xref: /openbmc/qemu/hw/virtio/vhost-user.c (revision 0b29090a)
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_BACKEND_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_AARCH64)
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_BACKEND_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_BACKEND_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_BACKEND_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_BACKEND_NONE = 0,
138     VHOST_USER_BACKEND_IOTLB_MSG = 1,
139     VHOST_USER_BACKEND_CONFIG_CHANGE_MSG = 2,
140     VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG = 3,
141     VHOST_USER_BACKEND_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 static int vhost_user_read(struct vhost_dev *dev, VhostUserMsg *msg)
309 {
310     struct vhost_user *u = dev->opaque;
311     CharBackend *chr = u->user->chr;
312     uint8_t *p = (uint8_t *) msg;
313     int r, size;
314 
315     r = vhost_user_read_header(dev, msg);
316     if (r < 0) {
317         return r;
318     }
319 
320     /* validate message size is sane */
321     if (msg->hdr.size > VHOST_USER_PAYLOAD_SIZE) {
322         error_report("Failed to read msg header."
323                 " Size %d exceeds the maximum %zu.", msg->hdr.size,
324                 VHOST_USER_PAYLOAD_SIZE);
325         return -EPROTO;
326     }
327 
328     if (msg->hdr.size) {
329         p += VHOST_USER_HDR_SIZE;
330         size = msg->hdr.size;
331         r = qemu_chr_fe_read_all(chr, p, size);
332         if (r != size) {
333             int saved_errno = errno;
334             error_report("Failed to read msg payload."
335                          " Read %d instead of %d.", r, msg->hdr.size);
336             return r < 0 ? -saved_errno : -EIO;
337         }
338     }
339 
340     return 0;
341 }
342 
343 static int process_message_reply(struct vhost_dev *dev,
344                                  const VhostUserMsg *msg)
345 {
346     int ret;
347     VhostUserMsg msg_reply;
348 
349     if ((msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) == 0) {
350         return 0;
351     }
352 
353     ret = vhost_user_read(dev, &msg_reply);
354     if (ret < 0) {
355         return ret;
356     }
357 
358     if (msg_reply.hdr.request != msg->hdr.request) {
359         error_report("Received unexpected msg type. "
360                      "Expected %d received %d",
361                      msg->hdr.request, msg_reply.hdr.request);
362         return -EPROTO;
363     }
364 
365     return msg_reply.payload.u64 ? -EIO : 0;
366 }
367 
368 static bool vhost_user_one_time_request(VhostUserRequest request)
369 {
370     switch (request) {
371     case VHOST_USER_SET_OWNER:
372     case VHOST_USER_RESET_OWNER:
373     case VHOST_USER_SET_MEM_TABLE:
374     case VHOST_USER_GET_QUEUE_NUM:
375     case VHOST_USER_NET_SET_MTU:
376     case VHOST_USER_ADD_MEM_REG:
377     case VHOST_USER_REM_MEM_REG:
378         return true;
379     default:
380         return false;
381     }
382 }
383 
384 /* most non-init callers ignore the error */
385 static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
386                             int *fds, int fd_num)
387 {
388     struct vhost_user *u = dev->opaque;
389     CharBackend *chr = u->user->chr;
390     int ret, size = VHOST_USER_HDR_SIZE + msg->hdr.size;
391 
392     /*
393      * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
394      * we just need send it once in the first time. For later such
395      * request, we just ignore it.
396      */
397     if (vhost_user_one_time_request(msg->hdr.request) && dev->vq_index != 0) {
398         msg->hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
399         return 0;
400     }
401 
402     if (qemu_chr_fe_set_msgfds(chr, fds, fd_num) < 0) {
403         error_report("Failed to set msg fds.");
404         return -EINVAL;
405     }
406 
407     ret = qemu_chr_fe_write_all(chr, (const uint8_t *) msg, size);
408     if (ret != size) {
409         int saved_errno = errno;
410         error_report("Failed to write msg."
411                      " Wrote %d instead of %d.", ret, size);
412         return ret < 0 ? -saved_errno : -EIO;
413     }
414 
415     trace_vhost_user_write(msg->hdr.request, msg->hdr.flags);
416 
417     return 0;
418 }
419 
420 int vhost_user_gpu_set_socket(struct vhost_dev *dev, int fd)
421 {
422     VhostUserMsg msg = {
423         .hdr.request = VHOST_USER_GPU_SET_SOCKET,
424         .hdr.flags = VHOST_USER_VERSION,
425     };
426 
427     return vhost_user_write(dev, &msg, &fd, 1);
428 }
429 
430 static int vhost_user_set_log_base(struct vhost_dev *dev, uint64_t base,
431                                    struct vhost_log *log)
432 {
433     int fds[VHOST_USER_MAX_RAM_SLOTS];
434     size_t fd_num = 0;
435     bool shmfd = virtio_has_feature(dev->protocol_features,
436                                     VHOST_USER_PROTOCOL_F_LOG_SHMFD);
437     int ret;
438     VhostUserMsg msg = {
439         .hdr.request = VHOST_USER_SET_LOG_BASE,
440         .hdr.flags = VHOST_USER_VERSION,
441         .payload.log.mmap_size = log->size * sizeof(*(log->log)),
442         .payload.log.mmap_offset = 0,
443         .hdr.size = sizeof(msg.payload.log),
444     };
445 
446     /* Send only once with first queue pair */
447     if (dev->vq_index != 0) {
448         return 0;
449     }
450 
451     if (shmfd && log->fd != -1) {
452         fds[fd_num++] = log->fd;
453     }
454 
455     ret = vhost_user_write(dev, &msg, fds, fd_num);
456     if (ret < 0) {
457         return ret;
458     }
459 
460     if (shmfd) {
461         msg.hdr.size = 0;
462         ret = vhost_user_read(dev, &msg);
463         if (ret < 0) {
464             return ret;
465         }
466 
467         if (msg.hdr.request != VHOST_USER_SET_LOG_BASE) {
468             error_report("Received unexpected msg type. "
469                          "Expected %d received %d",
470                          VHOST_USER_SET_LOG_BASE, msg.hdr.request);
471             return -EPROTO;
472         }
473     }
474 
475     return 0;
476 }
477 
478 static MemoryRegion *vhost_user_get_mr_data(uint64_t addr, ram_addr_t *offset,
479                                             int *fd)
480 {
481     MemoryRegion *mr;
482 
483     assert((uintptr_t)addr == addr);
484     mr = memory_region_from_host((void *)(uintptr_t)addr, offset);
485     *fd = memory_region_get_fd(mr);
486 
487     return mr;
488 }
489 
490 static void vhost_user_fill_msg_region(VhostUserMemoryRegion *dst,
491                                        struct vhost_memory_region *src,
492                                        uint64_t mmap_offset)
493 {
494     assert(src != NULL && dst != NULL);
495     dst->userspace_addr = src->userspace_addr;
496     dst->memory_size = src->memory_size;
497     dst->guest_phys_addr = src->guest_phys_addr;
498     dst->mmap_offset = mmap_offset;
499 }
500 
501 static int vhost_user_fill_set_mem_table_msg(struct vhost_user *u,
502                                              struct vhost_dev *dev,
503                                              VhostUserMsg *msg,
504                                              int *fds, size_t *fd_num,
505                                              bool track_ramblocks)
506 {
507     int i, fd;
508     ram_addr_t offset;
509     MemoryRegion *mr;
510     struct vhost_memory_region *reg;
511     VhostUserMemoryRegion region_buffer;
512 
513     msg->hdr.request = VHOST_USER_SET_MEM_TABLE;
514 
515     for (i = 0; i < dev->mem->nregions; ++i) {
516         reg = dev->mem->regions + i;
517 
518         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
519         if (fd > 0) {
520             if (track_ramblocks) {
521                 assert(*fd_num < VHOST_MEMORY_BASELINE_NREGIONS);
522                 trace_vhost_user_set_mem_table_withfd(*fd_num, mr->name,
523                                                       reg->memory_size,
524                                                       reg->guest_phys_addr,
525                                                       reg->userspace_addr,
526                                                       offset);
527                 u->region_rb_offset[i] = offset;
528                 u->region_rb[i] = mr->ram_block;
529             } else if (*fd_num == VHOST_MEMORY_BASELINE_NREGIONS) {
530                 error_report("Failed preparing vhost-user memory table msg");
531                 return -ENOBUFS;
532             }
533             vhost_user_fill_msg_region(&region_buffer, reg, offset);
534             msg->payload.memory.regions[*fd_num] = region_buffer;
535             fds[(*fd_num)++] = fd;
536         } else if (track_ramblocks) {
537             u->region_rb_offset[i] = 0;
538             u->region_rb[i] = NULL;
539         }
540     }
541 
542     msg->payload.memory.nregions = *fd_num;
543 
544     if (!*fd_num) {
545         error_report("Failed initializing vhost-user memory map, "
546                      "consider using -object memory-backend-file share=on");
547         return -EINVAL;
548     }
549 
550     msg->hdr.size = sizeof(msg->payload.memory.nregions);
551     msg->hdr.size += sizeof(msg->payload.memory.padding);
552     msg->hdr.size += *fd_num * sizeof(VhostUserMemoryRegion);
553 
554     return 0;
555 }
556 
557 static inline bool reg_equal(struct vhost_memory_region *shadow_reg,
558                              struct vhost_memory_region *vdev_reg)
559 {
560     return shadow_reg->guest_phys_addr == vdev_reg->guest_phys_addr &&
561         shadow_reg->userspace_addr == vdev_reg->userspace_addr &&
562         shadow_reg->memory_size == vdev_reg->memory_size;
563 }
564 
565 static void scrub_shadow_regions(struct vhost_dev *dev,
566                                  struct scrub_regions *add_reg,
567                                  int *nr_add_reg,
568                                  struct scrub_regions *rem_reg,
569                                  int *nr_rem_reg, uint64_t *shadow_pcb,
570                                  bool track_ramblocks)
571 {
572     struct vhost_user *u = dev->opaque;
573     bool found[VHOST_USER_MAX_RAM_SLOTS] = {};
574     struct vhost_memory_region *reg, *shadow_reg;
575     int i, j, fd, add_idx = 0, rm_idx = 0, fd_num = 0;
576     ram_addr_t offset;
577     MemoryRegion *mr;
578     bool matching;
579 
580     /*
581      * Find memory regions present in our shadow state which are not in
582      * the device's current memory state.
583      *
584      * Mark regions in both the shadow and device state as "found".
585      */
586     for (i = 0; i < u->num_shadow_regions; i++) {
587         shadow_reg = &u->shadow_regions[i];
588         matching = false;
589 
590         for (j = 0; j < dev->mem->nregions; j++) {
591             reg = &dev->mem->regions[j];
592 
593             mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
594 
595             if (reg_equal(shadow_reg, reg)) {
596                 matching = true;
597                 found[j] = true;
598                 if (track_ramblocks) {
599                     /*
600                      * Reset postcopy client bases, region_rb, and
601                      * region_rb_offset in case regions are removed.
602                      */
603                     if (fd > 0) {
604                         u->region_rb_offset[j] = offset;
605                         u->region_rb[j] = mr->ram_block;
606                         shadow_pcb[j] = u->postcopy_client_bases[i];
607                     } else {
608                         u->region_rb_offset[j] = 0;
609                         u->region_rb[j] = NULL;
610                     }
611                 }
612                 break;
613             }
614         }
615 
616         /*
617          * If the region was not found in the current device memory state
618          * create an entry for it in the removed list.
619          */
620         if (!matching) {
621             rem_reg[rm_idx].region = shadow_reg;
622             rem_reg[rm_idx++].reg_idx = i;
623         }
624     }
625 
626     /*
627      * For regions not marked "found", create entries in the added list.
628      *
629      * Note their indexes in the device memory state and the indexes of their
630      * file descriptors.
631      */
632     for (i = 0; i < dev->mem->nregions; i++) {
633         reg = &dev->mem->regions[i];
634         vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
635         if (fd > 0) {
636             ++fd_num;
637         }
638 
639         /*
640          * If the region was in both the shadow and device state we don't
641          * need to send a VHOST_USER_ADD_MEM_REG message for it.
642          */
643         if (found[i]) {
644             continue;
645         }
646 
647         add_reg[add_idx].region = reg;
648         add_reg[add_idx].reg_idx = i;
649         add_reg[add_idx++].fd_idx = fd_num;
650     }
651     *nr_rem_reg = rm_idx;
652     *nr_add_reg = add_idx;
653 
654     return;
655 }
656 
657 static int send_remove_regions(struct vhost_dev *dev,
658                                struct scrub_regions *remove_reg,
659                                int nr_rem_reg, VhostUserMsg *msg,
660                                bool reply_supported)
661 {
662     struct vhost_user *u = dev->opaque;
663     struct vhost_memory_region *shadow_reg;
664     int i, fd, shadow_reg_idx, ret;
665     ram_addr_t offset;
666     VhostUserMemoryRegion region_buffer;
667 
668     /*
669      * The regions in remove_reg appear in the same order they do in the
670      * shadow table. Therefore we can minimize memory copies by iterating
671      * through remove_reg backwards.
672      */
673     for (i = nr_rem_reg - 1; i >= 0; i--) {
674         shadow_reg = remove_reg[i].region;
675         shadow_reg_idx = remove_reg[i].reg_idx;
676 
677         vhost_user_get_mr_data(shadow_reg->userspace_addr, &offset, &fd);
678 
679         if (fd > 0) {
680             msg->hdr.request = VHOST_USER_REM_MEM_REG;
681             vhost_user_fill_msg_region(&region_buffer, shadow_reg, 0);
682             msg->payload.mem_reg.region = region_buffer;
683 
684             ret = vhost_user_write(dev, msg, NULL, 0);
685             if (ret < 0) {
686                 return ret;
687             }
688 
689             if (reply_supported) {
690                 ret = process_message_reply(dev, msg);
691                 if (ret) {
692                     return ret;
693                 }
694             }
695         }
696 
697         /*
698          * At this point we know the backend has unmapped the region. It is now
699          * safe to remove it from the shadow table.
700          */
701         memmove(&u->shadow_regions[shadow_reg_idx],
702                 &u->shadow_regions[shadow_reg_idx + 1],
703                 sizeof(struct vhost_memory_region) *
704                 (u->num_shadow_regions - shadow_reg_idx - 1));
705         u->num_shadow_regions--;
706     }
707 
708     return 0;
709 }
710 
711 static int send_add_regions(struct vhost_dev *dev,
712                             struct scrub_regions *add_reg, int nr_add_reg,
713                             VhostUserMsg *msg, uint64_t *shadow_pcb,
714                             bool reply_supported, bool track_ramblocks)
715 {
716     struct vhost_user *u = dev->opaque;
717     int i, fd, ret, reg_idx, reg_fd_idx;
718     struct vhost_memory_region *reg;
719     MemoryRegion *mr;
720     ram_addr_t offset;
721     VhostUserMsg msg_reply;
722     VhostUserMemoryRegion region_buffer;
723 
724     for (i = 0; i < nr_add_reg; i++) {
725         reg = add_reg[i].region;
726         reg_idx = add_reg[i].reg_idx;
727         reg_fd_idx = add_reg[i].fd_idx;
728 
729         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
730 
731         if (fd > 0) {
732             if (track_ramblocks) {
733                 trace_vhost_user_set_mem_table_withfd(reg_fd_idx, mr->name,
734                                                       reg->memory_size,
735                                                       reg->guest_phys_addr,
736                                                       reg->userspace_addr,
737                                                       offset);
738                 u->region_rb_offset[reg_idx] = offset;
739                 u->region_rb[reg_idx] = mr->ram_block;
740             }
741             msg->hdr.request = VHOST_USER_ADD_MEM_REG;
742             vhost_user_fill_msg_region(&region_buffer, reg, offset);
743             msg->payload.mem_reg.region = region_buffer;
744 
745             ret = vhost_user_write(dev, msg, &fd, 1);
746             if (ret < 0) {
747                 return ret;
748             }
749 
750             if (track_ramblocks) {
751                 uint64_t reply_gpa;
752 
753                 ret = vhost_user_read(dev, &msg_reply);
754                 if (ret < 0) {
755                     return ret;
756                 }
757 
758                 reply_gpa = msg_reply.payload.mem_reg.region.guest_phys_addr;
759 
760                 if (msg_reply.hdr.request != VHOST_USER_ADD_MEM_REG) {
761                     error_report("%s: Received unexpected msg type."
762                                  "Expected %d received %d", __func__,
763                                  VHOST_USER_ADD_MEM_REG,
764                                  msg_reply.hdr.request);
765                     return -EPROTO;
766                 }
767 
768                 /*
769                  * We're using the same structure, just reusing one of the
770                  * fields, so it should be the same size.
771                  */
772                 if (msg_reply.hdr.size != msg->hdr.size) {
773                     error_report("%s: Unexpected size for postcopy reply "
774                                  "%d vs %d", __func__, msg_reply.hdr.size,
775                                  msg->hdr.size);
776                     return -EPROTO;
777                 }
778 
779                 /* Get the postcopy client base from the backend's reply. */
780                 if (reply_gpa == dev->mem->regions[reg_idx].guest_phys_addr) {
781                     shadow_pcb[reg_idx] =
782                         msg_reply.payload.mem_reg.region.userspace_addr;
783                     trace_vhost_user_set_mem_table_postcopy(
784                         msg_reply.payload.mem_reg.region.userspace_addr,
785                         msg->payload.mem_reg.region.userspace_addr,
786                         reg_fd_idx, reg_idx);
787                 } else {
788                     error_report("%s: invalid postcopy reply for region. "
789                                  "Got guest physical address %" PRIX64 ", expected "
790                                  "%" PRIX64, __func__, reply_gpa,
791                                  dev->mem->regions[reg_idx].guest_phys_addr);
792                     return -EPROTO;
793                 }
794             } else if (reply_supported) {
795                 ret = process_message_reply(dev, msg);
796                 if (ret) {
797                     return ret;
798                 }
799             }
800         } else if (track_ramblocks) {
801             u->region_rb_offset[reg_idx] = 0;
802             u->region_rb[reg_idx] = NULL;
803         }
804 
805         /*
806          * At this point, we know the backend has mapped in the new
807          * region, if the region has a valid file descriptor.
808          *
809          * The region should now be added to the shadow table.
810          */
811         u->shadow_regions[u->num_shadow_regions].guest_phys_addr =
812             reg->guest_phys_addr;
813         u->shadow_regions[u->num_shadow_regions].userspace_addr =
814             reg->userspace_addr;
815         u->shadow_regions[u->num_shadow_regions].memory_size =
816             reg->memory_size;
817         u->num_shadow_regions++;
818     }
819 
820     return 0;
821 }
822 
823 static int vhost_user_add_remove_regions(struct vhost_dev *dev,
824                                          VhostUserMsg *msg,
825                                          bool reply_supported,
826                                          bool track_ramblocks)
827 {
828     struct vhost_user *u = dev->opaque;
829     struct scrub_regions add_reg[VHOST_USER_MAX_RAM_SLOTS];
830     struct scrub_regions rem_reg[VHOST_USER_MAX_RAM_SLOTS];
831     uint64_t shadow_pcb[VHOST_USER_MAX_RAM_SLOTS] = {};
832     int nr_add_reg, nr_rem_reg;
833     int ret;
834 
835     msg->hdr.size = sizeof(msg->payload.mem_reg);
836 
837     /* Find the regions which need to be removed or added. */
838     scrub_shadow_regions(dev, add_reg, &nr_add_reg, rem_reg, &nr_rem_reg,
839                          shadow_pcb, track_ramblocks);
840 
841     if (nr_rem_reg) {
842         ret = send_remove_regions(dev, rem_reg, nr_rem_reg, msg,
843                                   reply_supported);
844         if (ret < 0) {
845             goto err;
846         }
847     }
848 
849     if (nr_add_reg) {
850         ret = send_add_regions(dev, add_reg, nr_add_reg, msg, shadow_pcb,
851                                reply_supported, track_ramblocks);
852         if (ret < 0) {
853             goto err;
854         }
855     }
856 
857     if (track_ramblocks) {
858         memcpy(u->postcopy_client_bases, shadow_pcb,
859                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
860         /*
861          * Now we've registered this with the postcopy code, we ack to the
862          * client, because now we're in the position to be able to deal with
863          * any faults it generates.
864          */
865         /* TODO: Use this for failure cases as well with a bad value. */
866         msg->hdr.size = sizeof(msg->payload.u64);
867         msg->payload.u64 = 0; /* OK */
868 
869         ret = vhost_user_write(dev, msg, NULL, 0);
870         if (ret < 0) {
871             return ret;
872         }
873     }
874 
875     return 0;
876 
877 err:
878     if (track_ramblocks) {
879         memcpy(u->postcopy_client_bases, shadow_pcb,
880                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
881     }
882 
883     return ret;
884 }
885 
886 static int vhost_user_set_mem_table_postcopy(struct vhost_dev *dev,
887                                              struct vhost_memory *mem,
888                                              bool reply_supported,
889                                              bool config_mem_slots)
890 {
891     struct vhost_user *u = dev->opaque;
892     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
893     size_t fd_num = 0;
894     VhostUserMsg msg_reply;
895     int region_i, msg_i;
896     int ret;
897 
898     VhostUserMsg msg = {
899         .hdr.flags = VHOST_USER_VERSION,
900     };
901 
902     if (u->region_rb_len < dev->mem->nregions) {
903         u->region_rb = g_renew(RAMBlock*, u->region_rb, dev->mem->nregions);
904         u->region_rb_offset = g_renew(ram_addr_t, u->region_rb_offset,
905                                       dev->mem->nregions);
906         memset(&(u->region_rb[u->region_rb_len]), '\0',
907                sizeof(RAMBlock *) * (dev->mem->nregions - u->region_rb_len));
908         memset(&(u->region_rb_offset[u->region_rb_len]), '\0',
909                sizeof(ram_addr_t) * (dev->mem->nregions - u->region_rb_len));
910         u->region_rb_len = dev->mem->nregions;
911     }
912 
913     if (config_mem_slots) {
914         ret = vhost_user_add_remove_regions(dev, &msg, reply_supported, true);
915         if (ret < 0) {
916             return ret;
917         }
918     } else {
919         ret = vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
920                                                 true);
921         if (ret < 0) {
922             return ret;
923         }
924 
925         ret = vhost_user_write(dev, &msg, fds, fd_num);
926         if (ret < 0) {
927             return ret;
928         }
929 
930         ret = vhost_user_read(dev, &msg_reply);
931         if (ret < 0) {
932             return ret;
933         }
934 
935         if (msg_reply.hdr.request != VHOST_USER_SET_MEM_TABLE) {
936             error_report("%s: Received unexpected msg type."
937                          "Expected %d received %d", __func__,
938                          VHOST_USER_SET_MEM_TABLE, msg_reply.hdr.request);
939             return -EPROTO;
940         }
941 
942         /*
943          * We're using the same structure, just reusing one of the
944          * fields, so it should be the same size.
945          */
946         if (msg_reply.hdr.size != msg.hdr.size) {
947             error_report("%s: Unexpected size for postcopy reply "
948                          "%d vs %d", __func__, msg_reply.hdr.size,
949                          msg.hdr.size);
950             return -EPROTO;
951         }
952 
953         memset(u->postcopy_client_bases, 0,
954                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
955 
956         /*
957          * They're in the same order as the regions that were sent
958          * but some of the regions were skipped (above) if they
959          * didn't have fd's
960          */
961         for (msg_i = 0, region_i = 0;
962              region_i < dev->mem->nregions;
963              region_i++) {
964             if (msg_i < fd_num &&
965                 msg_reply.payload.memory.regions[msg_i].guest_phys_addr ==
966                 dev->mem->regions[region_i].guest_phys_addr) {
967                 u->postcopy_client_bases[region_i] =
968                     msg_reply.payload.memory.regions[msg_i].userspace_addr;
969                 trace_vhost_user_set_mem_table_postcopy(
970                     msg_reply.payload.memory.regions[msg_i].userspace_addr,
971                     msg.payload.memory.regions[msg_i].userspace_addr,
972                     msg_i, region_i);
973                 msg_i++;
974             }
975         }
976         if (msg_i != fd_num) {
977             error_report("%s: postcopy reply not fully consumed "
978                          "%d vs %zd",
979                          __func__, msg_i, fd_num);
980             return -EIO;
981         }
982 
983         /*
984          * Now we've registered this with the postcopy code, we ack to the
985          * client, because now we're in the position to be able to deal
986          * with any faults it generates.
987          */
988         /* TODO: Use this for failure cases as well with a bad value. */
989         msg.hdr.size = sizeof(msg.payload.u64);
990         msg.payload.u64 = 0; /* OK */
991         ret = vhost_user_write(dev, &msg, NULL, 0);
992         if (ret < 0) {
993             return ret;
994         }
995     }
996 
997     return 0;
998 }
999 
1000 static int vhost_user_set_mem_table(struct vhost_dev *dev,
1001                                     struct vhost_memory *mem)
1002 {
1003     struct vhost_user *u = dev->opaque;
1004     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
1005     size_t fd_num = 0;
1006     bool do_postcopy = u->postcopy_listen && u->postcopy_fd.handler;
1007     bool reply_supported = virtio_has_feature(dev->protocol_features,
1008                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1009     bool config_mem_slots =
1010         virtio_has_feature(dev->protocol_features,
1011                            VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS);
1012     int ret;
1013 
1014     if (do_postcopy) {
1015         /*
1016          * Postcopy has enough differences that it's best done in it's own
1017          * version
1018          */
1019         return vhost_user_set_mem_table_postcopy(dev, mem, reply_supported,
1020                                                  config_mem_slots);
1021     }
1022 
1023     VhostUserMsg msg = {
1024         .hdr.flags = VHOST_USER_VERSION,
1025     };
1026 
1027     if (reply_supported) {
1028         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1029     }
1030 
1031     if (config_mem_slots) {
1032         ret = vhost_user_add_remove_regions(dev, &msg, reply_supported, false);
1033         if (ret < 0) {
1034             return ret;
1035         }
1036     } else {
1037         ret = vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
1038                                                 false);
1039         if (ret < 0) {
1040             return ret;
1041         }
1042 
1043         ret = vhost_user_write(dev, &msg, fds, fd_num);
1044         if (ret < 0) {
1045             return ret;
1046         }
1047 
1048         if (reply_supported) {
1049             return process_message_reply(dev, &msg);
1050         }
1051     }
1052 
1053     return 0;
1054 }
1055 
1056 static int vhost_user_set_vring_endian(struct vhost_dev *dev,
1057                                        struct vhost_vring_state *ring)
1058 {
1059     bool cross_endian = virtio_has_feature(dev->protocol_features,
1060                                            VHOST_USER_PROTOCOL_F_CROSS_ENDIAN);
1061     VhostUserMsg msg = {
1062         .hdr.request = VHOST_USER_SET_VRING_ENDIAN,
1063         .hdr.flags = VHOST_USER_VERSION,
1064         .payload.state = *ring,
1065         .hdr.size = sizeof(msg.payload.state),
1066     };
1067 
1068     if (!cross_endian) {
1069         error_report("vhost-user trying to send unhandled ioctl");
1070         return -ENOTSUP;
1071     }
1072 
1073     return vhost_user_write(dev, &msg, NULL, 0);
1074 }
1075 
1076 static int vhost_set_vring(struct vhost_dev *dev,
1077                            unsigned long int request,
1078                            struct vhost_vring_state *ring)
1079 {
1080     VhostUserMsg msg = {
1081         .hdr.request = request,
1082         .hdr.flags = VHOST_USER_VERSION,
1083         .payload.state = *ring,
1084         .hdr.size = sizeof(msg.payload.state),
1085     };
1086 
1087     return vhost_user_write(dev, &msg, NULL, 0);
1088 }
1089 
1090 static int vhost_user_set_vring_num(struct vhost_dev *dev,
1091                                     struct vhost_vring_state *ring)
1092 {
1093     return vhost_set_vring(dev, VHOST_USER_SET_VRING_NUM, ring);
1094 }
1095 
1096 static void vhost_user_host_notifier_free(VhostUserHostNotifier *n)
1097 {
1098     assert(n && n->unmap_addr);
1099     munmap(n->unmap_addr, qemu_real_host_page_size());
1100     n->unmap_addr = NULL;
1101 }
1102 
1103 /*
1104  * clean-up function for notifier, will finally free the structure
1105  * under rcu.
1106  */
1107 static void vhost_user_host_notifier_remove(VhostUserHostNotifier *n,
1108                                             VirtIODevice *vdev)
1109 {
1110     if (n->addr) {
1111         if (vdev) {
1112             virtio_queue_set_host_notifier_mr(vdev, n->idx, &n->mr, false);
1113         }
1114         assert(!n->unmap_addr);
1115         n->unmap_addr = n->addr;
1116         n->addr = NULL;
1117         call_rcu(n, vhost_user_host_notifier_free, rcu);
1118     }
1119 }
1120 
1121 static int vhost_user_set_vring_base(struct vhost_dev *dev,
1122                                      struct vhost_vring_state *ring)
1123 {
1124     return vhost_set_vring(dev, VHOST_USER_SET_VRING_BASE, ring);
1125 }
1126 
1127 static int vhost_user_set_vring_enable(struct vhost_dev *dev, int enable)
1128 {
1129     int i;
1130 
1131     if (!virtio_has_feature(dev->features, VHOST_USER_F_PROTOCOL_FEATURES)) {
1132         return -EINVAL;
1133     }
1134 
1135     for (i = 0; i < dev->nvqs; ++i) {
1136         int ret;
1137         struct vhost_vring_state state = {
1138             .index = dev->vq_index + i,
1139             .num   = enable,
1140         };
1141 
1142         ret = vhost_set_vring(dev, VHOST_USER_SET_VRING_ENABLE, &state);
1143         if (ret < 0) {
1144             /*
1145              * Restoring the previous state is likely infeasible, as well as
1146              * proceeding regardless the error, so just bail out and hope for
1147              * the device-level recovery.
1148              */
1149             return ret;
1150         }
1151     }
1152 
1153     return 0;
1154 }
1155 
1156 static VhostUserHostNotifier *fetch_notifier(VhostUserState *u,
1157                                              int idx)
1158 {
1159     if (idx >= u->notifiers->len) {
1160         return NULL;
1161     }
1162     return g_ptr_array_index(u->notifiers, idx);
1163 }
1164 
1165 static int vhost_user_get_vring_base(struct vhost_dev *dev,
1166                                      struct vhost_vring_state *ring)
1167 {
1168     int ret;
1169     VhostUserMsg msg = {
1170         .hdr.request = VHOST_USER_GET_VRING_BASE,
1171         .hdr.flags = VHOST_USER_VERSION,
1172         .payload.state = *ring,
1173         .hdr.size = sizeof(msg.payload.state),
1174     };
1175     struct vhost_user *u = dev->opaque;
1176 
1177     VhostUserHostNotifier *n = fetch_notifier(u->user, ring->index);
1178     if (n) {
1179         vhost_user_host_notifier_remove(n, dev->vdev);
1180     }
1181 
1182     ret = vhost_user_write(dev, &msg, NULL, 0);
1183     if (ret < 0) {
1184         return ret;
1185     }
1186 
1187     ret = vhost_user_read(dev, &msg);
1188     if (ret < 0) {
1189         return ret;
1190     }
1191 
1192     if (msg.hdr.request != VHOST_USER_GET_VRING_BASE) {
1193         error_report("Received unexpected msg type. Expected %d received %d",
1194                      VHOST_USER_GET_VRING_BASE, msg.hdr.request);
1195         return -EPROTO;
1196     }
1197 
1198     if (msg.hdr.size != sizeof(msg.payload.state)) {
1199         error_report("Received bad msg size.");
1200         return -EPROTO;
1201     }
1202 
1203     *ring = msg.payload.state;
1204 
1205     return 0;
1206 }
1207 
1208 static int vhost_set_vring_file(struct vhost_dev *dev,
1209                                 VhostUserRequest request,
1210                                 struct vhost_vring_file *file)
1211 {
1212     int fds[VHOST_USER_MAX_RAM_SLOTS];
1213     size_t fd_num = 0;
1214     VhostUserMsg msg = {
1215         .hdr.request = request,
1216         .hdr.flags = VHOST_USER_VERSION,
1217         .payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK,
1218         .hdr.size = sizeof(msg.payload.u64),
1219     };
1220 
1221     if (ioeventfd_enabled() && file->fd > 0) {
1222         fds[fd_num++] = file->fd;
1223     } else {
1224         msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
1225     }
1226 
1227     return vhost_user_write(dev, &msg, fds, fd_num);
1228 }
1229 
1230 static int vhost_user_set_vring_kick(struct vhost_dev *dev,
1231                                      struct vhost_vring_file *file)
1232 {
1233     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_KICK, file);
1234 }
1235 
1236 static int vhost_user_set_vring_call(struct vhost_dev *dev,
1237                                      struct vhost_vring_file *file)
1238 {
1239     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_CALL, file);
1240 }
1241 
1242 static int vhost_user_set_vring_err(struct vhost_dev *dev,
1243                                     struct vhost_vring_file *file)
1244 {
1245     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_ERR, file);
1246 }
1247 
1248 static int vhost_user_get_u64(struct vhost_dev *dev, int request, uint64_t *u64)
1249 {
1250     int ret;
1251     VhostUserMsg msg = {
1252         .hdr.request = request,
1253         .hdr.flags = VHOST_USER_VERSION,
1254     };
1255 
1256     if (vhost_user_one_time_request(request) && dev->vq_index != 0) {
1257         return 0;
1258     }
1259 
1260     ret = vhost_user_write(dev, &msg, NULL, 0);
1261     if (ret < 0) {
1262         return ret;
1263     }
1264 
1265     ret = vhost_user_read(dev, &msg);
1266     if (ret < 0) {
1267         return ret;
1268     }
1269 
1270     if (msg.hdr.request != request) {
1271         error_report("Received unexpected msg type. Expected %d received %d",
1272                      request, msg.hdr.request);
1273         return -EPROTO;
1274     }
1275 
1276     if (msg.hdr.size != sizeof(msg.payload.u64)) {
1277         error_report("Received bad msg size.");
1278         return -EPROTO;
1279     }
1280 
1281     *u64 = msg.payload.u64;
1282 
1283     return 0;
1284 }
1285 
1286 static int vhost_user_get_features(struct vhost_dev *dev, uint64_t *features)
1287 {
1288     if (vhost_user_get_u64(dev, VHOST_USER_GET_FEATURES, features) < 0) {
1289         return -EPROTO;
1290     }
1291 
1292     return 0;
1293 }
1294 
1295 static int enforce_reply(struct vhost_dev *dev,
1296                          const VhostUserMsg *msg)
1297 {
1298     uint64_t dummy;
1299 
1300     if (msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1301         return process_message_reply(dev, msg);
1302     }
1303 
1304    /*
1305     * We need to wait for a reply but the backend does not
1306     * support replies for the command we just sent.
1307     * Send VHOST_USER_GET_FEATURES which makes all backends
1308     * send a reply.
1309     */
1310     return vhost_user_get_features(dev, &dummy);
1311 }
1312 
1313 static int vhost_user_set_vring_addr(struct vhost_dev *dev,
1314                                      struct vhost_vring_addr *addr)
1315 {
1316     int ret;
1317     VhostUserMsg msg = {
1318         .hdr.request = VHOST_USER_SET_VRING_ADDR,
1319         .hdr.flags = VHOST_USER_VERSION,
1320         .payload.addr = *addr,
1321         .hdr.size = sizeof(msg.payload.addr),
1322     };
1323 
1324     bool reply_supported = virtio_has_feature(dev->protocol_features,
1325                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1326 
1327     /*
1328      * wait for a reply if logging is enabled to make sure
1329      * backend is actually logging changes
1330      */
1331     bool wait_for_reply = addr->flags & (1 << VHOST_VRING_F_LOG);
1332 
1333     if (reply_supported && wait_for_reply) {
1334         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1335     }
1336 
1337     ret = vhost_user_write(dev, &msg, NULL, 0);
1338     if (ret < 0) {
1339         return ret;
1340     }
1341 
1342     if (wait_for_reply) {
1343         return enforce_reply(dev, &msg);
1344     }
1345 
1346     return 0;
1347 }
1348 
1349 static int vhost_user_set_u64(struct vhost_dev *dev, int request, uint64_t u64,
1350                               bool wait_for_reply)
1351 {
1352     VhostUserMsg msg = {
1353         .hdr.request = request,
1354         .hdr.flags = VHOST_USER_VERSION,
1355         .payload.u64 = u64,
1356         .hdr.size = sizeof(msg.payload.u64),
1357     };
1358     int ret;
1359 
1360     if (wait_for_reply) {
1361         bool reply_supported = virtio_has_feature(dev->protocol_features,
1362                                           VHOST_USER_PROTOCOL_F_REPLY_ACK);
1363         if (reply_supported) {
1364             msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1365         }
1366     }
1367 
1368     ret = vhost_user_write(dev, &msg, NULL, 0);
1369     if (ret < 0) {
1370         return ret;
1371     }
1372 
1373     if (wait_for_reply) {
1374         return enforce_reply(dev, &msg);
1375     }
1376 
1377     return 0;
1378 }
1379 
1380 static int vhost_user_set_status(struct vhost_dev *dev, uint8_t status)
1381 {
1382     return vhost_user_set_u64(dev, VHOST_USER_SET_STATUS, status, false);
1383 }
1384 
1385 static int vhost_user_get_status(struct vhost_dev *dev, uint8_t *status)
1386 {
1387     uint64_t value;
1388     int ret;
1389 
1390     ret = vhost_user_get_u64(dev, VHOST_USER_GET_STATUS, &value);
1391     if (ret < 0) {
1392         return ret;
1393     }
1394     *status = value;
1395 
1396     return 0;
1397 }
1398 
1399 static int vhost_user_add_status(struct vhost_dev *dev, uint8_t status)
1400 {
1401     uint8_t s;
1402     int ret;
1403 
1404     ret = vhost_user_get_status(dev, &s);
1405     if (ret < 0) {
1406         return ret;
1407     }
1408 
1409     if ((s & status) == status) {
1410         return 0;
1411     }
1412     s |= status;
1413 
1414     return vhost_user_set_status(dev, s);
1415 }
1416 
1417 static int vhost_user_set_features(struct vhost_dev *dev,
1418                                    uint64_t features)
1419 {
1420     /*
1421      * wait for a reply if logging is enabled to make sure
1422      * backend is actually logging changes
1423      */
1424     bool log_enabled = features & (0x1ULL << VHOST_F_LOG_ALL);
1425     int ret;
1426 
1427     /*
1428      * We need to include any extra backend only feature bits that
1429      * might be needed by our device. Currently this includes the
1430      * VHOST_USER_F_PROTOCOL_FEATURES bit for enabling protocol
1431      * features.
1432      */
1433     ret = vhost_user_set_u64(dev, VHOST_USER_SET_FEATURES,
1434                               features | dev->backend_features,
1435                               log_enabled);
1436 
1437     if (virtio_has_feature(dev->protocol_features,
1438                            VHOST_USER_PROTOCOL_F_STATUS)) {
1439         if (!ret) {
1440             return vhost_user_add_status(dev, VIRTIO_CONFIG_S_FEATURES_OK);
1441         }
1442     }
1443 
1444     return ret;
1445 }
1446 
1447 static int vhost_user_set_protocol_features(struct vhost_dev *dev,
1448                                             uint64_t features)
1449 {
1450     return vhost_user_set_u64(dev, VHOST_USER_SET_PROTOCOL_FEATURES, features,
1451                               false);
1452 }
1453 
1454 static int vhost_user_set_owner(struct vhost_dev *dev)
1455 {
1456     VhostUserMsg msg = {
1457         .hdr.request = VHOST_USER_SET_OWNER,
1458         .hdr.flags = VHOST_USER_VERSION,
1459     };
1460 
1461     return vhost_user_write(dev, &msg, NULL, 0);
1462 }
1463 
1464 static int vhost_user_get_max_memslots(struct vhost_dev *dev,
1465                                        uint64_t *max_memslots)
1466 {
1467     uint64_t backend_max_memslots;
1468     int err;
1469 
1470     err = vhost_user_get_u64(dev, VHOST_USER_GET_MAX_MEM_SLOTS,
1471                              &backend_max_memslots);
1472     if (err < 0) {
1473         return err;
1474     }
1475 
1476     *max_memslots = backend_max_memslots;
1477 
1478     return 0;
1479 }
1480 
1481 static int vhost_user_reset_device(struct vhost_dev *dev)
1482 {
1483     VhostUserMsg msg = {
1484         .hdr.flags = VHOST_USER_VERSION,
1485     };
1486 
1487     msg.hdr.request = virtio_has_feature(dev->protocol_features,
1488                                          VHOST_USER_PROTOCOL_F_RESET_DEVICE)
1489         ? VHOST_USER_RESET_DEVICE
1490         : VHOST_USER_RESET_OWNER;
1491 
1492     return vhost_user_write(dev, &msg, NULL, 0);
1493 }
1494 
1495 static int vhost_user_slave_handle_config_change(struct vhost_dev *dev)
1496 {
1497     if (!dev->config_ops || !dev->config_ops->vhost_dev_config_notifier) {
1498         return -ENOSYS;
1499     }
1500 
1501     return dev->config_ops->vhost_dev_config_notifier(dev);
1502 }
1503 
1504 /*
1505  * Fetch or create the notifier for a given idx. Newly created
1506  * notifiers are added to the pointer array that tracks them.
1507  */
1508 static VhostUserHostNotifier *fetch_or_create_notifier(VhostUserState *u,
1509                                                        int idx)
1510 {
1511     VhostUserHostNotifier *n = NULL;
1512     if (idx >= u->notifiers->len) {
1513         g_ptr_array_set_size(u->notifiers, idx + 1);
1514     }
1515 
1516     n = g_ptr_array_index(u->notifiers, idx);
1517     if (!n) {
1518         /*
1519          * In case notification arrive out-of-order,
1520          * make room for current index.
1521          */
1522         g_ptr_array_remove_index(u->notifiers, idx);
1523         n = g_new0(VhostUserHostNotifier, 1);
1524         n->idx = idx;
1525         g_ptr_array_insert(u->notifiers, idx, n);
1526         trace_vhost_user_create_notifier(idx, n);
1527     }
1528 
1529     return n;
1530 }
1531 
1532 static int vhost_user_slave_handle_vring_host_notifier(struct vhost_dev *dev,
1533                                                        VhostUserVringArea *area,
1534                                                        int fd)
1535 {
1536     int queue_idx = area->u64 & VHOST_USER_VRING_IDX_MASK;
1537     size_t page_size = qemu_real_host_page_size();
1538     struct vhost_user *u = dev->opaque;
1539     VhostUserState *user = u->user;
1540     VirtIODevice *vdev = dev->vdev;
1541     VhostUserHostNotifier *n;
1542     void *addr;
1543     char *name;
1544 
1545     if (!virtio_has_feature(dev->protocol_features,
1546                             VHOST_USER_PROTOCOL_F_HOST_NOTIFIER) ||
1547         vdev == NULL || queue_idx >= virtio_get_num_queues(vdev)) {
1548         return -EINVAL;
1549     }
1550 
1551     /*
1552      * Fetch notifier and invalidate any old data before setting up
1553      * new mapped address.
1554      */
1555     n = fetch_or_create_notifier(user, queue_idx);
1556     vhost_user_host_notifier_remove(n, vdev);
1557 
1558     if (area->u64 & VHOST_USER_VRING_NOFD_MASK) {
1559         return 0;
1560     }
1561 
1562     /* Sanity check. */
1563     if (area->size != page_size) {
1564         return -EINVAL;
1565     }
1566 
1567     addr = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1568                 fd, area->offset);
1569     if (addr == MAP_FAILED) {
1570         return -EFAULT;
1571     }
1572 
1573     name = g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
1574                            user, queue_idx);
1575     if (!n->mr.ram) { /* Don't init again after suspend. */
1576         memory_region_init_ram_device_ptr(&n->mr, OBJECT(vdev), name,
1577                                           page_size, addr);
1578     } else {
1579         n->mr.ram_block->host = addr;
1580     }
1581     g_free(name);
1582 
1583     if (virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, true)) {
1584         object_unparent(OBJECT(&n->mr));
1585         munmap(addr, page_size);
1586         return -ENXIO;
1587     }
1588 
1589     n->addr = addr;
1590 
1591     return 0;
1592 }
1593 
1594 static void close_slave_channel(struct vhost_user *u)
1595 {
1596     g_source_destroy(u->slave_src);
1597     g_source_unref(u->slave_src);
1598     u->slave_src = NULL;
1599     object_unref(OBJECT(u->slave_ioc));
1600     u->slave_ioc = NULL;
1601 }
1602 
1603 static gboolean slave_read(QIOChannel *ioc, GIOCondition condition,
1604                            gpointer opaque)
1605 {
1606     struct vhost_dev *dev = opaque;
1607     struct vhost_user *u = dev->opaque;
1608     VhostUserHeader hdr = { 0, };
1609     VhostUserPayload payload = { 0, };
1610     Error *local_err = NULL;
1611     gboolean rc = G_SOURCE_CONTINUE;
1612     int ret = 0;
1613     struct iovec iov;
1614     g_autofree int *fd = NULL;
1615     size_t fdsize = 0;
1616     int i;
1617 
1618     /* Read header */
1619     iov.iov_base = &hdr;
1620     iov.iov_len = VHOST_USER_HDR_SIZE;
1621 
1622     if (qio_channel_readv_full_all(ioc, &iov, 1, &fd, &fdsize, &local_err)) {
1623         error_report_err(local_err);
1624         goto err;
1625     }
1626 
1627     if (hdr.size > VHOST_USER_PAYLOAD_SIZE) {
1628         error_report("Failed to read msg header."
1629                 " Size %d exceeds the maximum %zu.", hdr.size,
1630                 VHOST_USER_PAYLOAD_SIZE);
1631         goto err;
1632     }
1633 
1634     /* Read payload */
1635     if (qio_channel_read_all(ioc, (char *) &payload, hdr.size, &local_err)) {
1636         error_report_err(local_err);
1637         goto err;
1638     }
1639 
1640     switch (hdr.request) {
1641     case VHOST_USER_BACKEND_IOTLB_MSG:
1642         ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
1643         break;
1644     case VHOST_USER_BACKEND_CONFIG_CHANGE_MSG:
1645         ret = vhost_user_slave_handle_config_change(dev);
1646         break;
1647     case VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG:
1648         ret = vhost_user_slave_handle_vring_host_notifier(dev, &payload.area,
1649                                                           fd ? fd[0] : -1);
1650         break;
1651     default:
1652         error_report("Received unexpected msg type: %d.", hdr.request);
1653         ret = -EINVAL;
1654     }
1655 
1656     /*
1657      * REPLY_ACK feature handling. Other reply types has to be managed
1658      * directly in their request handlers.
1659      */
1660     if (hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1661         struct iovec iovec[2];
1662 
1663 
1664         hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
1665         hdr.flags |= VHOST_USER_REPLY_MASK;
1666 
1667         payload.u64 = !!ret;
1668         hdr.size = sizeof(payload.u64);
1669 
1670         iovec[0].iov_base = &hdr;
1671         iovec[0].iov_len = VHOST_USER_HDR_SIZE;
1672         iovec[1].iov_base = &payload;
1673         iovec[1].iov_len = hdr.size;
1674 
1675         if (qio_channel_writev_all(ioc, iovec, ARRAY_SIZE(iovec), &local_err)) {
1676             error_report_err(local_err);
1677             goto err;
1678         }
1679     }
1680 
1681     goto fdcleanup;
1682 
1683 err:
1684     close_slave_channel(u);
1685     rc = G_SOURCE_REMOVE;
1686 
1687 fdcleanup:
1688     if (fd) {
1689         for (i = 0; i < fdsize; i++) {
1690             close(fd[i]);
1691         }
1692     }
1693     return rc;
1694 }
1695 
1696 static int vhost_setup_slave_channel(struct vhost_dev *dev)
1697 {
1698     VhostUserMsg msg = {
1699         .hdr.request = VHOST_USER_SET_BACKEND_REQ_FD,
1700         .hdr.flags = VHOST_USER_VERSION,
1701     };
1702     struct vhost_user *u = dev->opaque;
1703     int sv[2], ret = 0;
1704     bool reply_supported = virtio_has_feature(dev->protocol_features,
1705                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1706     Error *local_err = NULL;
1707     QIOChannel *ioc;
1708 
1709     if (!virtio_has_feature(dev->protocol_features,
1710                             VHOST_USER_PROTOCOL_F_BACKEND_REQ)) {
1711         return 0;
1712     }
1713 
1714     if (qemu_socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
1715         int saved_errno = errno;
1716         error_report("socketpair() failed");
1717         return -saved_errno;
1718     }
1719 
1720     ioc = QIO_CHANNEL(qio_channel_socket_new_fd(sv[0], &local_err));
1721     if (!ioc) {
1722         error_report_err(local_err);
1723         return -ECONNREFUSED;
1724     }
1725     u->slave_ioc = ioc;
1726     u->slave_src = qio_channel_add_watch_source(u->slave_ioc,
1727                                                 G_IO_IN | G_IO_HUP,
1728                                                 slave_read, dev, NULL, NULL);
1729 
1730     if (reply_supported) {
1731         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1732     }
1733 
1734     ret = vhost_user_write(dev, &msg, &sv[1], 1);
1735     if (ret) {
1736         goto out;
1737     }
1738 
1739     if (reply_supported) {
1740         ret = process_message_reply(dev, &msg);
1741     }
1742 
1743 out:
1744     close(sv[1]);
1745     if (ret) {
1746         close_slave_channel(u);
1747     }
1748 
1749     return ret;
1750 }
1751 
1752 #ifdef CONFIG_LINUX
1753 /*
1754  * Called back from the postcopy fault thread when a fault is received on our
1755  * ufd.
1756  * TODO: This is Linux specific
1757  */
1758 static int vhost_user_postcopy_fault_handler(struct PostCopyFD *pcfd,
1759                                              void *ufd)
1760 {
1761     struct vhost_dev *dev = pcfd->data;
1762     struct vhost_user *u = dev->opaque;
1763     struct uffd_msg *msg = ufd;
1764     uint64_t faultaddr = msg->arg.pagefault.address;
1765     RAMBlock *rb = NULL;
1766     uint64_t rb_offset;
1767     int i;
1768 
1769     trace_vhost_user_postcopy_fault_handler(pcfd->idstr, faultaddr,
1770                                             dev->mem->nregions);
1771     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1772         trace_vhost_user_postcopy_fault_handler_loop(i,
1773                 u->postcopy_client_bases[i], dev->mem->regions[i].memory_size);
1774         if (faultaddr >= u->postcopy_client_bases[i]) {
1775             /* Ofset of the fault address in the vhost region */
1776             uint64_t region_offset = faultaddr - u->postcopy_client_bases[i];
1777             if (region_offset < dev->mem->regions[i].memory_size) {
1778                 rb_offset = region_offset + u->region_rb_offset[i];
1779                 trace_vhost_user_postcopy_fault_handler_found(i,
1780                         region_offset, rb_offset);
1781                 rb = u->region_rb[i];
1782                 return postcopy_request_shared_page(pcfd, rb, faultaddr,
1783                                                     rb_offset);
1784             }
1785         }
1786     }
1787     error_report("%s: Failed to find region for fault %" PRIx64,
1788                  __func__, faultaddr);
1789     return -1;
1790 }
1791 
1792 static int vhost_user_postcopy_waker(struct PostCopyFD *pcfd, RAMBlock *rb,
1793                                      uint64_t offset)
1794 {
1795     struct vhost_dev *dev = pcfd->data;
1796     struct vhost_user *u = dev->opaque;
1797     int i;
1798 
1799     trace_vhost_user_postcopy_waker(qemu_ram_get_idstr(rb), offset);
1800 
1801     if (!u) {
1802         return 0;
1803     }
1804     /* Translate the offset into an address in the clients address space */
1805     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1806         if (u->region_rb[i] == rb &&
1807             offset >= u->region_rb_offset[i] &&
1808             offset < (u->region_rb_offset[i] +
1809                       dev->mem->regions[i].memory_size)) {
1810             uint64_t client_addr = (offset - u->region_rb_offset[i]) +
1811                                    u->postcopy_client_bases[i];
1812             trace_vhost_user_postcopy_waker_found(client_addr);
1813             return postcopy_wake_shared(pcfd, client_addr, rb);
1814         }
1815     }
1816 
1817     trace_vhost_user_postcopy_waker_nomatch(qemu_ram_get_idstr(rb), offset);
1818     return 0;
1819 }
1820 #endif
1821 
1822 /*
1823  * Called at the start of an inbound postcopy on reception of the
1824  * 'advise' command.
1825  */
1826 static int vhost_user_postcopy_advise(struct vhost_dev *dev, Error **errp)
1827 {
1828 #ifdef CONFIG_LINUX
1829     struct vhost_user *u = dev->opaque;
1830     CharBackend *chr = u->user->chr;
1831     int ufd;
1832     int ret;
1833     VhostUserMsg msg = {
1834         .hdr.request = VHOST_USER_POSTCOPY_ADVISE,
1835         .hdr.flags = VHOST_USER_VERSION,
1836     };
1837 
1838     ret = vhost_user_write(dev, &msg, NULL, 0);
1839     if (ret < 0) {
1840         error_setg(errp, "Failed to send postcopy_advise to vhost");
1841         return ret;
1842     }
1843 
1844     ret = vhost_user_read(dev, &msg);
1845     if (ret < 0) {
1846         error_setg(errp, "Failed to get postcopy_advise reply from vhost");
1847         return ret;
1848     }
1849 
1850     if (msg.hdr.request != VHOST_USER_POSTCOPY_ADVISE) {
1851         error_setg(errp, "Unexpected msg type. Expected %d received %d",
1852                      VHOST_USER_POSTCOPY_ADVISE, msg.hdr.request);
1853         return -EPROTO;
1854     }
1855 
1856     if (msg.hdr.size) {
1857         error_setg(errp, "Received bad msg size.");
1858         return -EPROTO;
1859     }
1860     ufd = qemu_chr_fe_get_msgfd(chr);
1861     if (ufd < 0) {
1862         error_setg(errp, "%s: Failed to get ufd", __func__);
1863         return -EIO;
1864     }
1865     qemu_socket_set_nonblock(ufd);
1866 
1867     /* register ufd with userfault thread */
1868     u->postcopy_fd.fd = ufd;
1869     u->postcopy_fd.data = dev;
1870     u->postcopy_fd.handler = vhost_user_postcopy_fault_handler;
1871     u->postcopy_fd.waker = vhost_user_postcopy_waker;
1872     u->postcopy_fd.idstr = "vhost-user"; /* Need to find unique name */
1873     postcopy_register_shared_ufd(&u->postcopy_fd);
1874     return 0;
1875 #else
1876     error_setg(errp, "Postcopy not supported on non-Linux systems");
1877     return -ENOSYS;
1878 #endif
1879 }
1880 
1881 /*
1882  * Called at the switch to postcopy on reception of the 'listen' command.
1883  */
1884 static int vhost_user_postcopy_listen(struct vhost_dev *dev, Error **errp)
1885 {
1886     struct vhost_user *u = dev->opaque;
1887     int ret;
1888     VhostUserMsg msg = {
1889         .hdr.request = VHOST_USER_POSTCOPY_LISTEN,
1890         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
1891     };
1892     u->postcopy_listen = true;
1893 
1894     trace_vhost_user_postcopy_listen();
1895 
1896     ret = vhost_user_write(dev, &msg, NULL, 0);
1897     if (ret < 0) {
1898         error_setg(errp, "Failed to send postcopy_listen to vhost");
1899         return ret;
1900     }
1901 
1902     ret = process_message_reply(dev, &msg);
1903     if (ret) {
1904         error_setg(errp, "Failed to receive reply to postcopy_listen");
1905         return ret;
1906     }
1907 
1908     return 0;
1909 }
1910 
1911 /*
1912  * Called at the end of postcopy
1913  */
1914 static int vhost_user_postcopy_end(struct vhost_dev *dev, Error **errp)
1915 {
1916     VhostUserMsg msg = {
1917         .hdr.request = VHOST_USER_POSTCOPY_END,
1918         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
1919     };
1920     int ret;
1921     struct vhost_user *u = dev->opaque;
1922 
1923     trace_vhost_user_postcopy_end_entry();
1924 
1925     ret = vhost_user_write(dev, &msg, NULL, 0);
1926     if (ret < 0) {
1927         error_setg(errp, "Failed to send postcopy_end to vhost");
1928         return ret;
1929     }
1930 
1931     ret = process_message_reply(dev, &msg);
1932     if (ret) {
1933         error_setg(errp, "Failed to receive reply to postcopy_end");
1934         return ret;
1935     }
1936     postcopy_unregister_shared_ufd(&u->postcopy_fd);
1937     close(u->postcopy_fd.fd);
1938     u->postcopy_fd.handler = NULL;
1939 
1940     trace_vhost_user_postcopy_end_exit();
1941 
1942     return 0;
1943 }
1944 
1945 static int vhost_user_postcopy_notifier(NotifierWithReturn *notifier,
1946                                         void *opaque)
1947 {
1948     struct PostcopyNotifyData *pnd = opaque;
1949     struct vhost_user *u = container_of(notifier, struct vhost_user,
1950                                          postcopy_notifier);
1951     struct vhost_dev *dev = u->dev;
1952 
1953     switch (pnd->reason) {
1954     case POSTCOPY_NOTIFY_PROBE:
1955         if (!virtio_has_feature(dev->protocol_features,
1956                                 VHOST_USER_PROTOCOL_F_PAGEFAULT)) {
1957             /* TODO: Get the device name into this error somehow */
1958             error_setg(pnd->errp,
1959                        "vhost-user backend not capable of postcopy");
1960             return -ENOENT;
1961         }
1962         break;
1963 
1964     case POSTCOPY_NOTIFY_INBOUND_ADVISE:
1965         return vhost_user_postcopy_advise(dev, pnd->errp);
1966 
1967     case POSTCOPY_NOTIFY_INBOUND_LISTEN:
1968         return vhost_user_postcopy_listen(dev, pnd->errp);
1969 
1970     case POSTCOPY_NOTIFY_INBOUND_END:
1971         return vhost_user_postcopy_end(dev, pnd->errp);
1972 
1973     default:
1974         /* We ignore notifications we don't know */
1975         break;
1976     }
1977 
1978     return 0;
1979 }
1980 
1981 static int vhost_user_backend_init(struct vhost_dev *dev, void *opaque,
1982                                    Error **errp)
1983 {
1984     uint64_t features, ram_slots;
1985     struct vhost_user *u;
1986     VhostUserState *vus = (VhostUserState *) opaque;
1987     int err;
1988 
1989     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
1990 
1991     u = g_new0(struct vhost_user, 1);
1992     u->user = vus;
1993     u->dev = dev;
1994     dev->opaque = u;
1995 
1996     err = vhost_user_get_features(dev, &features);
1997     if (err < 0) {
1998         error_setg_errno(errp, -err, "vhost_backend_init failed");
1999         return err;
2000     }
2001 
2002     if (virtio_has_feature(features, VHOST_USER_F_PROTOCOL_FEATURES)) {
2003         bool supports_f_config = vus->supports_config ||
2004             (dev->config_ops && dev->config_ops->vhost_dev_config_notifier);
2005         uint64_t protocol_features;
2006 
2007         dev->backend_features |= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
2008 
2009         err = vhost_user_get_u64(dev, VHOST_USER_GET_PROTOCOL_FEATURES,
2010                                  &protocol_features);
2011         if (err < 0) {
2012             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2013             return -EPROTO;
2014         }
2015 
2016         /*
2017          * We will use all the protocol features we support - although
2018          * we suppress F_CONFIG if we know QEMUs internal code can not support
2019          * it.
2020          */
2021         protocol_features &= VHOST_USER_PROTOCOL_FEATURE_MASK;
2022 
2023         if (supports_f_config) {
2024             if (!virtio_has_feature(protocol_features,
2025                                     VHOST_USER_PROTOCOL_F_CONFIG)) {
2026                 error_setg(errp, "vhost-user device expecting "
2027                            "VHOST_USER_PROTOCOL_F_CONFIG but the vhost-user backend does "
2028                            "not support it.");
2029                 return -EPROTO;
2030             }
2031         } else {
2032             if (virtio_has_feature(protocol_features,
2033                                    VHOST_USER_PROTOCOL_F_CONFIG)) {
2034                 warn_report("vhost-user backend supports "
2035                             "VHOST_USER_PROTOCOL_F_CONFIG but QEMU does not.");
2036                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG);
2037             }
2038         }
2039 
2040         /* final set of protocol features */
2041         dev->protocol_features = protocol_features;
2042         err = vhost_user_set_protocol_features(dev, dev->protocol_features);
2043         if (err < 0) {
2044             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2045             return -EPROTO;
2046         }
2047 
2048         /* query the max queues we support if backend supports Multiple Queue */
2049         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)) {
2050             err = vhost_user_get_u64(dev, VHOST_USER_GET_QUEUE_NUM,
2051                                      &dev->max_queues);
2052             if (err < 0) {
2053                 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2054                 return -EPROTO;
2055             }
2056         } else {
2057             dev->max_queues = 1;
2058         }
2059 
2060         if (dev->num_queues && dev->max_queues < dev->num_queues) {
2061             error_setg(errp, "The maximum number of queues supported by the "
2062                        "backend is %" PRIu64, dev->max_queues);
2063             return -EINVAL;
2064         }
2065 
2066         if (virtio_has_feature(features, VIRTIO_F_IOMMU_PLATFORM) &&
2067                 !(virtio_has_feature(dev->protocol_features,
2068                     VHOST_USER_PROTOCOL_F_BACKEND_REQ) &&
2069                  virtio_has_feature(dev->protocol_features,
2070                     VHOST_USER_PROTOCOL_F_REPLY_ACK))) {
2071             error_setg(errp, "IOMMU support requires reply-ack and "
2072                        "slave-req protocol features.");
2073             return -EINVAL;
2074         }
2075 
2076         /* get max memory regions if backend supports configurable RAM slots */
2077         if (!virtio_has_feature(dev->protocol_features,
2078                                 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS)) {
2079             u->user->memory_slots = VHOST_MEMORY_BASELINE_NREGIONS;
2080         } else {
2081             err = vhost_user_get_max_memslots(dev, &ram_slots);
2082             if (err < 0) {
2083                 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2084                 return -EPROTO;
2085             }
2086 
2087             if (ram_slots < u->user->memory_slots) {
2088                 error_setg(errp, "The backend specified a max ram slots limit "
2089                            "of %" PRIu64", when the prior validated limit was "
2090                            "%d. This limit should never decrease.", ram_slots,
2091                            u->user->memory_slots);
2092                 return -EINVAL;
2093             }
2094 
2095             u->user->memory_slots = MIN(ram_slots, VHOST_USER_MAX_RAM_SLOTS);
2096         }
2097     }
2098 
2099     if (dev->migration_blocker == NULL &&
2100         !virtio_has_feature(dev->protocol_features,
2101                             VHOST_USER_PROTOCOL_F_LOG_SHMFD)) {
2102         error_setg(&dev->migration_blocker,
2103                    "Migration disabled: vhost-user backend lacks "
2104                    "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
2105     }
2106 
2107     if (dev->vq_index == 0) {
2108         err = vhost_setup_slave_channel(dev);
2109         if (err < 0) {
2110             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2111             return -EPROTO;
2112         }
2113     }
2114 
2115     u->postcopy_notifier.notify = vhost_user_postcopy_notifier;
2116     postcopy_add_notifier(&u->postcopy_notifier);
2117 
2118     return 0;
2119 }
2120 
2121 static int vhost_user_backend_cleanup(struct vhost_dev *dev)
2122 {
2123     struct vhost_user *u;
2124 
2125     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2126 
2127     u = dev->opaque;
2128     if (u->postcopy_notifier.notify) {
2129         postcopy_remove_notifier(&u->postcopy_notifier);
2130         u->postcopy_notifier.notify = NULL;
2131     }
2132     u->postcopy_listen = false;
2133     if (u->postcopy_fd.handler) {
2134         postcopy_unregister_shared_ufd(&u->postcopy_fd);
2135         close(u->postcopy_fd.fd);
2136         u->postcopy_fd.handler = NULL;
2137     }
2138     if (u->slave_ioc) {
2139         close_slave_channel(u);
2140     }
2141     g_free(u->region_rb);
2142     u->region_rb = NULL;
2143     g_free(u->region_rb_offset);
2144     u->region_rb_offset = NULL;
2145     u->region_rb_len = 0;
2146     g_free(u);
2147     dev->opaque = 0;
2148 
2149     return 0;
2150 }
2151 
2152 static int vhost_user_get_vq_index(struct vhost_dev *dev, int idx)
2153 {
2154     assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
2155 
2156     return idx;
2157 }
2158 
2159 static int vhost_user_memslots_limit(struct vhost_dev *dev)
2160 {
2161     struct vhost_user *u = dev->opaque;
2162 
2163     return u->user->memory_slots;
2164 }
2165 
2166 static bool vhost_user_requires_shm_log(struct vhost_dev *dev)
2167 {
2168     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2169 
2170     return virtio_has_feature(dev->protocol_features,
2171                               VHOST_USER_PROTOCOL_F_LOG_SHMFD);
2172 }
2173 
2174 static int vhost_user_migration_done(struct vhost_dev *dev, char* mac_addr)
2175 {
2176     VhostUserMsg msg = { };
2177 
2178     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2179 
2180     /* If guest supports GUEST_ANNOUNCE do nothing */
2181     if (virtio_has_feature(dev->acked_features, VIRTIO_NET_F_GUEST_ANNOUNCE)) {
2182         return 0;
2183     }
2184 
2185     /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
2186     if (virtio_has_feature(dev->protocol_features,
2187                            VHOST_USER_PROTOCOL_F_RARP)) {
2188         msg.hdr.request = VHOST_USER_SEND_RARP;
2189         msg.hdr.flags = VHOST_USER_VERSION;
2190         memcpy((char *)&msg.payload.u64, mac_addr, 6);
2191         msg.hdr.size = sizeof(msg.payload.u64);
2192 
2193         return vhost_user_write(dev, &msg, NULL, 0);
2194     }
2195     return -ENOTSUP;
2196 }
2197 
2198 static bool vhost_user_can_merge(struct vhost_dev *dev,
2199                                  uint64_t start1, uint64_t size1,
2200                                  uint64_t start2, uint64_t size2)
2201 {
2202     ram_addr_t offset;
2203     int mfd, rfd;
2204 
2205     (void)vhost_user_get_mr_data(start1, &offset, &mfd);
2206     (void)vhost_user_get_mr_data(start2, &offset, &rfd);
2207 
2208     return mfd == rfd;
2209 }
2210 
2211 static int vhost_user_net_set_mtu(struct vhost_dev *dev, uint16_t mtu)
2212 {
2213     VhostUserMsg msg;
2214     bool reply_supported = virtio_has_feature(dev->protocol_features,
2215                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2216     int ret;
2217 
2218     if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU))) {
2219         return 0;
2220     }
2221 
2222     msg.hdr.request = VHOST_USER_NET_SET_MTU;
2223     msg.payload.u64 = mtu;
2224     msg.hdr.size = sizeof(msg.payload.u64);
2225     msg.hdr.flags = VHOST_USER_VERSION;
2226     if (reply_supported) {
2227         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2228     }
2229 
2230     ret = vhost_user_write(dev, &msg, NULL, 0);
2231     if (ret < 0) {
2232         return ret;
2233     }
2234 
2235     /* If reply_ack supported, slave has to ack specified MTU is valid */
2236     if (reply_supported) {
2237         return process_message_reply(dev, &msg);
2238     }
2239 
2240     return 0;
2241 }
2242 
2243 static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
2244                                             struct vhost_iotlb_msg *imsg)
2245 {
2246     int ret;
2247     VhostUserMsg msg = {
2248         .hdr.request = VHOST_USER_IOTLB_MSG,
2249         .hdr.size = sizeof(msg.payload.iotlb),
2250         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
2251         .payload.iotlb = *imsg,
2252     };
2253 
2254     ret = vhost_user_write(dev, &msg, NULL, 0);
2255     if (ret < 0) {
2256         return ret;
2257     }
2258 
2259     return process_message_reply(dev, &msg);
2260 }
2261 
2262 
2263 static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
2264 {
2265     /* No-op as the receive channel is not dedicated to IOTLB messages. */
2266 }
2267 
2268 static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
2269                                  uint32_t config_len, Error **errp)
2270 {
2271     int ret;
2272     VhostUserMsg msg = {
2273         .hdr.request = VHOST_USER_GET_CONFIG,
2274         .hdr.flags = VHOST_USER_VERSION,
2275         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + config_len,
2276     };
2277 
2278     if (!virtio_has_feature(dev->protocol_features,
2279                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2280         error_setg(errp, "VHOST_USER_PROTOCOL_F_CONFIG not supported");
2281         return -EINVAL;
2282     }
2283 
2284     assert(config_len <= VHOST_USER_MAX_CONFIG_SIZE);
2285 
2286     msg.payload.config.offset = 0;
2287     msg.payload.config.size = config_len;
2288     ret = vhost_user_write(dev, &msg, NULL, 0);
2289     if (ret < 0) {
2290         error_setg_errno(errp, -ret, "vhost_get_config failed");
2291         return ret;
2292     }
2293 
2294     ret = vhost_user_read(dev, &msg);
2295     if (ret < 0) {
2296         error_setg_errno(errp, -ret, "vhost_get_config failed");
2297         return ret;
2298     }
2299 
2300     if (msg.hdr.request != VHOST_USER_GET_CONFIG) {
2301         error_setg(errp,
2302                    "Received unexpected msg type. Expected %d received %d",
2303                    VHOST_USER_GET_CONFIG, msg.hdr.request);
2304         return -EPROTO;
2305     }
2306 
2307     if (msg.hdr.size != VHOST_USER_CONFIG_HDR_SIZE + config_len) {
2308         error_setg(errp, "Received bad msg size.");
2309         return -EPROTO;
2310     }
2311 
2312     memcpy(config, msg.payload.config.region, config_len);
2313 
2314     return 0;
2315 }
2316 
2317 static int vhost_user_set_config(struct vhost_dev *dev, const uint8_t *data,
2318                                  uint32_t offset, uint32_t size, uint32_t flags)
2319 {
2320     int ret;
2321     uint8_t *p;
2322     bool reply_supported = virtio_has_feature(dev->protocol_features,
2323                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2324 
2325     VhostUserMsg msg = {
2326         .hdr.request = VHOST_USER_SET_CONFIG,
2327         .hdr.flags = VHOST_USER_VERSION,
2328         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + size,
2329     };
2330 
2331     if (!virtio_has_feature(dev->protocol_features,
2332                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2333         return -ENOTSUP;
2334     }
2335 
2336     if (reply_supported) {
2337         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2338     }
2339 
2340     if (size > VHOST_USER_MAX_CONFIG_SIZE) {
2341         return -EINVAL;
2342     }
2343 
2344     msg.payload.config.offset = offset,
2345     msg.payload.config.size = size,
2346     msg.payload.config.flags = flags,
2347     p = msg.payload.config.region;
2348     memcpy(p, data, size);
2349 
2350     ret = vhost_user_write(dev, &msg, NULL, 0);
2351     if (ret < 0) {
2352         return ret;
2353     }
2354 
2355     if (reply_supported) {
2356         return process_message_reply(dev, &msg);
2357     }
2358 
2359     return 0;
2360 }
2361 
2362 static int vhost_user_crypto_create_session(struct vhost_dev *dev,
2363                                             void *session_info,
2364                                             uint64_t *session_id)
2365 {
2366     int ret;
2367     bool crypto_session = virtio_has_feature(dev->protocol_features,
2368                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2369     CryptoDevBackendSymSessionInfo *sess_info = session_info;
2370     VhostUserMsg msg = {
2371         .hdr.request = VHOST_USER_CREATE_CRYPTO_SESSION,
2372         .hdr.flags = VHOST_USER_VERSION,
2373         .hdr.size = sizeof(msg.payload.session),
2374     };
2375 
2376     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2377 
2378     if (!crypto_session) {
2379         error_report("vhost-user trying to send unhandled ioctl");
2380         return -ENOTSUP;
2381     }
2382 
2383     memcpy(&msg.payload.session.session_setup_data, sess_info,
2384               sizeof(CryptoDevBackendSymSessionInfo));
2385     if (sess_info->key_len) {
2386         memcpy(&msg.payload.session.key, sess_info->cipher_key,
2387                sess_info->key_len);
2388     }
2389     if (sess_info->auth_key_len > 0) {
2390         memcpy(&msg.payload.session.auth_key, sess_info->auth_key,
2391                sess_info->auth_key_len);
2392     }
2393     ret = vhost_user_write(dev, &msg, NULL, 0);
2394     if (ret < 0) {
2395         error_report("vhost_user_write() return %d, create session failed",
2396                      ret);
2397         return ret;
2398     }
2399 
2400     ret = vhost_user_read(dev, &msg);
2401     if (ret < 0) {
2402         error_report("vhost_user_read() return %d, create session failed",
2403                      ret);
2404         return ret;
2405     }
2406 
2407     if (msg.hdr.request != VHOST_USER_CREATE_CRYPTO_SESSION) {
2408         error_report("Received unexpected msg type. Expected %d received %d",
2409                      VHOST_USER_CREATE_CRYPTO_SESSION, msg.hdr.request);
2410         return -EPROTO;
2411     }
2412 
2413     if (msg.hdr.size != sizeof(msg.payload.session)) {
2414         error_report("Received bad msg size.");
2415         return -EPROTO;
2416     }
2417 
2418     if (msg.payload.session.session_id < 0) {
2419         error_report("Bad session id: %" PRId64 "",
2420                               msg.payload.session.session_id);
2421         return -EINVAL;
2422     }
2423     *session_id = msg.payload.session.session_id;
2424 
2425     return 0;
2426 }
2427 
2428 static int
2429 vhost_user_crypto_close_session(struct vhost_dev *dev, uint64_t session_id)
2430 {
2431     int ret;
2432     bool crypto_session = virtio_has_feature(dev->protocol_features,
2433                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2434     VhostUserMsg msg = {
2435         .hdr.request = VHOST_USER_CLOSE_CRYPTO_SESSION,
2436         .hdr.flags = VHOST_USER_VERSION,
2437         .hdr.size = sizeof(msg.payload.u64),
2438     };
2439     msg.payload.u64 = session_id;
2440 
2441     if (!crypto_session) {
2442         error_report("vhost-user trying to send unhandled ioctl");
2443         return -ENOTSUP;
2444     }
2445 
2446     ret = vhost_user_write(dev, &msg, NULL, 0);
2447     if (ret < 0) {
2448         error_report("vhost_user_write() return %d, close session failed",
2449                      ret);
2450         return ret;
2451     }
2452 
2453     return 0;
2454 }
2455 
2456 static bool vhost_user_mem_section_filter(struct vhost_dev *dev,
2457                                           MemoryRegionSection *section)
2458 {
2459     return memory_region_get_fd(section->mr) >= 0;
2460 }
2461 
2462 static int vhost_user_get_inflight_fd(struct vhost_dev *dev,
2463                                       uint16_t queue_size,
2464                                       struct vhost_inflight *inflight)
2465 {
2466     void *addr;
2467     int fd;
2468     int ret;
2469     struct vhost_user *u = dev->opaque;
2470     CharBackend *chr = u->user->chr;
2471     VhostUserMsg msg = {
2472         .hdr.request = VHOST_USER_GET_INFLIGHT_FD,
2473         .hdr.flags = VHOST_USER_VERSION,
2474         .payload.inflight.num_queues = dev->nvqs,
2475         .payload.inflight.queue_size = queue_size,
2476         .hdr.size = sizeof(msg.payload.inflight),
2477     };
2478 
2479     if (!virtio_has_feature(dev->protocol_features,
2480                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2481         return 0;
2482     }
2483 
2484     ret = vhost_user_write(dev, &msg, NULL, 0);
2485     if (ret < 0) {
2486         return ret;
2487     }
2488 
2489     ret = vhost_user_read(dev, &msg);
2490     if (ret < 0) {
2491         return ret;
2492     }
2493 
2494     if (msg.hdr.request != VHOST_USER_GET_INFLIGHT_FD) {
2495         error_report("Received unexpected msg type. "
2496                      "Expected %d received %d",
2497                      VHOST_USER_GET_INFLIGHT_FD, msg.hdr.request);
2498         return -EPROTO;
2499     }
2500 
2501     if (msg.hdr.size != sizeof(msg.payload.inflight)) {
2502         error_report("Received bad msg size.");
2503         return -EPROTO;
2504     }
2505 
2506     if (!msg.payload.inflight.mmap_size) {
2507         return 0;
2508     }
2509 
2510     fd = qemu_chr_fe_get_msgfd(chr);
2511     if (fd < 0) {
2512         error_report("Failed to get mem fd");
2513         return -EIO;
2514     }
2515 
2516     addr = mmap(0, msg.payload.inflight.mmap_size, PROT_READ | PROT_WRITE,
2517                 MAP_SHARED, fd, msg.payload.inflight.mmap_offset);
2518 
2519     if (addr == MAP_FAILED) {
2520         error_report("Failed to mmap mem fd");
2521         close(fd);
2522         return -EFAULT;
2523     }
2524 
2525     inflight->addr = addr;
2526     inflight->fd = fd;
2527     inflight->size = msg.payload.inflight.mmap_size;
2528     inflight->offset = msg.payload.inflight.mmap_offset;
2529     inflight->queue_size = queue_size;
2530 
2531     return 0;
2532 }
2533 
2534 static int vhost_user_set_inflight_fd(struct vhost_dev *dev,
2535                                       struct vhost_inflight *inflight)
2536 {
2537     VhostUserMsg msg = {
2538         .hdr.request = VHOST_USER_SET_INFLIGHT_FD,
2539         .hdr.flags = VHOST_USER_VERSION,
2540         .payload.inflight.mmap_size = inflight->size,
2541         .payload.inflight.mmap_offset = inflight->offset,
2542         .payload.inflight.num_queues = dev->nvqs,
2543         .payload.inflight.queue_size = inflight->queue_size,
2544         .hdr.size = sizeof(msg.payload.inflight),
2545     };
2546 
2547     if (!virtio_has_feature(dev->protocol_features,
2548                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2549         return 0;
2550     }
2551 
2552     return vhost_user_write(dev, &msg, &inflight->fd, 1);
2553 }
2554 
2555 static void vhost_user_state_destroy(gpointer data)
2556 {
2557     VhostUserHostNotifier *n = (VhostUserHostNotifier *) data;
2558     if (n) {
2559         vhost_user_host_notifier_remove(n, NULL);
2560         object_unparent(OBJECT(&n->mr));
2561         /*
2562          * We can't free until vhost_user_host_notifier_remove has
2563          * done it's thing so schedule the free with RCU.
2564          */
2565         g_free_rcu(n, rcu);
2566     }
2567 }
2568 
2569 bool vhost_user_init(VhostUserState *user, CharBackend *chr, Error **errp)
2570 {
2571     if (user->chr) {
2572         error_setg(errp, "Cannot initialize vhost-user state");
2573         return false;
2574     }
2575     user->chr = chr;
2576     user->memory_slots = 0;
2577     user->notifiers = g_ptr_array_new_full(VIRTIO_QUEUE_MAX / 4,
2578                                            &vhost_user_state_destroy);
2579     return true;
2580 }
2581 
2582 void vhost_user_cleanup(VhostUserState *user)
2583 {
2584     if (!user->chr) {
2585         return;
2586     }
2587     memory_region_transaction_begin();
2588     user->notifiers = (GPtrArray *) g_ptr_array_free(user->notifiers, true);
2589     memory_region_transaction_commit();
2590     user->chr = NULL;
2591 }
2592 
2593 
2594 typedef struct {
2595     vu_async_close_fn cb;
2596     DeviceState *dev;
2597     CharBackend *cd;
2598     struct vhost_dev *vhost;
2599 } VhostAsyncCallback;
2600 
2601 static void vhost_user_async_close_bh(void *opaque)
2602 {
2603     VhostAsyncCallback *data = opaque;
2604     struct vhost_dev *vhost = data->vhost;
2605 
2606     /*
2607      * If the vhost_dev has been cleared in the meantime there is
2608      * nothing left to do as some other path has completed the
2609      * cleanup.
2610      */
2611     if (vhost->vdev) {
2612         data->cb(data->dev);
2613     }
2614 
2615     g_free(data);
2616 }
2617 
2618 /*
2619  * We only schedule the work if the machine is running. If suspended
2620  * we want to keep all the in-flight data as is for migration
2621  * purposes.
2622  */
2623 void vhost_user_async_close(DeviceState *d,
2624                             CharBackend *chardev, struct vhost_dev *vhost,
2625                             vu_async_close_fn cb)
2626 {
2627     if (!runstate_check(RUN_STATE_SHUTDOWN)) {
2628         /*
2629          * A close event may happen during a read/write, but vhost
2630          * code assumes the vhost_dev remains setup, so delay the
2631          * stop & clear.
2632          */
2633         AioContext *ctx = qemu_get_current_aio_context();
2634         VhostAsyncCallback *data = g_new0(VhostAsyncCallback, 1);
2635 
2636         /* Save data for the callback */
2637         data->cb = cb;
2638         data->dev = d;
2639         data->cd = chardev;
2640         data->vhost = vhost;
2641 
2642         /* Disable any further notifications on the chardev */
2643         qemu_chr_fe_set_handlers(chardev,
2644                                  NULL, NULL, NULL, NULL, NULL, NULL,
2645                                  false);
2646 
2647         aio_bh_schedule_oneshot(ctx, vhost_user_async_close_bh, data);
2648 
2649         /*
2650          * Move vhost device to the stopped state. The vhost-user device
2651          * will be clean up and disconnected in BH. This can be useful in
2652          * the vhost migration code. If disconnect was caught there is an
2653          * option for the general vhost code to get the dev state without
2654          * knowing its type (in this case vhost-user).
2655          *
2656          * Note if the vhost device is fully cleared by the time we
2657          * execute the bottom half we won't continue with the cleanup.
2658          */
2659         vhost->started = false;
2660     }
2661 }
2662 
2663 static int vhost_user_dev_start(struct vhost_dev *dev, bool started)
2664 {
2665     if (!virtio_has_feature(dev->protocol_features,
2666                             VHOST_USER_PROTOCOL_F_STATUS)) {
2667         return 0;
2668     }
2669 
2670     /* Set device status only for last queue pair */
2671     if (dev->vq_index + dev->nvqs != dev->vq_index_end) {
2672         return 0;
2673     }
2674 
2675     if (started) {
2676         return vhost_user_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE |
2677                                           VIRTIO_CONFIG_S_DRIVER |
2678                                           VIRTIO_CONFIG_S_DRIVER_OK);
2679     } else {
2680         return vhost_user_set_status(dev, 0);
2681     }
2682 }
2683 
2684 const VhostOps user_ops = {
2685         .backend_type = VHOST_BACKEND_TYPE_USER,
2686         .vhost_backend_init = vhost_user_backend_init,
2687         .vhost_backend_cleanup = vhost_user_backend_cleanup,
2688         .vhost_backend_memslots_limit = vhost_user_memslots_limit,
2689         .vhost_set_log_base = vhost_user_set_log_base,
2690         .vhost_set_mem_table = vhost_user_set_mem_table,
2691         .vhost_set_vring_addr = vhost_user_set_vring_addr,
2692         .vhost_set_vring_endian = vhost_user_set_vring_endian,
2693         .vhost_set_vring_num = vhost_user_set_vring_num,
2694         .vhost_set_vring_base = vhost_user_set_vring_base,
2695         .vhost_get_vring_base = vhost_user_get_vring_base,
2696         .vhost_set_vring_kick = vhost_user_set_vring_kick,
2697         .vhost_set_vring_call = vhost_user_set_vring_call,
2698         .vhost_set_vring_err = vhost_user_set_vring_err,
2699         .vhost_set_features = vhost_user_set_features,
2700         .vhost_get_features = vhost_user_get_features,
2701         .vhost_set_owner = vhost_user_set_owner,
2702         .vhost_reset_device = vhost_user_reset_device,
2703         .vhost_get_vq_index = vhost_user_get_vq_index,
2704         .vhost_set_vring_enable = vhost_user_set_vring_enable,
2705         .vhost_requires_shm_log = vhost_user_requires_shm_log,
2706         .vhost_migration_done = vhost_user_migration_done,
2707         .vhost_backend_can_merge = vhost_user_can_merge,
2708         .vhost_net_set_mtu = vhost_user_net_set_mtu,
2709         .vhost_set_iotlb_callback = vhost_user_set_iotlb_callback,
2710         .vhost_send_device_iotlb_msg = vhost_user_send_device_iotlb_msg,
2711         .vhost_get_config = vhost_user_get_config,
2712         .vhost_set_config = vhost_user_set_config,
2713         .vhost_crypto_create_session = vhost_user_crypto_create_session,
2714         .vhost_crypto_close_session = vhost_user_crypto_close_session,
2715         .vhost_backend_mem_section_filter = vhost_user_mem_section_filter,
2716         .vhost_get_inflight_fd = vhost_user_get_inflight_fd,
2717         .vhost_set_inflight_fd = vhost_user_set_inflight_fd,
2718         .vhost_dev_start = vhost_user_dev_start,
2719 };
2720