800         error_report("socketpair() failed");
1801         return -saved_errno;
1802     }
1803 
1804     ioc = QIO_CHANNEL(qio_channel_socket_new_fd(sv[0], &local_err));
1805     if (!ioc) {
1806         error_report_err(local_err);
1807         return -ECONNREFUSED;
1808     }
1809     u->slave_ioc = ioc;
1810     slave_update_read_handler(dev, NULL);
1811 
1812     if (reply_supported) {
1813         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1814     }
1815 
1816     ret = vhost_user_write(dev, &msg, &sv[1], 1);
1817     if (ret) {
1818         goto out;
1819     }
1820 
1821     if (reply_supported) {
1822         ret = process_message_reply(dev, &msg);
1823     }
1824 
1825 out:
1826     close(sv[1]);
1827     if (ret) {
1828         close_slave_channel(u);
1829     }
1830 
1831     return ret;
1832 }
1833 
1834 #ifdef CONFIG_LINUX
1835 /*
1836  * Called back from the postcopy fault thread when a fault is received on our
1837  * ufd.
1838  * TODO: This is Linux specific
1839  */
1840 static int vhost_user_postcopy_fault_handler(struct PostCopyFD *pcfd,
1841                                              void *ufd)
1842 {
1843     struct vhost_dev *dev = pcfd->data;
1844     struct vhost_user *u = dev->opaque;
1845     struct uffd_msg *msg = ufd;
1846     uint64_t faultaddr = msg->arg.pagefault.address;
1847     RAMBlock *rb = NULL;
1848     uint64_t rb_offset;
1849     int i;
1850 
1851     trace_vhost_user_postcopy_fault_handler(pcfd->idstr, faultaddr,
1852                                             dev->mem->nregions);
1853     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1854         trace_vhost_user_postcopy_fault_handler_loop(i,
1855                 u->postcopy_client_bases[i], dev->mem->regions[i].memory_size);
1856         if (faultaddr >= u->postcopy_client_bases[i]) {
1857             /* Ofset of the fault address in the vhost region */
1858             uint64_t region_offset = faultaddr - u->postcopy_client_bases[i];
1859             if (region_offset < dev->mem->regions[i].memory_size) {
1860                 rb_offset = region_offset + u->region_rb_offset[i];
1861                 trace_vhost_user_postcopy_fault_handler_found(i,
1862                         region_offset, rb_offset);
1863                 rb = u->region_rb[i];
1864                 return postcopy_request_shared_page(pcfd, rb, faultaddr,
1865                                                     rb_offset);
1866             }
1867         }
1868     }
1869     error_report("%s: Failed to find region for fault %" PRIx64,
1870                  __func__, faultaddr);
1871     return -1;
1872 }
1873 
1874 static int vhost_user_postcopy_waker(struct PostCopyFD *pcfd, RAMBlock *rb,
1875                                      uint64_t offset)
1876 {
1877     struct vhost_dev *dev = pcfd->data;
1878     struct vhost_user *u = dev->opaque;
1879     int i;
1880 
1881     trace_vhost_user_postcopy_waker(qemu_ram_get_idstr(rb), offset);
1882 
1883     if (!u) {
1884         return 0;
1885     }
1886     /* Translate the offset into an address in the clients address space */
1887     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1888         if (u->region_rb[i] == rb &&
1889             offset >= u->region_rb_offset[i] &&
1890             offset < (u->region_rb_offset[i] +
1891                       dev->mem->regions[i].memory_size)) {
1892             uint64_t client_addr = (offset - u->region_rb_offset[i]) +
1893                                    u->postcopy_client_bases[i];
1894             trace_vhost_user_postcopy_waker_found(client_addr);
1895             return postcopy_wake_shared(pcfd, client_addr, rb);
1896         }
1897     }
1898 
1899     trace_vhost_user_postcopy_waker_nomatch(qemu_ram_get_idstr(rb), offset);
1900     return 0;
1901 }
1902 #endif
1903 
1904 /*
1905  * Called at the start of an inbound postcopy on reception of the
1906  * 'advise' command.
1907  */
1908 static int vhost_user_postcopy_advise(struct vhost_dev *dev, Error **errp)
1909 {
1910 #ifdef CONFIG_LINUX
1911     struct vhost_user *u = dev->opaque;
1912     CharBackend *chr = u->user->chr;
1913     int ufd;
1914     int ret;
1915     VhostUserMsg msg = {
1916         .hdr.request = VHOST_USER_POSTCOPY_ADVISE,
1917         .hdr.flags = VHOST_USER_VERSION,
1918     };
1919 
1920     ret = vhost_user_write(dev, &msg, NULL, 0);
1921     if (ret < 0) {
1922         error_setg(errp, "Failed to send postcopy_advise to vhost");
1923         return ret;
1924     }
1925 
1926     ret = vhost_user_read(dev, &msg);
1927     if (ret < 0) {
1928         error_setg(errp, "Failed to get postcopy_advise reply from vhost");
1929         return ret;
1930     }
1931 
1932     if (msg.hdr.request != VHOST_USER_POSTCOPY_ADVISE) {
1933         error_setg(errp, "Unexpected msg type. Expected %d received %d",
1934                      VHOST_USER_POSTCOPY_ADVISE, msg.hdr.request);
1935         return -EPROTO;
1936     }
1937 
1938     if (msg.hdr.size) {
1939         error_setg(errp, "Received bad msg size.");
1940         return -EPROTO;
1941     }
1942     ufd = qemu_chr_fe_get_msgfd(chr);
1943     if (ufd < 0) {
1944         error_setg(errp, "%s: Failed to get ufd", __func__);
1945         return -EIO;
1946     }
1947     qemu_socket_set_nonblock(ufd);
1948 
1949     /* register ufd with userfault thread */
1950     u->postcopy_fd.fd = ufd;
1951     u->postcopy_fd.data = dev;
1952     u->postcopy_fd.handler = vhost_user_postcopy_fault_handler;
1953     u->postcopy_fd.waker = vhost_user_postcopy_waker;
1954     u->postcopy_fd.idstr = "vhost-user"; /* Need to find unique name */
1955     postcopy_register_shared_ufd(&u->postcopy_fd);
1956     return 0;
1957 #else
1958     error_setg(errp, "Postcopy not supported on non-Linux systems");
1959     return -ENOSYS;
1960 #endif
1961 }
1962 
1963 /*
1964  * Called at the switch to postcopy on reception of the 'listen' command.
1965  */
1966 static int vhost_user_postcopy_listen(struct vhost_dev *dev, Error **errp)
1967 {
1968     struct vhost_user *u = dev->opaque;
1969     int ret;
1970     VhostUserMsg msg = {
1971         .hdr.request = VHOST_USER_POSTCOPY_LISTEN,
1972         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
1973     };
1974     u->postcopy_listen = true;
1975 
1976     trace_vhost_user_postcopy_listen();
1977 
1978     ret = vhost_user_write(dev, &msg, NULL, 0);
1979     if (ret < 0) {
1980         error_setg(errp, "Failed to send postcopy_listen to vhost");
1981         return ret;
1982     }
1983 
1984     ret = process_message_reply(dev, &msg);
1985     if (ret) {
1986         error_setg(errp, "Failed to receive reply to postcopy_listen");
1987         return ret;
1988     }
1989 
1990     return 0;
1991 }
1992 
1993 /*
1994  * Called at the end of postcopy
1995  */
1996 static int vhost_user_postcopy_end(struct vhost_dev *dev, Error **errp)
1997 {
1998     VhostUserMsg msg = {
1999         .hdr.request = VHOST_USER_POSTCOPY_END,
2000         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
2001     };
2002     int ret;
2003     struct vhost_user *u = dev->opaque;
2004 
2005     trace_vhost_user_postcopy_end_entry();
2006 
2007     ret = vhost_user_write(dev, &msg, NULL, 0);
2008     if (ret < 0) {
2009         error_setg(errp, "Failed to send postcopy_end to vhost");
2010         return ret;
2011     }
2012 
2013     ret = process_message_reply(dev, &msg);
2014     if (ret) {
2015         error_setg(errp, "Failed to receive reply to postcopy_end");
2016         return ret;
2017     }
2018     postcopy_unregister_shared_ufd(&u->postcopy_fd);
2019     close(u->postcopy_fd.fd);
2020     u->postcopy_fd.handler = NULL;
2021 
2022     trace_vhost_user_postcopy_end_exit();
2023 
2024     return 0;
2025 }
2026 
2027 static int vhost_user_postcopy_notifier(NotifierWithReturn *notifier,
2028                                         void *opaque)
2029 {
2030     struct PostcopyNotifyData *pnd = opaque;
2031     struct vhost_user *u = container_of(notifier, struct vhost_user,
2032                                          postcopy_notifier);
2033     struct vhost_dev *dev = u->dev;
2034 
2035     switch (pnd->reason) {
2036     case POSTCOPY_NOTIFY_PROBE:
2037         if (!virtio_has_feature(dev->protocol_features,
2038                                 VHOST_USER_PROTOCOL_F_PAGEFAULT)) {
2039             /* TODO: Get the device name into this error somehow */
2040             error_setg(pnd->errp,
2041                        "vhost-user backend not capable of postcopy");
2042             return -ENOENT;
2043         }
2044         break;
2045 
2046     case POSTCOPY_NOTIFY_INBOUND_ADVISE:
2047         return vhost_user_postcopy_advise(dev, pnd->errp);
2048 
2049     case POSTCOPY_NOTIFY_INBOUND_LISTEN:
2050         return vhost_user_postcopy_listen(dev, pnd->errp);
2051 
2052     case POSTCOPY_NOTIFY_INBOUND_END:
2053         return vhost_user_postcopy_end(dev, pnd->errp);
2054 
2055     default:
2056         /* We ignore notifications we don't know */
2057         break;
2058     }
2059 
2060     return 0;
2061 }
2062 
2063 static int vhost_user_backend_init(struct vhost_dev *dev, void *opaque,
2064                                    Error **errp)
2065 {
2066     uint64_t features, ram_slots;
2067     struct vhost_user *u;
2068     VhostUserState *vus = (VhostUserState *) opaque;
2069     int err;
2070 
2071     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2072 
2073     u = g_new0(struct vhost_user, 1);
2074     u->user = vus;
2075     u->dev = dev;
2076     dev->opaque = u;
2077 
2078     err = vhost_user_get_features(dev, &features);
2079     if (err < 0) {
2080         error_setg_errno(errp, -err, "vhost_backend_init failed");
2081         return err;
2082     }
2083 
2084     if (virtio_has_feature(features, VHOST_USER_F_PROTOCOL_FEATURES)) {
2085         bool supports_f_config = vus->supports_config ||
2086             (dev->config_ops && dev->config_ops->vhost_dev_config_notifier);
2087         uint64_t protocol_features;
2088 
2089         dev->backend_features |= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
2090 
2091         err = vhost_user_get_u64(dev, VHOST_USER_GET_PROTOCOL_FEATURES,
2092                                  &protocol_features);
2093         if (err < 0) {
2094             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2095             return -EPROTO;
2096         }
2097 
2098         /*
2099          * We will use all the protocol features we support - although
2100          * we suppress F_CONFIG if we know QEMUs internal code can not support
2101          * it.
2102          */
2103         protocol_features &= VHOST_USER_PROTOCOL_FEATURE_MASK;
2104 
2105         if (supports_f_config) {
2106             if (!virtio_has_feature(protocol_features,
2107                                     VHOST_USER_PROTOCOL_F_CONFIG)) {
2108                 error_setg(errp, "vhost-user device expecting "
2109                            "VHOST_USER_PROTOCOL_F_CONFIG but the vhost-user backend does "
2110                            "not support it.");
2111                 return -EPROTO;
2112             }
2113         } else {
2114             if (virtio_has_feature(protocol_features,
2115                                    VHOST_USER_PROTOCOL_F_CONFIG)) {
2116                 warn_reportf_err(*errp, "vhost-user backend supports "
2117                                  "VHOST_USER_PROTOCOL_F_CONFIG but QEMU does not.");
2118                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG);
2119             }
2120         }
2121 
2122         /* final set of protocol features */
2123         dev->protocol_features = protocol_features;
2124         err = vhost_user_set_protocol_features(dev, dev->protocol_features);
2125         if (err < 0) {
2126             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2127             return -EPROTO;
2128         }
2129 
2130         /* query the max queues we support if backend supports Multiple Queue */
2131         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)) {
2132             err = vhost_user_get_u64(dev, VHOST_USER_GET_QUEUE_NUM,
2133                                      &dev->max_queues);
2134             if (err < 0) {
2135                 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2136                 return -EPROTO;
2137             }
2138         } else {
2139             dev->max_queues = 1;
2140         }
2141 
2142         if (dev->num_queues && dev->max_queues < dev->num_queues) {
2143             error_setg(errp, "The maximum number of queues supported by the "
2144                        "backend is %" PRIu64, dev->max_queues);
2145             return -EINVAL;
2146         }
2147 
2148         if (virtio_has_feature(features, VIRTIO_F_IOMMU_PLATFORM) &&
2149                 !(virtio_has_feature(dev->protocol_features,
2150                     VHOST_USER_PROTOCOL_F_SLAVE_REQ) &&
2151                  virtio_has_feature(dev->protocol_features,
2152                     VHOST_USER_PROTOCOL_F_REPLY_ACK))) {
2153             error_setg(errp, "IOMMU support requires reply-ack and "
2154                        "slave-req protocol features.");
2155             return -EINVAL;
2156         }
2157 
2158         /* get max memory regions if backend supports configurable RAM slots */
2159         if (!virtio_has_feature(dev->protocol_features,
2160                                 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS)) {
2161             u->user->memory_slots = VHOST_MEMORY_BASELINE_NREGIONS;
2162         } else {
2163             err = vhost_user_get_max_memslots(dev, &ram_slots);
2164             if (err < 0) {
2165                 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2166                 return -EPROTO;
2167             }
2168 
2169             if (ram_slots < u->user->memory_slots) {
2170                 error_setg(errp, "The backend specified a max ram slots limit "
2171                            "of %" PRIu64", when the prior validated limit was "
2172                            "%d. This limit should never decrease.", ram_slots,
2173                            u->user->memory_slots);
2174                 return -EINVAL;
2175             }
2176 
2177             u->user->memory_slots = MIN(ram_slots, VHOST_USER_MAX_RAM_SLOTS);
2178         }
2179     }
2180 
2181     if (dev->migration_blocker == NULL &&
2182         !virtio_has_feature(dev->protocol_features,
2183                             VHOST_USER_PROTOCOL_F_LOG_SHMFD)) {
2184         error_setg(&dev->migration_blocker,
2185                    "Migration disabled: vhost-user backend lacks "
2186                    "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
2187     }
2188 
2189     if (dev->vq_index == 0) {
2190         err = vhost_setup_slave_channel(dev);
2191         if (err < 0) {
2192             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2193             return -EPROTO;
2194         }
2195     }
2196 
2197     u->postcopy_notifier.notify = vhost_user_postcopy_notifier;
2198     postcopy_add_notifier(&u->postcopy_notifier);
2199 
2200     return 0;
2201 }
2202 
2203 static int vhost_user_backend_cleanup(struct vhost_dev *dev)
2204 {
2205     struct vhost_user *u;
2206 
2207     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2208 
2209     u = dev->opaque;
2210     if (u->postcopy_notifier.notify) {
2211         postcopy_remove_notifier(&u->postcopy_notifier);
2212         u->postcopy_notifier.notify = NULL;
2213     }
2214     u->postcopy_listen = false;
2215     if (u->postcopy_fd.handler) {
2216         postcopy_unregister_shared_ufd(&u->postcopy_fd);
2217         close(u->postcopy_fd.fd);
2218         u->postcopy_fd.handler = NULL;
2219     }
2220     if (u->slave_ioc) {
2221         close_slave_channel(u);
2222     }
2223     g_free(u->region_rb);
2224     u->region_rb = NULL;
2225     g_free(u->region_rb_offset);
2226     u->region_rb_offset = NULL;
2227     u->region_rb_len = 0;
2228     g_free(u);
2229     dev->opaque = 0;
2230 
2231     return 0;
2232 }
2233 
2234 static int vhost_user_get_vq_index(struct vhost_dev *dev, int idx)
2235 {
2236     assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
2237 
2238     return idx;
2239 }
2240 
2241 static int vhost_user_memslots_limit(struct vhost_dev *dev)
2242 {
2243     struct vhost_user *u = dev->opaque;
2244 
2245     return u->user->memory_slots;
2246 }
2247 
2248 static bool vhost_user_requires_shm_log(struct vhost_dev *dev)
2249 {
2250     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2251 
2252     return virtio_has_feature(dev->protocol_features,
2253                               VHOST_USER_PROTOCOL_F_LOG_SHMFD);
2254 }
2255 
2256 static int vhost_user_migration_done(struct vhost_dev *dev, char* mac_addr)
2257 {
2258     VhostUserMsg msg = { };
2259 
2260     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2261 
2262     /* If guest supports GUEST_ANNOUNCE do nothing */
2263     if (virtio_has_feature(dev->acked_features, VIRTIO_NET_F_GUEST_ANNOUNCE)) {
2264         return 0;
2265     }
2266 
2267     /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
2268     if (virtio_has_feature(dev->protocol_features,
2269                            VHOST_USER_PROTOCOL_F_RARP)) {
2270         msg.hdr.request = VHOST_USER_SEND_RARP;
2271         msg.hdr.flags = VHOST_USER_VERSION;
2272         memcpy((char *)&msg.payload.u64, mac_addr, 6);
2273         msg.hdr.size = sizeof(msg.payload.u64);
2274 
2275         return vhost_user_write(dev, &msg, NULL, 0);
2276     }
2277     return -ENOTSUP;
2278 }
2279 
2280 static bool vhost_user_can_merge(struct vhost_dev *dev,
2281                                  uint64_t start1, uint64_t size1,
2282                                  uint64_t start2, uint64_t size2)
2283 {
2284     ram_addr_t offset;
2285     int mfd, rfd;
2286 
2287     (void)vhost_user_get_mr_data(start1, &offset, &mfd);
2288     (void)vhost_user_get_mr_data(start2, &offset, &rfd);
2289 
2290     return mfd == rfd;
2291 }
2292 
2293 static int vhost_user_net_set_mtu(struct vhost_dev *dev, uint16_t mtu)
2294 {
2295     VhostUserMsg msg;
2296     bool reply_supported = virtio_has_feature(dev->protocol_features,
2297                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2298     int ret;
2299 
2300     if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU))) {
2301         return 0;
2302     }
2303 
2304     msg.hdr.request = VHOST_USER_NET_SET_MTU;
2305     msg.payload.u64 = mtu;
2306     msg.hdr.size = sizeof(msg.payload.u64);
2307     msg.hdr.flags = VHOST_USER_VERSION;
2308     if (reply_supported) {
2309         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2310     }
2311 
2312     ret = vhost_user_write(dev, &msg, NULL, 0);
2313     if (ret < 0) {
2314         return ret;
2315     }
2316 
2317     /* If reply_ack supported, slave has to ack specified MTU is valid */
2318     if (reply_supported) {
2319         return process_message_reply(dev, &msg);
2320     }
2321 
2322     return 0;
2323 }
2324 
2325 static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
2326                                             struct vhost_iotlb_msg *imsg)
2327 {
2328     int ret;
2329     VhostUserMsg msg = {
2330         .hdr.request = VHOST_USER_IOTLB_MSG,
2331         .hdr.size = sizeof(msg.payload.iotlb),
2332         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
2333         .payload.iotlb = *imsg,
2334     };
2335 
2336     ret = vhost_user_write(dev, &msg, NULL, 0);
2337     if (ret < 0) {
2338         return ret;
2339     }
2340 
2341     return process_message_reply(dev, &msg);
2342 }
2343 
2344 
2345 static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
2346 {
2347     /* No-op as the receive channel is not dedicated to IOTLB messages. */
2348 }
2349 
2350 static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
2351                                  uint32_t config_len, Error **errp)
2352 {
2353     int ret;
2354     VhostUserMsg msg = {
2355         .hdr.request = VHOST_USER_GET_CONFIG,
2356         .hdr.flags = VHOST_USER_VERSION,
2357         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + config_len,
2358     };
2359 
2360     if (!virtio_has_feature(dev->protocol_features,
2361                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2362         error_setg(errp, "VHOST_USER_PROTOCOL_F_CONFIG not supported");
2363         return -EINVAL;
2364     }
2365 
2366     assert(config_len <= VHOST_USER_MAX_CONFIG_SIZE);
2367 
2368     msg.payload.config.offset = 0;
2369     msg.payload.config.size = config_len;
2370     ret = vhost_user_write(dev, &msg, NULL, 0);
2371     if (ret < 0) {
2372         error_setg_errno(errp, -ret, "vhost_get_config failed");
2373         return ret;
2374     }
2375 
2376     ret = vhost_user_read(dev, &msg);
2377     if (ret < 0) {
2378         error_setg_errno(errp, -ret, "vhost_get_config failed");
2379         return ret;
2380     }
2381 
2382     if (msg.hdr.request != VHOST_USER_GET_CONFIG) {
2383         error_setg(errp,
2384                    "Received unexpected msg type. Expected %d received %d",
2385                    VHOST_USER_GET_CONFIG, msg.hdr.request);
2386         return -EPROTO;
2387     }
2388 
2389     if (msg.hdr.size != VHOST_USER_CONFIG_HDR_SIZE + config_len) {
2390         error_setg(errp, "Received bad msg size.");
2391         return -EPROTO;
2392     }
2393 
2394     memcpy(config, msg.payload.config.region, config_len);
2395 
2396     return 0;
2397 }
2398 
2399 static int vhost_user_set_config(struct vhost_dev *dev, const uint8_t *data,
2400                                  uint32_t offset, uint32_t size, uint32_t flags)
2401 {
2402     int ret;
2403     uint8_t *p;
2404     bool reply_supported = virtio_has_feature(dev->protocol_features,
2405                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2406 
2407     VhostUserMsg msg = {
2408         .hdr.request = VHOST_USER_SET_CONFIG,
2409         .hdr.flags = VHOST_USER_VERSION,
2410         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + size,
2411     };
2412 
2413     if (!virtio_has_feature(dev->protocol_features,
2414                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2415         return -ENOTSUP;
2416     }
2417 
2418     if (reply_supported) {
2419         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2420     }
2421 
2422     if (size > VHOST_USER_MAX_CONFIG_SIZE) {
2423         return -EINVAL;
2424     }
2425 
2426     msg.payload.config.offset = offset,
2427     msg.payload.config.size = size,
2428     msg.payload.config.flags = flags,
2429     p = msg.payload.config.region;
2430     memcpy(p, data, size);
2431 
2432     ret = vhost_user_write(dev, &msg, NULL, 0);
2433     if (ret < 0) {
2434         return ret;
2435     }
2436 
2437     if (reply_supported) {
2438         return process_message_reply(dev, &msg);
2439     }
2440 
2441     return 0;
2442 }
2443 
2444 static int vhost_user_crypto_create_session(struct vhost_dev *dev,
2445                                             void *session_info,
2446                                             uint64_t *session_id)
2447 {
2448     int ret;
2449     bool crypto_session = virtio_has_feature(dev->protocol_features,
2450                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2451     CryptoDevBackendSymSessionInfo *sess_info = session_info;
2452     VhostUserMsg msg = {
2453         .hdr.request = VHOST_USER_CREATE_CRYPTO_SESSION,
2454         .hdr.flags = VHOST_USER_VERSION,
2455         .hdr.size = sizeof(msg.payload.session),
2456     };
2457 
2458     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2459 
2460     if (!crypto_session) {
2461         error_report("vhost-user trying to send unhandled ioctl");
2462         return -ENOTSUP;
2463     }
2464 
2465     memcpy(&msg.payload.session.session_setup_data, sess_info,
2466               sizeof(CryptoDevBackendSymSessionInfo));
2467     if (sess_info->key_len) {
2468         memcpy(&msg.payload.session.key, sess_info->cipher_key,
2469                sess_info->key_len);
2470     }
2471     if (sess_info->auth_key_len > 0) {
2472         memcpy(&msg.payload.session.auth_key, sess_info->auth_key,
2473                sess_info->auth_key_len);
2474     }
2475     ret = vhost_user_write(dev, &msg, NULL, 0);
2476     if (ret < 0) {
2477         error_report("vhost_user_write() return %d, create session failed",
2478                      ret);
2479         return ret;
2480     }
2481 
2482     ret = vhost_user_read(dev, &msg);
2483     if (ret < 0) {
2484         error_report("vhost_user_read() return %d, create session failed",
2485                      ret);
2486         return ret;
2487     }
2488 
2489     if (msg.hdr.request != VHOST_USER_CREATE_CRYPTO_SESSION) {
2490         error_report("Received unexpected msg type. Expected %d received %d",
2491                      VHOST_USER_CREATE_CRYPTO_SESSION, msg.hdr.request);
2492         return -EPROTO;
2493     }
2494 
2495     if (msg.hdr.size != sizeof(msg.payload.session)) {
2496         error_report("Received bad msg size.");
2497         return -EPROTO;
2498     }
2499 
2500     if (msg.payload.session.session_id < 0) {
2501         error_report("Bad session id: %" PRId64 "",
2502                               msg.payload.session.session_id);
2503         return -EINVAL;
2504     }
2505     *session_id = msg.payload.session.session_id;
2506 
2507     return 0;
2508 }
2509 
2510 static int
2511 vhost_user_crypto_close_session(struct vhost_dev *dev, uint64_t session_id)
2512 {
2513     int ret;
2514     bool crypto_session = virtio_has_feature(dev->protocol_features,
2515                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2516     VhostUserMsg msg = {
2517         .hdr.request = VHOST_USER_CLOSE_CRYPTO_SESSION,
2518         .hdr.flags = VHOST_USER_VERSION,
2519         .hdr.size = sizeof(msg.payload.u64),
2520     };
2521     msg.payload.u64 = session_id;
2522 
2523     if (!crypto_session) {
2524         error_report("vhost-user trying to send unhandled ioctl");
2525         return -ENOTSUP;
2526     }
2527 
2528     ret = vhost_user_write(dev, &msg, NULL, 0);
2529     if (ret < 0) {
2530         error_report("vhost_user_write() return %d, close session failed",
2531                      ret);
2532         return ret;
2533     }
2534 
2535     return 0;
2536 }
2537 
2538 static bool vhost_user_mem_section_filter(struct vhost_dev *dev,
2539                                           MemoryRegionSection *section)
2540 {
2541     return memory_region_get_fd(section->mr) >= 0;
2542 }
2543 
2544 static int vhost_user_get_inflight_fd(struct vhost_dev *dev,
2545                                       uint16_t queue_size,
2546                                       struct vhost_inflight *inflight)
2547 {
2548     void *addr;
2549     int fd;
2550     int ret;
2551     struct vhost_user *u = dev->opaque;
2552     CharBackend *chr = u->user->chr;
2553     VhostUserMsg msg = {
2554         .hdr.request = VHOST_USER_GET_INFLIGHT_FD,
2555         .hdr.flags = VHOST_USER_VERSION,
2556         .payload.inflight.num_queues = dev->nvqs,
2557         .payload.inflight.queue_size = queue_size,
2558         .hdr.size = sizeof(msg.payload.inflight),
2559     };
2560 
2561     if (!virtio_has_feature(dev->protocol_features,
2562                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2563         return 0;
2564     }
2565 
2566     ret = vhost_user_write(dev, &msg, NULL, 0);
2567     if (ret < 0) {
2568         return ret;
2569     }
2570 
2571     ret = vhost_user_read(dev, &msg);
2572     if (ret < 0) {
2573         return ret;
2574     }
2575 
2576     if (msg.hdr.request != VHOST_USER_GET_INFLIGHT_FD) {
2577         error_report("Received unexpected msg type. "
2578                      "Expected %d received %d",
2579                      VHOST_USER_GET_INFLIGHT_FD, msg.hdr.request);
2580         return -EPROTO;
2581     }
2582 
2583     if (msg.hdr.size != sizeof(msg.payload.inflight)) {
2584         error_report("Received bad msg size.");
2585         return -EPROTO;
2586     }
2587 
2588     if (!msg.payload.inflight.mmap_size) {
2589         return 0;
2590     }
2591 
2592     fd = qemu_chr_fe_get_msgfd(chr);
2593     if (fd < 0) {
2594         error_report("Failed to get mem fd");
2595         return -EIO;
2596     }
2597 
2598     addr = mmap(0, msg.payload.inflight.mmap_size, PROT_READ | PROT_WRITE,
2599                 MAP_SHARED, fd, msg.payload.inflight.mmap_offset);
2600 
2601     if (addr == MAP_FAILED) {
2602         error_report("Failed to mmap mem fd");
2603         close(fd);
2604         return -EFAULT;
2605     }
2606 
2607     inflight->addr = addr;
2608     inflight->fd = fd;
2609     inflight->size = msg.payload.inflight.mmap_size;
2610     inflight->offset = msg.payload.inflight.mmap_offset;
2611     inflight->queue_size = queue_size;
2612 
2613     return 0;
2614 }
2615 
2616 static int vhost_user_set_inflight_fd(struct vhost_dev *dev,
2617                                       struct vhost_inflight *inflight)
2618 {
2619     VhostUserMsg msg = {
2620         .hdr.request = VHOST_USER_SET_INFLIGHT_FD,
2621         .hdr.flags = VHOST_USER_VERSION,
2622         .payload.inflight.mmap_size = inflight->size,
2623         .payload.inflight.mmap_offset = inflight->offset,
2624         .payload.inflight.num_queues = dev->nvqs,
2625         .payload.inflight.queue_size = inflight->queue_size,
2626         .hdr.size = sizeof(msg.payload.inflight),
2627     };
2628 
2629     if (!virtio_has_feature(dev->protocol_features,
2630                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2631         return 0;
2632     }
2633 
2634     return vhost_user_write(dev, &msg, &inflight->fd, 1);
2635 }
2636 
2637 static void vhost_user_state_destroy(gpointer data)
2638 {
2639     VhostUserHostNotifier *n = (VhostUserHostNotifier *) data;
2640     if (n) {
2641         vhost_user_host_notifier_remove(n, NULL);
2642         object_unparent(OBJECT(&n->mr));
2643         /*
2644          * We can't free until vhost_user_host_notifier_remove has
2645          * done it's thing so schedule the free with RCU.
2646          */
2647         g_free_rcu(n, rcu);
2648     }
2649 }
2650 
2651 bool vhost_user_init(VhostUserState *user, CharBackend *chr, Error **errp)
2652 {
2653     if (user->chr) {
2654         error_setg(errp, "Cannot initialize vhost-user state");
2655         return false;
2656     }
2657     user->chr = chr;
2658     user->memory_slots = 0;
2659     user->notifiers = g_ptr_array_new_full(VIRTIO_QUEUE_MAX / 4,
2660                                            &vhost_user_state_destroy);
2661     return true;
2662 }
2663 
2664 void vhost_user_cleanup(VhostUserState *user)
2665 {
2666     if (!user->chr) {
2667         return;
2668     }
2669     memory_region_transaction_begin();
2670     user->notifiers = (GPtrArray *) g_ptr_array_free(user->notifiers, true);
2671     memory_region_transaction_commit();
2672     user->chr = NULL;
2673 }
2674 
2675 
2676 typedef struct {
2677     vu_async_close_fn cb;
2678     DeviceState *dev;
2679     CharBackend *cd;
2680     struct vhost_dev *vhost;
2681 } VhostAsyncCallback;
2682 
2683 static void vhost_user_async_close_bh(void *opaque)
2684 {
2685     VhostAsyncCallback *data = opaque;
2686     struct vhost_dev *vhost = data->vhost;
2687 
2688     /*
2689      * If the vhost_dev has been cleared in the meantime there is
2690      * nothing left to do as some other path has completed the
2691      * cleanup.
2692      */
2693     if (vhost->vdev) {
2694         data->cb(data->dev);
2695     }
2696 
2697     g_free(data);
2698 }
2699 
2700 /*
2701  * We only schedule the work if the machine is running. If suspended
2702  * we want to keep all the in-flight data as is for migration
2703  * purposes.
2704  */
2705 void vhost_user_async_close(DeviceState *d,
2706                             CharBackend *chardev, struct vhost_dev *vhost,
2707                             vu_async_close_fn cb)
2708 {
2709     if (!runstate_check(RUN_STATE_SHUTDOWN)) {
2710         /*
2711          * A close event may happen during a read/write, but vhost
2712          * code assumes the vhost_dev remains setup, so delay the
2713          * stop & clear.
2714          */
2715         AioContext *ctx = qemu_get_current_aio_context();
2716         VhostAsyncCallback *data = g_new0(VhostAsyncCallback, 1);
2717 
2718         /* Save data for the callback */
2719         data->cb = cb;
2720         data->dev = d;
2721         data->cd = chardev;
2722         data->vhost = vhost;
2723 
2724         /* Disable any further notifications on the chardev */
2725         qemu_chr_fe_set_handlers(chardev,
2726                                  NULL, NULL, NULL, NULL, NULL, NULL,
2727                                  false);
2728 
2729         aio_bh_schedule_oneshot(ctx, vhost_user_async_close_bh, data);
2730 
2731         /*
2732          * Move vhost device to the stopped state. The vhost-user device
2733          * will be clean up and disconnected in BH. This can be useful in
2734          * the vhost migration code. If disconnect was caught there is an
2735          * option for the general vhost code to get the dev state without
2736          * knowing its type (in this case vhost-user).
2737          *
2738          * Note if the vhost device is fully cleared by the time we
2739          * execute the bottom half we won't continue with the cleanup.
2740          */
2741         vhost->started = false;
2742     }
2743 }
2744 
2745 static int vhost_user_dev_start(struct vhost_dev *dev, bool started)
2746 {
2747     if (!virtio_has_feature(dev->protocol_features,
2748                             VHOST_USER_PROTOCOL_F_STATUS)) {
2749         return 0;
2750     }
2751 
2752     /* Set device status only for last queue pair */
2753     if (dev->vq_index + dev->nvqs != dev->vq_index_end) {
2754         return 0;
2755     }
2756 
2757     if (started) {
2758         return vhost_user_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE |
2759                                           VIRTIO_CONFIG_S_DRIVER |
2760                                           VIRTIO_CONFIG_S_DRIVER_OK);
2761     } else {
2762         return vhost_user_set_status(dev, 0);
2763     }
2764 }
2765 
2766 const VhostOps user_ops = {
2767         .backend_type = VHOST_BACKEND_TYPE_USER,
2768         .vhost_backend_init = vhost_user_backend_init,
2769         .vhost_backend_cleanup = vhost_user_backend_cleanup,
2770         .vhost_backend_memslots_limit = vhost_user_memslots_limit,
2771         .vhost_set_log_base = vhost_user_set_log_base,
2772         .vhost_set_mem_table = vhost_user_set_mem_table,
2773         .vhost_set_vring_addr = vhost_user_set_vring_addr,
2774         .vhost_set_vring_endian = vhost_user_set_vring_endian,
2775         .vhost_set_vring_num = vhost_user_set_vring_num,
2776         .vhost_set_vring_base = vhost_user_set_vring_base,
2777         .vhost_get_vring_base = vhost_user_get_vring_base,
2778         .vhost_set_vring_kick = vhost_user_set_vring_kick,
2779         .vhost_set_vring_call = vhost_user_set_vring_call,
2780         .vhost_set_vring_err = vhost_user_set_vring_err,
2781         .vhost_set_features = vhost_user_set_features,
2782         .vhost_get_features = vhost_user_get_features,
2783         .vhost_set_owner = vhost_user_set_owner,
2784         .vhost_reset_device = vhost_user_reset_device,
2785         .vhost_get_vq_index = vhost_user_get_vq_index,
2786         .vhost_set_vring_enable = vhost_user_set_vring_enable,
2787         .vhost_requires_shm_log = vhost_user_requires_shm_log,
2788         .vhost_migration_done = vhost_user_migration_done,
2789         .vhost_backend_can_merge = vhost_user_can_merge,
2790         .vhost_net_set_mtu = vhost_user_net_set_mtu,
2791         .vhost_set_iotlb_callback = vhost_user_set_iotlb_callback,
2792         .vhost_send_device_iotlb_msg = vhost_user_send_device_iotlb_msg,
2793         .vhost_get_config = vhost_user_get_config,
2794         .vhost_set_config = vhost_user_set_config,
2795         .vhost_crypto_create_session = vhost_user_crypto_create_session,
2796         .vhost_crypto_close_session = vhost_user_crypto_close_session,
2797         .vhost_backend_mem_section_filter = vhost_user_mem_section_filter,
2798         .vhost_get_inflight_fd = vhost_user_get_inflight_fd,
2799         .vhost_set_inflight_fd = vhost_user_set_inflight_fd,
2800         .vhost_dev_start = vhost_user_dev_start,
2801 };
2802