xref: /openbmc/qemu/hw/virtio/vhost-user.c (revision c85cad81)
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/virtio-crypto.h"
15 #include "hw/virtio/vhost-user.h"
16 #include "hw/virtio/vhost-backend.h"
17 #include "hw/virtio/virtio.h"
18 #include "hw/virtio/virtio-net.h"
19 #include "chardev/char-fe.h"
20 #include "io/channel-socket.h"
21 #include "sysemu/kvm.h"
22 #include "qemu/error-report.h"
23 #include "qemu/main-loop.h"
24 #include "qemu/sockets.h"
25 #include "sysemu/runstate.h"
26 #include "sysemu/cryptodev.h"
27 #include "migration/migration.h"
28 #include "migration/postcopy-ram.h"
29 #include "trace.h"
30 #include "exec/ramblock.h"
31 
32 #include <sys/ioctl.h>
33 #include <sys/socket.h>
34 #include <sys/un.h>
35 
36 #include "standard-headers/linux/vhost_types.h"
37 
38 #ifdef CONFIG_LINUX
39 #include <linux/userfaultfd.h>
40 #endif
41 
42 #define VHOST_MEMORY_BASELINE_NREGIONS    8
43 #define VHOST_USER_F_PROTOCOL_FEATURES 30
44 #define VHOST_USER_BACKEND_MAX_FDS     8
45 
46 #if defined(TARGET_PPC) || defined(TARGET_PPC64)
47 #include "hw/ppc/spapr.h"
48 #define VHOST_USER_MAX_RAM_SLOTS SPAPR_MAX_RAM_SLOTS
49 
50 #else
51 #define VHOST_USER_MAX_RAM_SLOTS 512
52 #endif
53 
54 /*
55  * Maximum size of virtio device config space
56  */
57 #define VHOST_USER_MAX_CONFIG_SIZE 256
58 
59 enum VhostUserProtocolFeature {
60     VHOST_USER_PROTOCOL_F_MQ = 0,
61     VHOST_USER_PROTOCOL_F_LOG_SHMFD = 1,
62     VHOST_USER_PROTOCOL_F_RARP = 2,
63     VHOST_USER_PROTOCOL_F_REPLY_ACK = 3,
64     VHOST_USER_PROTOCOL_F_NET_MTU = 4,
65     VHOST_USER_PROTOCOL_F_BACKEND_REQ = 5,
66     VHOST_USER_PROTOCOL_F_CROSS_ENDIAN = 6,
67     VHOST_USER_PROTOCOL_F_CRYPTO_SESSION = 7,
68     VHOST_USER_PROTOCOL_F_PAGEFAULT = 8,
69     VHOST_USER_PROTOCOL_F_CONFIG = 9,
70     VHOST_USER_PROTOCOL_F_BACKEND_SEND_FD = 10,
71     VHOST_USER_PROTOCOL_F_HOST_NOTIFIER = 11,
72     VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD = 12,
73     VHOST_USER_PROTOCOL_F_RESET_DEVICE = 13,
74     /* Feature 14 reserved for VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS. */
75     VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS = 15,
76     VHOST_USER_PROTOCOL_F_STATUS = 16,
77     VHOST_USER_PROTOCOL_F_MAX
78 };
79 
80 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
81 
82 typedef enum VhostUserRequest {
83     VHOST_USER_NONE = 0,
84     VHOST_USER_GET_FEATURES = 1,
85     VHOST_USER_SET_FEATURES = 2,
86     VHOST_USER_SET_OWNER = 3,
87     VHOST_USER_RESET_OWNER = 4,
88     VHOST_USER_SET_MEM_TABLE = 5,
89     VHOST_USER_SET_LOG_BASE = 6,
90     VHOST_USER_SET_LOG_FD = 7,
91     VHOST_USER_SET_VRING_NUM = 8,
92     VHOST_USER_SET_VRING_ADDR = 9,
93     VHOST_USER_SET_VRING_BASE = 10,
94     VHOST_USER_GET_VRING_BASE = 11,
95     VHOST_USER_SET_VRING_KICK = 12,
96     VHOST_USER_SET_VRING_CALL = 13,
97     VHOST_USER_SET_VRING_ERR = 14,
98     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
99     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
100     VHOST_USER_GET_QUEUE_NUM = 17,
101     VHOST_USER_SET_VRING_ENABLE = 18,
102     VHOST_USER_SEND_RARP = 19,
103     VHOST_USER_NET_SET_MTU = 20,
104     VHOST_USER_SET_BACKEND_REQ_FD = 21,
105     VHOST_USER_IOTLB_MSG = 22,
106     VHOST_USER_SET_VRING_ENDIAN = 23,
107     VHOST_USER_GET_CONFIG = 24,
108     VHOST_USER_SET_CONFIG = 25,
109     VHOST_USER_CREATE_CRYPTO_SESSION = 26,
110     VHOST_USER_CLOSE_CRYPTO_SESSION = 27,
111     VHOST_USER_POSTCOPY_ADVISE  = 28,
112     VHOST_USER_POSTCOPY_LISTEN  = 29,
113     VHOST_USER_POSTCOPY_END     = 30,
114     VHOST_USER_GET_INFLIGHT_FD = 31,
115     VHOST_USER_SET_INFLIGHT_FD = 32,
116     VHOST_USER_GPU_SET_SOCKET = 33,
117     VHOST_USER_RESET_DEVICE = 34,
118     /* Message number 35 reserved for VHOST_USER_VRING_KICK. */
119     VHOST_USER_GET_MAX_MEM_SLOTS = 36,
120     VHOST_USER_ADD_MEM_REG = 37,
121     VHOST_USER_REM_MEM_REG = 38,
122     VHOST_USER_SET_STATUS = 39,
123     VHOST_USER_GET_STATUS = 40,
124     VHOST_USER_MAX
125 } VhostUserRequest;
126 
127 typedef enum VhostUserSlaveRequest {
128     VHOST_USER_BACKEND_NONE = 0,
129     VHOST_USER_BACKEND_IOTLB_MSG = 1,
130     VHOST_USER_BACKEND_CONFIG_CHANGE_MSG = 2,
131     VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG = 3,
132     VHOST_USER_BACKEND_MAX
133 }  VhostUserSlaveRequest;
134 
135 typedef struct VhostUserMemoryRegion {
136     uint64_t guest_phys_addr;
137     uint64_t memory_size;
138     uint64_t userspace_addr;
139     uint64_t mmap_offset;
140 } VhostUserMemoryRegion;
141 
142 typedef struct VhostUserMemory {
143     uint32_t nregions;
144     uint32_t padding;
145     VhostUserMemoryRegion regions[VHOST_MEMORY_BASELINE_NREGIONS];
146 } VhostUserMemory;
147 
148 typedef struct VhostUserMemRegMsg {
149     uint64_t padding;
150     VhostUserMemoryRegion region;
151 } VhostUserMemRegMsg;
152 
153 typedef struct VhostUserLog {
154     uint64_t mmap_size;
155     uint64_t mmap_offset;
156 } VhostUserLog;
157 
158 typedef struct VhostUserConfig {
159     uint32_t offset;
160     uint32_t size;
161     uint32_t flags;
162     uint8_t region[VHOST_USER_MAX_CONFIG_SIZE];
163 } VhostUserConfig;
164 
165 #define VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN    512
166 #define VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN  64
167 #define VHOST_CRYPTO_ASYM_MAX_KEY_LEN  1024
168 
169 typedef struct VhostUserCryptoSession {
170     uint64_t op_code;
171     union {
172         struct {
173             CryptoDevBackendSymSessionInfo session_setup_data;
174             uint8_t key[VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN];
175             uint8_t auth_key[VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN];
176         } sym;
177         struct {
178             CryptoDevBackendAsymSessionInfo session_setup_data;
179             uint8_t key[VHOST_CRYPTO_ASYM_MAX_KEY_LEN];
180         } asym;
181     } u;
182 
183     /* session id for success, -1 on errors */
184     int64_t session_id;
185 } VhostUserCryptoSession;
186 
187 static VhostUserConfig c __attribute__ ((unused));
188 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
189                                    + sizeof(c.size) \
190                                    + sizeof(c.flags))
191 
192 typedef struct VhostUserVringArea {
193     uint64_t u64;
194     uint64_t size;
195     uint64_t offset;
196 } VhostUserVringArea;
197 
198 typedef struct VhostUserInflight {
199     uint64_t mmap_size;
200     uint64_t mmap_offset;
201     uint16_t num_queues;
202     uint16_t queue_size;
203 } VhostUserInflight;
204 
205 typedef struct {
206     VhostUserRequest request;
207 
208 #define VHOST_USER_VERSION_MASK     (0x3)
209 #define VHOST_USER_REPLY_MASK       (0x1 << 2)
210 #define VHOST_USER_NEED_REPLY_MASK  (0x1 << 3)
211     uint32_t flags;
212     uint32_t size; /* the following payload size */
213 } QEMU_PACKED VhostUserHeader;
214 
215 typedef union {
216 #define VHOST_USER_VRING_IDX_MASK   (0xff)
217 #define VHOST_USER_VRING_NOFD_MASK  (0x1 << 8)
218         uint64_t u64;
219         struct vhost_vring_state state;
220         struct vhost_vring_addr addr;
221         VhostUserMemory memory;
222         VhostUserMemRegMsg mem_reg;
223         VhostUserLog log;
224         struct vhost_iotlb_msg iotlb;
225         VhostUserConfig config;
226         VhostUserCryptoSession session;
227         VhostUserVringArea area;
228         VhostUserInflight inflight;
229 } VhostUserPayload;
230 
231 typedef struct VhostUserMsg {
232     VhostUserHeader hdr;
233     VhostUserPayload payload;
234 } QEMU_PACKED VhostUserMsg;
235 
236 static VhostUserMsg m __attribute__ ((unused));
237 #define VHOST_USER_HDR_SIZE (sizeof(VhostUserHeader))
238 
239 #define VHOST_USER_PAYLOAD_SIZE (sizeof(VhostUserPayload))
240 
241 /* The version of the protocol we support */
242 #define VHOST_USER_VERSION    (0x1)
243 
244 struct vhost_user {
245     struct vhost_dev *dev;
246     /* Shared between vhost devs of the same virtio device */
247     VhostUserState *user;
248     QIOChannel *slave_ioc;
249     GSource *slave_src;
250     NotifierWithReturn postcopy_notifier;
251     struct PostCopyFD  postcopy_fd;
252     uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
253     /* Length of the region_rb and region_rb_offset arrays */
254     size_t             region_rb_len;
255     /* RAMBlock associated with a given region */
256     RAMBlock         **region_rb;
257     /*
258      * The offset from the start of the RAMBlock to the start of the
259      * vhost region.
260      */
261     ram_addr_t        *region_rb_offset;
262 
263     /* True once we've entered postcopy_listen */
264     bool               postcopy_listen;
265 
266     /* Our current regions */
267     int num_shadow_regions;
268     struct vhost_memory_region shadow_regions[VHOST_USER_MAX_RAM_SLOTS];
269 };
270 
271 struct scrub_regions {
272     struct vhost_memory_region *region;
273     int reg_idx;
274     int fd_idx;
275 };
276 
277 static bool ioeventfd_enabled(void)
278 {
279     return !kvm_enabled() || kvm_eventfds_enabled();
280 }
281 
282 static int vhost_user_read_header(struct vhost_dev *dev, VhostUserMsg *msg)
283 {
284     struct vhost_user *u = dev->opaque;
285     CharBackend *chr = u->user->chr;
286     uint8_t *p = (uint8_t *) msg;
287     int r, size = VHOST_USER_HDR_SIZE;
288 
289     r = qemu_chr_fe_read_all(chr, p, size);
290     if (r != size) {
291         int saved_errno = errno;
292         error_report("Failed to read msg header. Read %d instead of %d."
293                      " Original request %d.", r, size, msg->hdr.request);
294         return r < 0 ? -saved_errno : -EIO;
295     }
296 
297     /* validate received flags */
298     if (msg->hdr.flags != (VHOST_USER_REPLY_MASK | VHOST_USER_VERSION)) {
299         error_report("Failed to read msg header."
300                 " Flags 0x%x instead of 0x%x.", msg->hdr.flags,
301                 VHOST_USER_REPLY_MASK | VHOST_USER_VERSION);
302         return -EPROTO;
303     }
304 
305     trace_vhost_user_read(msg->hdr.request, msg->hdr.flags);
306 
307     return 0;
308 }
309 
310 static int vhost_user_read(struct vhost_dev *dev, VhostUserMsg *msg)
311 {
312     struct vhost_user *u = dev->opaque;
313     CharBackend *chr = u->user->chr;
314     uint8_t *p = (uint8_t *) msg;
315     int r, size;
316 
317     r = vhost_user_read_header(dev, msg);
318     if (r < 0) {
319         return r;
320     }
321 
322     /* validate message size is sane */
323     if (msg->hdr.size > VHOST_USER_PAYLOAD_SIZE) {
324         error_report("Failed to read msg header."
325                 " Size %d exceeds the maximum %zu.", msg->hdr.size,
326                 VHOST_USER_PAYLOAD_SIZE);
327         return -EPROTO;
328     }
329 
330     if (msg->hdr.size) {
331         p += VHOST_USER_HDR_SIZE;
332         size = msg->hdr.size;
333         r = qemu_chr_fe_read_all(chr, p, size);
334         if (r != size) {
335             int saved_errno = errno;
336             error_report("Failed to read msg payload."
337                          " Read %d instead of %d.", r, msg->hdr.size);
338             return r < 0 ? -saved_errno : -EIO;
339         }
340     }
341 
342     return 0;
343 }
344 
345 static int process_message_reply(struct vhost_dev *dev,
346                                  const VhostUserMsg *msg)
347 {
348     int ret;
349     VhostUserMsg msg_reply;
350 
351     if ((msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) == 0) {
352         return 0;
353     }
354 
355     ret = vhost_user_read(dev, &msg_reply);
356     if (ret < 0) {
357         return ret;
358     }
359 
360     if (msg_reply.hdr.request != msg->hdr.request) {
361         error_report("Received unexpected msg type. "
362                      "Expected %d received %d",
363                      msg->hdr.request, msg_reply.hdr.request);
364         return -EPROTO;
365     }
366 
367     return msg_reply.payload.u64 ? -EIO : 0;
368 }
369 
370 static bool vhost_user_one_time_request(VhostUserRequest request)
371 {
372     switch (request) {
373     case VHOST_USER_SET_OWNER:
374     case VHOST_USER_RESET_OWNER:
375     case VHOST_USER_SET_MEM_TABLE:
376     case VHOST_USER_GET_QUEUE_NUM:
377     case VHOST_USER_NET_SET_MTU:
378     case VHOST_USER_ADD_MEM_REG:
379     case VHOST_USER_REM_MEM_REG:
380         return true;
381     default:
382         return false;
383     }
384 }
385 
386 /* most non-init callers ignore the error */
387 static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
388                             int *fds, int fd_num)
389 {
390     struct vhost_user *u = dev->opaque;
391     CharBackend *chr = u->user->chr;
392     int ret, size = VHOST_USER_HDR_SIZE + msg->hdr.size;
393 
394     /*
395      * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
396      * we just need send it once in the first time. For later such
397      * request, we just ignore it.
398      */
399     if (vhost_user_one_time_request(msg->hdr.request) && dev->vq_index != 0) {
400         msg->hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
401         return 0;
402     }
403 
404     if (qemu_chr_fe_set_msgfds(chr, fds, fd_num) < 0) {
405         error_report("Failed to set msg fds.");
406         return -EINVAL;
407     }
408 
409     ret = qemu_chr_fe_write_all(chr, (const uint8_t *) msg, size);
410     if (ret != size) {
411         int saved_errno = errno;
412         error_report("Failed to write msg."
413                      " Wrote %d instead of %d.", ret, size);
414         return ret < 0 ? -saved_errno : -EIO;
415     }
416 
417     trace_vhost_user_write(msg->hdr.request, msg->hdr.flags);
418 
419     return 0;
420 }
421 
422 int vhost_user_gpu_set_socket(struct vhost_dev *dev, int fd)
423 {
424     VhostUserMsg msg = {
425         .hdr.request = VHOST_USER_GPU_SET_SOCKET,
426         .hdr.flags = VHOST_USER_VERSION,
427     };
428 
429     return vhost_user_write(dev, &msg, &fd, 1);
430 }
431 
432 static int vhost_user_set_log_base(struct vhost_dev *dev, uint64_t base,
433                                    struct vhost_log *log)
434 {
435     int fds[VHOST_USER_MAX_RAM_SLOTS];
436     size_t fd_num = 0;
437     bool shmfd = virtio_has_feature(dev->protocol_features,
438                                     VHOST_USER_PROTOCOL_F_LOG_SHMFD);
439     int ret;
440     VhostUserMsg msg = {
441         .hdr.request = VHOST_USER_SET_LOG_BASE,
442         .hdr.flags = VHOST_USER_VERSION,
443         .payload.log.mmap_size = log->size * sizeof(*(log->log)),
444         .payload.log.mmap_offset = 0,
445         .hdr.size = sizeof(msg.payload.log),
446     };
447 
448     /* Send only once with first queue pair */
449     if (dev->vq_index != 0) {
450         return 0;
451     }
452 
453     if (shmfd && log->fd != -1) {
454         fds[fd_num++] = log->fd;
455     }
456 
457     ret = vhost_user_write(dev, &msg, fds, fd_num);
458     if (ret < 0) {
459         return ret;
460     }
461 
462     if (shmfd) {
463         msg.hdr.size = 0;
464         ret = vhost_user_read(dev, &msg);
465         if (ret < 0) {
466             return ret;
467         }
468 
469         if (msg.hdr.request != VHOST_USER_SET_LOG_BASE) {
470             error_report("Received unexpected msg type. "
471                          "Expected %d received %d",
472                          VHOST_USER_SET_LOG_BASE, msg.hdr.request);
473             return -EPROTO;
474         }
475     }
476 
477     return 0;
478 }
479 
480 static MemoryRegion *vhost_user_get_mr_data(uint64_t addr, ram_addr_t *offset,
481                                             int *fd)
482 {
483     MemoryRegion *mr;
484 
485     assert((uintptr_t)addr == addr);
486     mr = memory_region_from_host((void *)(uintptr_t)addr, offset);
487     *fd = memory_region_get_fd(mr);
488     *offset += mr->ram_block->fd_offset;
489 
490     return mr;
491 }
492 
493 static void vhost_user_fill_msg_region(VhostUserMemoryRegion *dst,
494                                        struct vhost_memory_region *src,
495                                        uint64_t mmap_offset)
496 {
497     assert(src != NULL && dst != NULL);
498     dst->userspace_addr = src->userspace_addr;
499     dst->memory_size = src->memory_size;
500     dst->guest_phys_addr = src->guest_phys_addr;
501     dst->mmap_offset = mmap_offset;
502 }
503 
504 static int vhost_user_fill_set_mem_table_msg(struct vhost_user *u,
505                                              struct vhost_dev *dev,
506                                              VhostUserMsg *msg,
507                                              int *fds, size_t *fd_num,
508                                              bool track_ramblocks)
509 {
510     int i, fd;
511     ram_addr_t offset;
512     MemoryRegion *mr;
513     struct vhost_memory_region *reg;
514     VhostUserMemoryRegion region_buffer;
515 
516     msg->hdr.request = VHOST_USER_SET_MEM_TABLE;
517 
518     for (i = 0; i < dev->mem->nregions; ++i) {
519         reg = dev->mem->regions + i;
520 
521         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
522         if (fd > 0) {
523             if (track_ramblocks) {
524                 assert(*fd_num < VHOST_MEMORY_BASELINE_NREGIONS);
525                 trace_vhost_user_set_mem_table_withfd(*fd_num, mr->name,
526                                                       reg->memory_size,
527                                                       reg->guest_phys_addr,
528                                                       reg->userspace_addr,
529                                                       offset);
530                 u->region_rb_offset[i] = offset;
531                 u->region_rb[i] = mr->ram_block;
532             } else if (*fd_num == VHOST_MEMORY_BASELINE_NREGIONS) {
533                 error_report("Failed preparing vhost-user memory table msg");
534                 return -ENOBUFS;
535             }
536             vhost_user_fill_msg_region(&region_buffer, reg, offset);
537             msg->payload.memory.regions[*fd_num] = region_buffer;
538             fds[(*fd_num)++] = fd;
539         } else if (track_ramblocks) {
540             u->region_rb_offset[i] = 0;
541             u->region_rb[i] = NULL;
542         }
543     }
544 
545     msg->payload.memory.nregions = *fd_num;
546 
547     if (!*fd_num) {
548         error_report("Failed initializing vhost-user memory map, "
549                      "consider using -object memory-backend-file share=on");
550         return -EINVAL;
551     }
552 
553     msg->hdr.size = sizeof(msg->payload.memory.nregions);
554     msg->hdr.size += sizeof(msg->payload.memory.padding);
555     msg->hdr.size += *fd_num * sizeof(VhostUserMemoryRegion);
556 
557     return 0;
558 }
559 
560 static inline bool reg_equal(struct vhost_memory_region *shadow_reg,
561                              struct vhost_memory_region *vdev_reg)
562 {
563     return shadow_reg->guest_phys_addr == vdev_reg->guest_phys_addr &&
564         shadow_reg->userspace_addr == vdev_reg->userspace_addr &&
565         shadow_reg->memory_size == vdev_reg->memory_size;
566 }
567 
568 static void scrub_shadow_regions(struct vhost_dev *dev,
569                                  struct scrub_regions *add_reg,
570                                  int *nr_add_reg,
571                                  struct scrub_regions *rem_reg,
572                                  int *nr_rem_reg, uint64_t *shadow_pcb,
573                                  bool track_ramblocks)
574 {
575     struct vhost_user *u = dev->opaque;
576     bool found[VHOST_USER_MAX_RAM_SLOTS] = {};
577     struct vhost_memory_region *reg, *shadow_reg;
578     int i, j, fd, add_idx = 0, rm_idx = 0, fd_num = 0;
579     ram_addr_t offset;
580     MemoryRegion *mr;
581     bool matching;
582 
583     /*
584      * Find memory regions present in our shadow state which are not in
585      * the device's current memory state.
586      *
587      * Mark regions in both the shadow and device state as "found".
588      */
589     for (i = 0; i < u->num_shadow_regions; i++) {
590         shadow_reg = &u->shadow_regions[i];
591         matching = false;
592 
593         for (j = 0; j < dev->mem->nregions; j++) {
594             reg = &dev->mem->regions[j];
595 
596             mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
597 
598             if (reg_equal(shadow_reg, reg)) {
599                 matching = true;
600                 found[j] = true;
601                 if (track_ramblocks) {
602                     /*
603                      * Reset postcopy client bases, region_rb, and
604                      * region_rb_offset in case regions are removed.
605                      */
606                     if (fd > 0) {
607                         u->region_rb_offset[j] = offset;
608                         u->region_rb[j] = mr->ram_block;
609                         shadow_pcb[j] = u->postcopy_client_bases[i];
610                     } else {
611                         u->region_rb_offset[j] = 0;
612                         u->region_rb[j] = NULL;
613                     }
614                 }
615                 break;
616             }
617         }
618 
619         /*
620          * If the region was not found in the current device memory state
621          * create an entry for it in the removed list.
622          */
623         if (!matching) {
624             rem_reg[rm_idx].region = shadow_reg;
625             rem_reg[rm_idx++].reg_idx = i;
626         }
627     }
628 
629     /*
630      * For regions not marked "found", create entries in the added list.
631      *
632      * Note their indexes in the device memory state and the indexes of their
633      * file descriptors.
634      */
635     for (i = 0; i < dev->mem->nregions; i++) {
636         reg = &dev->mem->regions[i];
637         vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
638         if (fd > 0) {
639             ++fd_num;
640         }
641 
642         /*
643          * If the region was in both the shadow and device state we don't
644          * need to send a VHOST_USER_ADD_MEM_REG message for it.
645          */
646         if (found[i]) {
647             continue;
648         }
649 
650         add_reg[add_idx].region = reg;
651         add_reg[add_idx].reg_idx = i;
652         add_reg[add_idx++].fd_idx = fd_num;
653     }
654     *nr_rem_reg = rm_idx;
655     *nr_add_reg = add_idx;
656 
657     return;
658 }
659 
660 static int send_remove_regions(struct vhost_dev *dev,
661                                struct scrub_regions *remove_reg,
662                                int nr_rem_reg, VhostUserMsg *msg,
663                                bool reply_supported)
664 {
665     struct vhost_user *u = dev->opaque;
666     struct vhost_memory_region *shadow_reg;
667     int i, fd, shadow_reg_idx, ret;
668     ram_addr_t offset;
669     VhostUserMemoryRegion region_buffer;
670 
671     /*
672      * The regions in remove_reg appear in the same order they do in the
673      * shadow table. Therefore we can minimize memory copies by iterating
674      * through remove_reg backwards.
675      */
676     for (i = nr_rem_reg - 1; i >= 0; i--) {
677         shadow_reg = remove_reg[i].region;
678         shadow_reg_idx = remove_reg[i].reg_idx;
679 
680         vhost_user_get_mr_data(shadow_reg->userspace_addr, &offset, &fd);
681 
682         if (fd > 0) {
683             msg->hdr.request = VHOST_USER_REM_MEM_REG;
684             vhost_user_fill_msg_region(&region_buffer, shadow_reg, 0);
685             msg->payload.mem_reg.region = region_buffer;
686 
687             ret = vhost_user_write(dev, msg, NULL, 0);
688             if (ret < 0) {
689                 return ret;
690             }
691 
692             if (reply_supported) {
693                 ret = process_message_reply(dev, msg);
694                 if (ret) {
695                     return ret;
696                 }
697             }
698         }
699 
700         /*
701          * At this point we know the backend has unmapped the region. It is now
702          * safe to remove it from the shadow table.
703          */
704         memmove(&u->shadow_regions[shadow_reg_idx],
705                 &u->shadow_regions[shadow_reg_idx + 1],
706                 sizeof(struct vhost_memory_region) *
707                 (u->num_shadow_regions - shadow_reg_idx - 1));
708         u->num_shadow_regions--;
709     }
710 
711     return 0;
712 }
713 
714 static int send_add_regions(struct vhost_dev *dev,
715                             struct scrub_regions *add_reg, int nr_add_reg,
716                             VhostUserMsg *msg, uint64_t *shadow_pcb,
717                             bool reply_supported, bool track_ramblocks)
718 {
719     struct vhost_user *u = dev->opaque;
720     int i, fd, ret, reg_idx, reg_fd_idx;
721     struct vhost_memory_region *reg;
722     MemoryRegion *mr;
723     ram_addr_t offset;
724     VhostUserMsg msg_reply;
725     VhostUserMemoryRegion region_buffer;
726 
727     for (i = 0; i < nr_add_reg; i++) {
728         reg = add_reg[i].region;
729         reg_idx = add_reg[i].reg_idx;
730         reg_fd_idx = add_reg[i].fd_idx;
731 
732         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
733 
734         if (fd > 0) {
735             if (track_ramblocks) {
736                 trace_vhost_user_set_mem_table_withfd(reg_fd_idx, mr->name,
737                                                       reg->memory_size,
738                                                       reg->guest_phys_addr,
739                                                       reg->userspace_addr,
740                                                       offset);
741                 u->region_rb_offset[reg_idx] = offset;
742                 u->region_rb[reg_idx] = mr->ram_block;
743             }
744             msg->hdr.request = VHOST_USER_ADD_MEM_REG;
745             vhost_user_fill_msg_region(&region_buffer, reg, offset);
746             msg->payload.mem_reg.region = region_buffer;
747 
748             ret = vhost_user_write(dev, msg, &fd, 1);
749             if (ret < 0) {
750                 return ret;
751             }
752 
753             if (track_ramblocks) {
754                 uint64_t reply_gpa;
755 
756                 ret = vhost_user_read(dev, &msg_reply);
757                 if (ret < 0) {
758                     return ret;
759                 }
760 
761                 reply_gpa = msg_reply.payload.mem_reg.region.guest_phys_addr;
762 
763                 if (msg_reply.hdr.request != VHOST_USER_ADD_MEM_REG) {
764                     error_report("%s: Received unexpected msg type."
765                                  "Expected %d received %d", __func__,
766                                  VHOST_USER_ADD_MEM_REG,
767                                  msg_reply.hdr.request);
768                     return -EPROTO;
769                 }
770 
771                 /*
772                  * We're using the same structure, just reusing one of the
773                  * fields, so it should be the same size.
774                  */
775                 if (msg_reply.hdr.size != msg->hdr.size) {
776                     error_report("%s: Unexpected size for postcopy reply "
777                                  "%d vs %d", __func__, msg_reply.hdr.size,
778                                  msg->hdr.size);
779                     return -EPROTO;
780                 }
781 
782                 /* Get the postcopy client base from the backend's reply. */
783                 if (reply_gpa == dev->mem->regions[reg_idx].guest_phys_addr) {
784                     shadow_pcb[reg_idx] =
785                         msg_reply.payload.mem_reg.region.userspace_addr;
786                     trace_vhost_user_set_mem_table_postcopy(
787                         msg_reply.payload.mem_reg.region.userspace_addr,
788                         msg->payload.mem_reg.region.userspace_addr,
789                         reg_fd_idx, reg_idx);
790                 } else {
791                     error_report("%s: invalid postcopy reply for region. "
792                                  "Got guest physical address %" PRIX64 ", expected "
793                                  "%" PRIX64, __func__, reply_gpa,
794                                  dev->mem->regions[reg_idx].guest_phys_addr);
795                     return -EPROTO;
796                 }
797             } else if (reply_supported) {
798                 ret = process_message_reply(dev, msg);
799                 if (ret) {
800                     return ret;
801                 }
802             }
803         } else if (track_ramblocks) {
804             u->region_rb_offset[reg_idx] = 0;
805             u->region_rb[reg_idx] = NULL;
806         }
807 
808         /*
809          * At this point, we know the backend has mapped in the new
810          * region, if the region has a valid file descriptor.
811          *
812          * The region should now be added to the shadow table.
813          */
814         u->shadow_regions[u->num_shadow_regions].guest_phys_addr =
815             reg->guest_phys_addr;
816         u->shadow_regions[u->num_shadow_regions].userspace_addr =
817             reg->userspace_addr;
818         u->shadow_regions[u->num_shadow_regions].memory_size =
819             reg->memory_size;
820         u->num_shadow_regions++;
821     }
822 
823     return 0;
824 }
825 
826 static int vhost_user_add_remove_regions(struct vhost_dev *dev,
827                                          VhostUserMsg *msg,
828                                          bool reply_supported,
829                                          bool track_ramblocks)
830 {
831     struct vhost_user *u = dev->opaque;
832     struct scrub_regions add_reg[VHOST_USER_MAX_RAM_SLOTS];
833     struct scrub_regions rem_reg[VHOST_USER_MAX_RAM_SLOTS];
834     uint64_t shadow_pcb[VHOST_USER_MAX_RAM_SLOTS] = {};
835     int nr_add_reg, nr_rem_reg;
836     int ret;
837 
838     msg->hdr.size = sizeof(msg->payload.mem_reg);
839 
840     /* Find the regions which need to be removed or added. */
841     scrub_shadow_regions(dev, add_reg, &nr_add_reg, rem_reg, &nr_rem_reg,
842                          shadow_pcb, track_ramblocks);
843 
844     if (nr_rem_reg) {
845         ret = send_remove_regions(dev, rem_reg, nr_rem_reg, msg,
846                                   reply_supported);
847         if (ret < 0) {
848             goto err;
849         }
850     }
851 
852     if (nr_add_reg) {
853         ret = send_add_regions(dev, add_reg, nr_add_reg, msg, shadow_pcb,
854                                reply_supported, track_ramblocks);
855         if (ret < 0) {
856             goto err;
857         }
858     }
859 
860     if (track_ramblocks) {
861         memcpy(u->postcopy_client_bases, shadow_pcb,
862                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
863         /*
864          * Now we've registered this with the postcopy code, we ack to the
865          * client, because now we're in the position to be able to deal with
866          * any faults it generates.
867          */
868         /* TODO: Use this for failure cases as well with a bad value. */
869         msg->hdr.size = sizeof(msg->payload.u64);
870         msg->payload.u64 = 0; /* OK */
871 
872         ret = vhost_user_write(dev, msg, NULL, 0);
873         if (ret < 0) {
874             return ret;
875         }
876     }
877 
878     return 0;
879 
880 err:
881     if (track_ramblocks) {
882         memcpy(u->postcopy_client_bases, shadow_pcb,
883                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
884     }
885 
886     return ret;
887 }
888 
889 static int vhost_user_set_mem_table_postcopy(struct vhost_dev *dev,
890                                              struct vhost_memory *mem,
891                                              bool reply_supported,
892                                              bool config_mem_slots)
893 {
894     struct vhost_user *u = dev->opaque;
895     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
896     size_t fd_num = 0;
897     VhostUserMsg msg_reply;
898     int region_i, msg_i;
899     int ret;
900 
901     VhostUserMsg msg = {
902         .hdr.flags = VHOST_USER_VERSION,
903     };
904 
905     if (u->region_rb_len < dev->mem->nregions) {
906         u->region_rb = g_renew(RAMBlock*, u->region_rb, dev->mem->nregions);
907         u->region_rb_offset = g_renew(ram_addr_t, u->region_rb_offset,
908                                       dev->mem->nregions);
909         memset(&(u->region_rb[u->region_rb_len]), '\0',
910                sizeof(RAMBlock *) * (dev->mem->nregions - u->region_rb_len));
911         memset(&(u->region_rb_offset[u->region_rb_len]), '\0',
912                sizeof(ram_addr_t) * (dev->mem->nregions - u->region_rb_len));
913         u->region_rb_len = dev->mem->nregions;
914     }
915 
916     if (config_mem_slots) {
917         ret = vhost_user_add_remove_regions(dev, &msg, reply_supported, true);
918         if (ret < 0) {
919             return ret;
920         }
921     } else {
922         ret = vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
923                                                 true);
924         if (ret < 0) {
925             return ret;
926         }
927 
928         ret = vhost_user_write(dev, &msg, fds, fd_num);
929         if (ret < 0) {
930             return ret;
931         }
932 
933         ret = vhost_user_read(dev, &msg_reply);
934         if (ret < 0) {
935             return ret;
936         }
937 
938         if (msg_reply.hdr.request != VHOST_USER_SET_MEM_TABLE) {
939             error_report("%s: Received unexpected msg type."
940                          "Expected %d received %d", __func__,
941                          VHOST_USER_SET_MEM_TABLE, msg_reply.hdr.request);
942             return -EPROTO;
943         }
944 
945         /*
946          * We're using the same structure, just reusing one of the
947          * fields, so it should be the same size.
948          */
949         if (msg_reply.hdr.size != msg.hdr.size) {
950             error_report("%s: Unexpected size for postcopy reply "
951                          "%d vs %d", __func__, msg_reply.hdr.size,
952                          msg.hdr.size);
953             return -EPROTO;
954         }
955 
956         memset(u->postcopy_client_bases, 0,
957                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
958 
959         /*
960          * They're in the same order as the regions that were sent
961          * but some of the regions were skipped (above) if they
962          * didn't have fd's
963          */
964         for (msg_i = 0, region_i = 0;
965              region_i < dev->mem->nregions;
966              region_i++) {
967             if (msg_i < fd_num &&
968                 msg_reply.payload.memory.regions[msg_i].guest_phys_addr ==
969                 dev->mem->regions[region_i].guest_phys_addr) {
970                 u->postcopy_client_bases[region_i] =
971                     msg_reply.payload.memory.regions[msg_i].userspace_addr;
972                 trace_vhost_user_set_mem_table_postcopy(
973                     msg_reply.payload.memory.regions[msg_i].userspace_addr,
974                     msg.payload.memory.regions[msg_i].userspace_addr,
975                     msg_i, region_i);
976                 msg_i++;
977             }
978         }
979         if (msg_i != fd_num) {
980             error_report("%s: postcopy reply not fully consumed "
981                          "%d vs %zd",
982                          __func__, msg_i, fd_num);
983             return -EIO;
984         }
985 
986         /*
987          * Now we've registered this with the postcopy code, we ack to the
988          * client, because now we're in the position to be able to deal
989          * with any faults it generates.
990          */
991         /* TODO: Use this for failure cases as well with a bad value. */
992         msg.hdr.size = sizeof(msg.payload.u64);
993         msg.payload.u64 = 0; /* OK */
994         ret = vhost_user_write(dev, &msg, NULL, 0);
995         if (ret < 0) {
996             return ret;
997         }
998     }
999 
1000     return 0;
1001 }
1002 
1003 static int vhost_user_set_mem_table(struct vhost_dev *dev,
1004                                     struct vhost_memory *mem)
1005 {
1006     struct vhost_user *u = dev->opaque;
1007     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
1008     size_t fd_num = 0;
1009     bool do_postcopy = u->postcopy_listen && u->postcopy_fd.handler;
1010     bool reply_supported = virtio_has_feature(dev->protocol_features,
1011                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1012     bool config_mem_slots =
1013         virtio_has_feature(dev->protocol_features,
1014                            VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS);
1015     int ret;
1016 
1017     if (do_postcopy) {
1018         /*
1019          * Postcopy has enough differences that it's best done in it's own
1020          * version
1021          */
1022         return vhost_user_set_mem_table_postcopy(dev, mem, reply_supported,
1023                                                  config_mem_slots);
1024     }
1025 
1026     VhostUserMsg msg = {
1027         .hdr.flags = VHOST_USER_VERSION,
1028     };
1029 
1030     if (reply_supported) {
1031         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1032     }
1033 
1034     if (config_mem_slots) {
1035         ret = vhost_user_add_remove_regions(dev, &msg, reply_supported, false);
1036         if (ret < 0) {
1037             return ret;
1038         }
1039     } else {
1040         ret = vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
1041                                                 false);
1042         if (ret < 0) {
1043             return ret;
1044         }
1045 
1046         ret = vhost_user_write(dev, &msg, fds, fd_num);
1047         if (ret < 0) {
1048             return ret;
1049         }
1050 
1051         if (reply_supported) {
1052             return process_message_reply(dev, &msg);
1053         }
1054     }
1055 
1056     return 0;
1057 }
1058 
1059 static int vhost_user_set_vring_endian(struct vhost_dev *dev,
1060                                        struct vhost_vring_state *ring)
1061 {
1062     bool cross_endian = virtio_has_feature(dev->protocol_features,
1063                                            VHOST_USER_PROTOCOL_F_CROSS_ENDIAN);
1064     VhostUserMsg msg = {
1065         .hdr.request = VHOST_USER_SET_VRING_ENDIAN,
1066         .hdr.flags = VHOST_USER_VERSION,
1067         .payload.state = *ring,
1068         .hdr.size = sizeof(msg.payload.state),
1069     };
1070 
1071     if (!cross_endian) {
1072         error_report("vhost-user trying to send unhandled ioctl");
1073         return -ENOTSUP;
1074     }
1075 
1076     return vhost_user_write(dev, &msg, NULL, 0);
1077 }
1078 
1079 static int vhost_set_vring(struct vhost_dev *dev,
1080                            unsigned long int request,
1081                            struct vhost_vring_state *ring)
1082 {
1083     VhostUserMsg msg = {
1084         .hdr.request = request,
1085         .hdr.flags = VHOST_USER_VERSION,
1086         .payload.state = *ring,
1087         .hdr.size = sizeof(msg.payload.state),
1088     };
1089 
1090     return vhost_user_write(dev, &msg, NULL, 0);
1091 }
1092 
1093 static int vhost_user_set_vring_num(struct vhost_dev *dev,
1094                                     struct vhost_vring_state *ring)
1095 {
1096     return vhost_set_vring(dev, VHOST_USER_SET_VRING_NUM, ring);
1097 }
1098 
1099 static void vhost_user_host_notifier_free(VhostUserHostNotifier *n)
1100 {
1101     assert(n && n->unmap_addr);
1102     munmap(n->unmap_addr, qemu_real_host_page_size());
1103     n->unmap_addr = NULL;
1104 }
1105 
1106 /*
1107  * clean-up function for notifier, will finally free the structure
1108  * under rcu.
1109  */
1110 static void vhost_user_host_notifier_remove(VhostUserHostNotifier *n,
1111                                             VirtIODevice *vdev)
1112 {
1113     if (n->addr) {
1114         if (vdev) {
1115             virtio_queue_set_host_notifier_mr(vdev, n->idx, &n->mr, false);
1116         }
1117         assert(!n->unmap_addr);
1118         n->unmap_addr = n->addr;
1119         n->addr = NULL;
1120         call_rcu(n, vhost_user_host_notifier_free, rcu);
1121     }
1122 }
1123 
1124 static int vhost_user_set_vring_base(struct vhost_dev *dev,
1125                                      struct vhost_vring_state *ring)
1126 {
1127     return vhost_set_vring(dev, VHOST_USER_SET_VRING_BASE, ring);
1128 }
1129 
1130 static int vhost_user_set_vring_enable(struct vhost_dev *dev, int enable)
1131 {
1132     int i;
1133 
1134     if (!virtio_has_feature(dev->features, VHOST_USER_F_PROTOCOL_FEATURES)) {
1135         return -EINVAL;
1136     }
1137 
1138     for (i = 0; i < dev->nvqs; ++i) {
1139         int ret;
1140         struct vhost_vring_state state = {
1141             .index = dev->vq_index + i,
1142             .num   = enable,
1143         };
1144 
1145         ret = vhost_set_vring(dev, VHOST_USER_SET_VRING_ENABLE, &state);
1146         if (ret < 0) {
1147             /*
1148              * Restoring the previous state is likely infeasible, as well as
1149              * proceeding regardless the error, so just bail out and hope for
1150              * the device-level recovery.
1151              */
1152             return ret;
1153         }
1154     }
1155 
1156     return 0;
1157 }
1158 
1159 static VhostUserHostNotifier *fetch_notifier(VhostUserState *u,
1160                                              int idx)
1161 {
1162     if (idx >= u->notifiers->len) {
1163         return NULL;
1164     }
1165     return g_ptr_array_index(u->notifiers, idx);
1166 }
1167 
1168 static int vhost_user_get_vring_base(struct vhost_dev *dev,
1169                                      struct vhost_vring_state *ring)
1170 {
1171     int ret;
1172     VhostUserMsg msg = {
1173         .hdr.request = VHOST_USER_GET_VRING_BASE,
1174         .hdr.flags = VHOST_USER_VERSION,
1175         .payload.state = *ring,
1176         .hdr.size = sizeof(msg.payload.state),
1177     };
1178     struct vhost_user *u = dev->opaque;
1179 
1180     VhostUserHostNotifier *n = fetch_notifier(u->user, ring->index);
1181     if (n) {
1182         vhost_user_host_notifier_remove(n, dev->vdev);
1183     }
1184 
1185     ret = vhost_user_write(dev, &msg, NULL, 0);
1186     if (ret < 0) {
1187         return ret;
1188     }
1189 
1190     ret = vhost_user_read(dev, &msg);
1191     if (ret < 0) {
1192         return ret;
1193     }
1194 
1195     if (msg.hdr.request != VHOST_USER_GET_VRING_BASE) {
1196         error_report("Received unexpected msg type. Expected %d received %d",
1197                      VHOST_USER_GET_VRING_BASE, msg.hdr.request);
1198         return -EPROTO;
1199     }
1200 
1201     if (msg.hdr.size != sizeof(msg.payload.state)) {
1202         error_report("Received bad msg size.");
1203         return -EPROTO;
1204     }
1205 
1206     *ring = msg.payload.state;
1207 
1208     return 0;
1209 }
1210 
1211 static int vhost_set_vring_file(struct vhost_dev *dev,
1212                                 VhostUserRequest request,
1213                                 struct vhost_vring_file *file)
1214 {
1215     int fds[VHOST_USER_MAX_RAM_SLOTS];
1216     size_t fd_num = 0;
1217     VhostUserMsg msg = {
1218         .hdr.request = request,
1219         .hdr.flags = VHOST_USER_VERSION,
1220         .payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK,
1221         .hdr.size = sizeof(msg.payload.u64),
1222     };
1223 
1224     if (ioeventfd_enabled() && file->fd > 0) {
1225         fds[fd_num++] = file->fd;
1226     } else {
1227         msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
1228     }
1229 
1230     return vhost_user_write(dev, &msg, fds, fd_num);
1231 }
1232 
1233 static int vhost_user_set_vring_kick(struct vhost_dev *dev,
1234                                      struct vhost_vring_file *file)
1235 {
1236     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_KICK, file);
1237 }
1238 
1239 static int vhost_user_set_vring_call(struct vhost_dev *dev,
1240                                      struct vhost_vring_file *file)
1241 {
1242     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_CALL, file);
1243 }
1244 
1245 static int vhost_user_set_vring_err(struct vhost_dev *dev,
1246                                     struct vhost_vring_file *file)
1247 {
1248     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_ERR, file);
1249 }
1250 
1251 static int vhost_user_get_u64(struct vhost_dev *dev, int request, uint64_t *u64)
1252 {
1253     int ret;
1254     VhostUserMsg msg = {
1255         .hdr.request = request,
1256         .hdr.flags = VHOST_USER_VERSION,
1257     };
1258 
1259     if (vhost_user_one_time_request(request) && dev->vq_index != 0) {
1260         return 0;
1261     }
1262 
1263     ret = vhost_user_write(dev, &msg, NULL, 0);
1264     if (ret < 0) {
1265         return ret;
1266     }
1267 
1268     ret = vhost_user_read(dev, &msg);
1269     if (ret < 0) {
1270         return ret;
1271     }
1272 
1273     if (msg.hdr.request != request) {
1274         error_report("Received unexpected msg type. Expected %d received %d",
1275                      request, msg.hdr.request);
1276         return -EPROTO;
1277     }
1278 
1279     if (msg.hdr.size != sizeof(msg.payload.u64)) {
1280         error_report("Received bad msg size.");
1281         return -EPROTO;
1282     }
1283 
1284     *u64 = msg.payload.u64;
1285 
1286     return 0;
1287 }
1288 
1289 static int vhost_user_get_features(struct vhost_dev *dev, uint64_t *features)
1290 {
1291     if (vhost_user_get_u64(dev, VHOST_USER_GET_FEATURES, features) < 0) {
1292         return -EPROTO;
1293     }
1294 
1295     return 0;
1296 }
1297 
1298 static int enforce_reply(struct vhost_dev *dev,
1299                          const VhostUserMsg *msg)
1300 {
1301     uint64_t dummy;
1302 
1303     if (msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1304         return process_message_reply(dev, msg);
1305     }
1306 
1307    /*
1308     * We need to wait for a reply but the backend does not
1309     * support replies for the command we just sent.
1310     * Send VHOST_USER_GET_FEATURES which makes all backends
1311     * send a reply.
1312     */
1313     return vhost_user_get_features(dev, &dummy);
1314 }
1315 
1316 static int vhost_user_set_vring_addr(struct vhost_dev *dev,
1317                                      struct vhost_vring_addr *addr)
1318 {
1319     int ret;
1320     VhostUserMsg msg = {
1321         .hdr.request = VHOST_USER_SET_VRING_ADDR,
1322         .hdr.flags = VHOST_USER_VERSION,
1323         .payload.addr = *addr,
1324         .hdr.size = sizeof(msg.payload.addr),
1325     };
1326 
1327     bool reply_supported = virtio_has_feature(dev->protocol_features,
1328                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1329 
1330     /*
1331      * wait for a reply if logging is enabled to make sure
1332      * backend is actually logging changes
1333      */
1334     bool wait_for_reply = addr->flags & (1 << VHOST_VRING_F_LOG);
1335 
1336     if (reply_supported && wait_for_reply) {
1337         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1338     }
1339 
1340     ret = vhost_user_write(dev, &msg, NULL, 0);
1341     if (ret < 0) {
1342         return ret;
1343     }
1344 
1345     if (wait_for_reply) {
1346         return enforce_reply(dev, &msg);
1347     }
1348 
1349     return 0;
1350 }
1351 
1352 static int vhost_user_set_u64(struct vhost_dev *dev, int request, uint64_t u64,
1353                               bool wait_for_reply)
1354 {
1355     VhostUserMsg msg = {
1356         .hdr.request = request,
1357         .hdr.flags = VHOST_USER_VERSION,
1358         .payload.u64 = u64,
1359         .hdr.size = sizeof(msg.payload.u64),
1360     };
1361     int ret;
1362 
1363     if (wait_for_reply) {
1364         bool reply_supported = virtio_has_feature(dev->protocol_features,
1365                                           VHOST_USER_PROTOCOL_F_REPLY_ACK);
1366         if (reply_supported) {
1367             msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1368         }
1369     }
1370 
1371     ret = vhost_user_write(dev, &msg, NULL, 0);
1372     if (ret < 0) {
1373         return ret;
1374     }
1375 
1376     if (wait_for_reply) {
1377         return enforce_reply(dev, &msg);
1378     }
1379 
1380     return 0;
1381 }
1382 
1383 static int vhost_user_set_status(struct vhost_dev *dev, uint8_t status)
1384 {
1385     return vhost_user_set_u64(dev, VHOST_USER_SET_STATUS, status, false);
1386 }
1387 
1388 static int vhost_user_get_status(struct vhost_dev *dev, uint8_t *status)
1389 {
1390     uint64_t value;
1391     int ret;
1392 
1393     ret = vhost_user_get_u64(dev, VHOST_USER_GET_STATUS, &value);
1394     if (ret < 0) {
1395         return ret;
1396     }
1397     *status = value;
1398 
1399     return 0;
1400 }
1401 
1402 static int vhost_user_add_status(struct vhost_dev *dev, uint8_t status)
1403 {
1404     uint8_t s;
1405     int ret;
1406 
1407     ret = vhost_user_get_status(dev, &s);
1408     if (ret < 0) {
1409         return ret;
1410     }
1411 
1412     if ((s & status) == status) {
1413         return 0;
1414     }
1415     s |= status;
1416 
1417     return vhost_user_set_status(dev, s);
1418 }
1419 
1420 static int vhost_user_set_features(struct vhost_dev *dev,
1421                                    uint64_t features)
1422 {
1423     /*
1424      * wait for a reply if logging is enabled to make sure
1425      * backend is actually logging changes
1426      */
1427     bool log_enabled = features & (0x1ULL << VHOST_F_LOG_ALL);
1428     int ret;
1429 
1430     /*
1431      * We need to include any extra backend only feature bits that
1432      * might be needed by our device. Currently this includes the
1433      * VHOST_USER_F_PROTOCOL_FEATURES bit for enabling protocol
1434      * features.
1435      */
1436     ret = vhost_user_set_u64(dev, VHOST_USER_SET_FEATURES,
1437                               features | dev->backend_features,
1438                               log_enabled);
1439 
1440     if (virtio_has_feature(dev->protocol_features,
1441                            VHOST_USER_PROTOCOL_F_STATUS)) {
1442         if (!ret) {
1443             return vhost_user_add_status(dev, VIRTIO_CONFIG_S_FEATURES_OK);
1444         }
1445     }
1446 
1447     return ret;
1448 }
1449 
1450 static int vhost_user_set_protocol_features(struct vhost_dev *dev,
1451                                             uint64_t features)
1452 {
1453     return vhost_user_set_u64(dev, VHOST_USER_SET_PROTOCOL_FEATURES, features,
1454                               false);
1455 }
1456 
1457 static int vhost_user_set_owner(struct vhost_dev *dev)
1458 {
1459     VhostUserMsg msg = {
1460         .hdr.request = VHOST_USER_SET_OWNER,
1461         .hdr.flags = VHOST_USER_VERSION,
1462     };
1463 
1464     return vhost_user_write(dev, &msg, NULL, 0);
1465 }
1466 
1467 static int vhost_user_get_max_memslots(struct vhost_dev *dev,
1468                                        uint64_t *max_memslots)
1469 {
1470     uint64_t backend_max_memslots;
1471     int err;
1472 
1473     err = vhost_user_get_u64(dev, VHOST_USER_GET_MAX_MEM_SLOTS,
1474                              &backend_max_memslots);
1475     if (err < 0) {
1476         return err;
1477     }
1478 
1479     *max_memslots = backend_max_memslots;
1480 
1481     return 0;
1482 }
1483 
1484 static int vhost_user_reset_device(struct vhost_dev *dev)
1485 {
1486     VhostUserMsg msg = {
1487         .hdr.flags = VHOST_USER_VERSION,
1488     };
1489 
1490     msg.hdr.request = virtio_has_feature(dev->protocol_features,
1491                                          VHOST_USER_PROTOCOL_F_RESET_DEVICE)
1492         ? VHOST_USER_RESET_DEVICE
1493         : VHOST_USER_RESET_OWNER;
1494 
1495     return vhost_user_write(dev, &msg, NULL, 0);
1496 }
1497 
1498 static int vhost_user_slave_handle_config_change(struct vhost_dev *dev)
1499 {
1500     if (!dev->config_ops || !dev->config_ops->vhost_dev_config_notifier) {
1501         return -ENOSYS;
1502     }
1503 
1504     return dev->config_ops->vhost_dev_config_notifier(dev);
1505 }
1506 
1507 /*
1508  * Fetch or create the notifier for a given idx. Newly created
1509  * notifiers are added to the pointer array that tracks them.
1510  */
1511 static VhostUserHostNotifier *fetch_or_create_notifier(VhostUserState *u,
1512                                                        int idx)
1513 {
1514     VhostUserHostNotifier *n = NULL;
1515     if (idx >= u->notifiers->len) {
1516         g_ptr_array_set_size(u->notifiers, idx + 1);
1517     }
1518 
1519     n = g_ptr_array_index(u->notifiers, idx);
1520     if (!n) {
1521         /*
1522          * In case notification arrive out-of-order,
1523          * make room for current index.
1524          */
1525         g_ptr_array_remove_index(u->notifiers, idx);
1526         n = g_new0(VhostUserHostNotifier, 1);
1527         n->idx = idx;
1528         g_ptr_array_insert(u->notifiers, idx, n);
1529         trace_vhost_user_create_notifier(idx, n);
1530     }
1531 
1532     return n;
1533 }
1534 
1535 static int vhost_user_slave_handle_vring_host_notifier(struct vhost_dev *dev,
1536                                                        VhostUserVringArea *area,
1537                                                        int fd)
1538 {
1539     int queue_idx = area->u64 & VHOST_USER_VRING_IDX_MASK;
1540     size_t page_size = qemu_real_host_page_size();
1541     struct vhost_user *u = dev->opaque;
1542     VhostUserState *user = u->user;
1543     VirtIODevice *vdev = dev->vdev;
1544     VhostUserHostNotifier *n;
1545     void *addr;
1546     char *name;
1547 
1548     if (!virtio_has_feature(dev->protocol_features,
1549                             VHOST_USER_PROTOCOL_F_HOST_NOTIFIER) ||
1550         vdev == NULL || queue_idx >= virtio_get_num_queues(vdev)) {
1551         return -EINVAL;
1552     }
1553 
1554     /*
1555      * Fetch notifier and invalidate any old data before setting up
1556      * new mapped address.
1557      */
1558     n = fetch_or_create_notifier(user, queue_idx);
1559     vhost_user_host_notifier_remove(n, vdev);
1560 
1561     if (area->u64 & VHOST_USER_VRING_NOFD_MASK) {
1562         return 0;
1563     }
1564 
1565     /* Sanity check. */
1566     if (area->size != page_size) {
1567         return -EINVAL;
1568     }
1569 
1570     addr = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1571                 fd, area->offset);
1572     if (addr == MAP_FAILED) {
1573         return -EFAULT;
1574     }
1575 
1576     name = g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
1577                            user, queue_idx);
1578     if (!n->mr.ram) { /* Don't init again after suspend. */
1579         memory_region_init_ram_device_ptr(&n->mr, OBJECT(vdev), name,
1580                                           page_size, addr);
1581     } else {
1582         n->mr.ram_block->host = addr;
1583     }
1584     g_free(name);
1585 
1586     if (virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, true)) {
1587         object_unparent(OBJECT(&n->mr));
1588         munmap(addr, page_size);
1589         return -ENXIO;
1590     }
1591 
1592     n->addr = addr;
1593 
1594     return 0;
1595 }
1596 
1597 static void close_slave_channel(struct vhost_user *u)
1598 {
1599     g_source_destroy(u->slave_src);
1600     g_source_unref(u->slave_src);
1601     u->slave_src = NULL;
1602     object_unref(OBJECT(u->slave_ioc));
1603     u->slave_ioc = NULL;
1604 }
1605 
1606 static gboolean slave_read(QIOChannel *ioc, GIOCondition condition,
1607                            gpointer opaque)
1608 {
1609     struct vhost_dev *dev = opaque;
1610     struct vhost_user *u = dev->opaque;
1611     VhostUserHeader hdr = { 0, };
1612     VhostUserPayload payload = { 0, };
1613     Error *local_err = NULL;
1614     gboolean rc = G_SOURCE_CONTINUE;
1615     int ret = 0;
1616     struct iovec iov;
1617     g_autofree int *fd = NULL;
1618     size_t fdsize = 0;
1619     int i;
1620 
1621     /* Read header */
1622     iov.iov_base = &hdr;
1623     iov.iov_len = VHOST_USER_HDR_SIZE;
1624 
1625     if (qio_channel_readv_full_all(ioc, &iov, 1, &fd, &fdsize, &local_err)) {
1626         error_report_err(local_err);
1627         goto err;
1628     }
1629 
1630     if (hdr.size > VHOST_USER_PAYLOAD_SIZE) {
1631         error_report("Failed to read msg header."
1632                 " Size %d exceeds the maximum %zu.", hdr.size,
1633                 VHOST_USER_PAYLOAD_SIZE);
1634         goto err;
1635     }
1636 
1637     /* Read payload */
1638     if (qio_channel_read_all(ioc, (char *) &payload, hdr.size, &local_err)) {
1639         error_report_err(local_err);
1640         goto err;
1641     }
1642 
1643     switch (hdr.request) {
1644     case VHOST_USER_BACKEND_IOTLB_MSG:
1645         ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
1646         break;
1647     case VHOST_USER_BACKEND_CONFIG_CHANGE_MSG:
1648         ret = vhost_user_slave_handle_config_change(dev);
1649         break;
1650     case VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG:
1651         ret = vhost_user_slave_handle_vring_host_notifier(dev, &payload.area,
1652                                                           fd ? fd[0] : -1);
1653         break;
1654     default:
1655         error_report("Received unexpected msg type: %d.", hdr.request);
1656         ret = -EINVAL;
1657     }
1658 
1659     /*
1660      * REPLY_ACK feature handling. Other reply types has to be managed
1661      * directly in their request handlers.
1662      */
1663     if (hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1664         struct iovec iovec[2];
1665 
1666 
1667         hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
1668         hdr.flags |= VHOST_USER_REPLY_MASK;
1669 
1670         payload.u64 = !!ret;
1671         hdr.size = sizeof(payload.u64);
1672 
1673         iovec[0].iov_base = &hdr;
1674         iovec[0].iov_len = VHOST_USER_HDR_SIZE;
1675         iovec[1].iov_base = &payload;
1676         iovec[1].iov_len = hdr.size;
1677 
1678         if (qio_channel_writev_all(ioc, iovec, ARRAY_SIZE(iovec), &local_err)) {
1679             error_report_err(local_err);
1680             goto err;
1681         }
1682     }
1683 
1684     goto fdcleanup;
1685 
1686 err:
1687     close_slave_channel(u);
1688     rc = G_SOURCE_REMOVE;
1689 
1690 fdcleanup:
1691     if (fd) {
1692         for (i = 0; i < fdsize; i++) {
1693             close(fd[i]);
1694         }
1695     }
1696     return rc;
1697 }
1698 
1699 static int vhost_setup_slave_channel(struct vhost_dev *dev)
1700 {
1701     VhostUserMsg msg = {
1702         .hdr.request = VHOST_USER_SET_BACKEND_REQ_FD,
1703         .hdr.flags = VHOST_USER_VERSION,
1704     };
1705     struct vhost_user *u = dev->opaque;
1706     int sv[2], ret = 0;
1707     bool reply_supported = virtio_has_feature(dev->protocol_features,
1708                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1709     Error *local_err = NULL;
1710     QIOChannel *ioc;
1711 
1712     if (!virtio_has_feature(dev->protocol_features,
1713                             VHOST_USER_PROTOCOL_F_BACKEND_REQ)) {
1714         return 0;
1715     }
1716 
1717     if (qemu_socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
1718         int saved_errno = errno;
1719         error_report("socketpair() failed");
1720         return -saved_errno;
1721     }
1722 
1723     ioc = QIO_CHANNEL(qio_channel_socket_new_fd(sv[0], &local_err));
1724     if (!ioc) {
1725         error_report_err(local_err);
1726         return -ECONNREFUSED;
1727     }
1728     u->slave_ioc = ioc;
1729     u->slave_src = qio_channel_add_watch_source(u->slave_ioc,
1730                                                 G_IO_IN | G_IO_HUP,
1731                                                 slave_read, dev, NULL, NULL);
1732 
1733     if (reply_supported) {
1734         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1735     }
1736 
1737     ret = vhost_user_write(dev, &msg, &sv[1], 1);
1738     if (ret) {
1739         goto out;
1740     }
1741 
1742     if (reply_supported) {
1743         ret = process_message_reply(dev, &msg);
1744     }
1745 
1746 out:
1747     close(sv[1]);
1748     if (ret) {
1749         close_slave_channel(u);
1750     }
1751 
1752     return ret;
1753 }
1754 
1755 #ifdef CONFIG_LINUX
1756 /*
1757  * Called back from the postcopy fault thread when a fault is received on our
1758  * ufd.
1759  * TODO: This is Linux specific
1760  */
1761 static int vhost_user_postcopy_fault_handler(struct PostCopyFD *pcfd,
1762                                              void *ufd)
1763 {
1764     struct vhost_dev *dev = pcfd->data;
1765     struct vhost_user *u = dev->opaque;
1766     struct uffd_msg *msg = ufd;
1767     uint64_t faultaddr = msg->arg.pagefault.address;
1768     RAMBlock *rb = NULL;
1769     uint64_t rb_offset;
1770     int i;
1771 
1772     trace_vhost_user_postcopy_fault_handler(pcfd->idstr, faultaddr,
1773                                             dev->mem->nregions);
1774     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1775         trace_vhost_user_postcopy_fault_handler_loop(i,
1776                 u->postcopy_client_bases[i], dev->mem->regions[i].memory_size);
1777         if (faultaddr >= u->postcopy_client_bases[i]) {
1778             /* Ofset of the fault address in the vhost region */
1779             uint64_t region_offset = faultaddr - u->postcopy_client_bases[i];
1780             if (region_offset < dev->mem->regions[i].memory_size) {
1781                 rb_offset = region_offset + u->region_rb_offset[i];
1782                 trace_vhost_user_postcopy_fault_handler_found(i,
1783                         region_offset, rb_offset);
1784                 rb = u->region_rb[i];
1785                 return postcopy_request_shared_page(pcfd, rb, faultaddr,
1786                                                     rb_offset);
1787             }
1788         }
1789     }
1790     error_report("%s: Failed to find region for fault %" PRIx64,
1791                  __func__, faultaddr);
1792     return -1;
1793 }
1794 
1795 static int vhost_user_postcopy_waker(struct PostCopyFD *pcfd, RAMBlock *rb,
1796                                      uint64_t offset)
1797 {
1798     struct vhost_dev *dev = pcfd->data;
1799     struct vhost_user *u = dev->opaque;
1800     int i;
1801 
1802     trace_vhost_user_postcopy_waker(qemu_ram_get_idstr(rb), offset);
1803 
1804     if (!u) {
1805         return 0;
1806     }
1807     /* Translate the offset into an address in the clients address space */
1808     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1809         if (u->region_rb[i] == rb &&
1810             offset >= u->region_rb_offset[i] &&
1811             offset < (u->region_rb_offset[i] +
1812                       dev->mem->regions[i].memory_size)) {
1813             uint64_t client_addr = (offset - u->region_rb_offset[i]) +
1814                                    u->postcopy_client_bases[i];
1815             trace_vhost_user_postcopy_waker_found(client_addr);
1816             return postcopy_wake_shared(pcfd, client_addr, rb);
1817         }
1818     }
1819 
1820     trace_vhost_user_postcopy_waker_nomatch(qemu_ram_get_idstr(rb), offset);
1821     return 0;
1822 }
1823 #endif
1824 
1825 /*
1826  * Called at the start of an inbound postcopy on reception of the
1827  * 'advise' command.
1828  */
1829 static int vhost_user_postcopy_advise(struct vhost_dev *dev, Error **errp)
1830 {
1831 #ifdef CONFIG_LINUX
1832     struct vhost_user *u = dev->opaque;
1833     CharBackend *chr = u->user->chr;
1834     int ufd;
1835     int ret;
1836     VhostUserMsg msg = {
1837         .hdr.request = VHOST_USER_POSTCOPY_ADVISE,
1838         .hdr.flags = VHOST_USER_VERSION,
1839     };
1840 
1841     ret = vhost_user_write(dev, &msg, NULL, 0);
1842     if (ret < 0) {
1843         error_setg(errp, "Failed to send postcopy_advise to vhost");
1844         return ret;
1845     }
1846 
1847     ret = vhost_user_read(dev, &msg);
1848     if (ret < 0) {
1849         error_setg(errp, "Failed to get postcopy_advise reply from vhost");
1850         return ret;
1851     }
1852 
1853     if (msg.hdr.request != VHOST_USER_POSTCOPY_ADVISE) {
1854         error_setg(errp, "Unexpected msg type. Expected %d received %d",
1855                      VHOST_USER_POSTCOPY_ADVISE, msg.hdr.request);
1856         return -EPROTO;
1857     }
1858 
1859     if (msg.hdr.size) {
1860         error_setg(errp, "Received bad msg size.");
1861         return -EPROTO;
1862     }
1863     ufd = qemu_chr_fe_get_msgfd(chr);
1864     if (ufd < 0) {
1865         error_setg(errp, "%s: Failed to get ufd", __func__);
1866         return -EIO;
1867     }
1868     qemu_socket_set_nonblock(ufd);
1869 
1870     /* register ufd with userfault thread */
1871     u->postcopy_fd.fd = ufd;
1872     u->postcopy_fd.data = dev;
1873     u->postcopy_fd.handler = vhost_user_postcopy_fault_handler;
1874     u->postcopy_fd.waker = vhost_user_postcopy_waker;
1875     u->postcopy_fd.idstr = "vhost-user"; /* Need to find unique name */
1876     postcopy_register_shared_ufd(&u->postcopy_fd);
1877     return 0;
1878 #else
1879     error_setg(errp, "Postcopy not supported on non-Linux systems");
1880     return -ENOSYS;
1881 #endif
1882 }
1883 
1884 /*
1885  * Called at the switch to postcopy on reception of the 'listen' command.
1886  */
1887 static int vhost_user_postcopy_listen(struct vhost_dev *dev, Error **errp)
1888 {
1889     struct vhost_user *u = dev->opaque;
1890     int ret;
1891     VhostUserMsg msg = {
1892         .hdr.request = VHOST_USER_POSTCOPY_LISTEN,
1893         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
1894     };
1895     u->postcopy_listen = true;
1896 
1897     trace_vhost_user_postcopy_listen();
1898 
1899     ret = vhost_user_write(dev, &msg, NULL, 0);
1900     if (ret < 0) {
1901         error_setg(errp, "Failed to send postcopy_listen to vhost");
1902         return ret;
1903     }
1904 
1905     ret = process_message_reply(dev, &msg);
1906     if (ret) {
1907         error_setg(errp, "Failed to receive reply to postcopy_listen");
1908         return ret;
1909     }
1910 
1911     return 0;
1912 }
1913 
1914 /*
1915  * Called at the end of postcopy
1916  */
1917 static int vhost_user_postcopy_end(struct vhost_dev *dev, Error **errp)
1918 {
1919     VhostUserMsg msg = {
1920         .hdr.request = VHOST_USER_POSTCOPY_END,
1921         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
1922     };
1923     int ret;
1924     struct vhost_user *u = dev->opaque;
1925 
1926     trace_vhost_user_postcopy_end_entry();
1927 
1928     ret = vhost_user_write(dev, &msg, NULL, 0);
1929     if (ret < 0) {
1930         error_setg(errp, "Failed to send postcopy_end to vhost");
1931         return ret;
1932     }
1933 
1934     ret = process_message_reply(dev, &msg);
1935     if (ret) {
1936         error_setg(errp, "Failed to receive reply to postcopy_end");
1937         return ret;
1938     }
1939     postcopy_unregister_shared_ufd(&u->postcopy_fd);
1940     close(u->postcopy_fd.fd);
1941     u->postcopy_fd.handler = NULL;
1942 
1943     trace_vhost_user_postcopy_end_exit();
1944 
1945     return 0;
1946 }
1947 
1948 static int vhost_user_postcopy_notifier(NotifierWithReturn *notifier,
1949                                         void *opaque)
1950 {
1951     struct PostcopyNotifyData *pnd = opaque;
1952     struct vhost_user *u = container_of(notifier, struct vhost_user,
1953                                          postcopy_notifier);
1954     struct vhost_dev *dev = u->dev;
1955 
1956     switch (pnd->reason) {
1957     case POSTCOPY_NOTIFY_PROBE:
1958         if (!virtio_has_feature(dev->protocol_features,
1959                                 VHOST_USER_PROTOCOL_F_PAGEFAULT)) {
1960             /* TODO: Get the device name into this error somehow */
1961             error_setg(pnd->errp,
1962                        "vhost-user backend not capable of postcopy");
1963             return -ENOENT;
1964         }
1965         break;
1966 
1967     case POSTCOPY_NOTIFY_INBOUND_ADVISE:
1968         return vhost_user_postcopy_advise(dev, pnd->errp);
1969 
1970     case POSTCOPY_NOTIFY_INBOUND_LISTEN:
1971         return vhost_user_postcopy_listen(dev, pnd->errp);
1972 
1973     case POSTCOPY_NOTIFY_INBOUND_END:
1974         return vhost_user_postcopy_end(dev, pnd->errp);
1975 
1976     default:
1977         /* We ignore notifications we don't know */
1978         break;
1979     }
1980 
1981     return 0;
1982 }
1983 
1984 static int vhost_user_backend_init(struct vhost_dev *dev, void *opaque,
1985                                    Error **errp)
1986 {
1987     uint64_t features, ram_slots;
1988     struct vhost_user *u;
1989     VhostUserState *vus = (VhostUserState *) opaque;
1990     int err;
1991 
1992     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
1993 
1994     u = g_new0(struct vhost_user, 1);
1995     u->user = vus;
1996     u->dev = dev;
1997     dev->opaque = u;
1998 
1999     err = vhost_user_get_features(dev, &features);
2000     if (err < 0) {
2001         error_setg_errno(errp, -err, "vhost_backend_init failed");
2002         return err;
2003     }
2004 
2005     if (virtio_has_feature(features, VHOST_USER_F_PROTOCOL_FEATURES)) {
2006         bool supports_f_config = vus->supports_config ||
2007             (dev->config_ops && dev->config_ops->vhost_dev_config_notifier);
2008         uint64_t protocol_features;
2009 
2010         dev->backend_features |= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
2011 
2012         err = vhost_user_get_u64(dev, VHOST_USER_GET_PROTOCOL_FEATURES,
2013                                  &protocol_features);
2014         if (err < 0) {
2015             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2016             return -EPROTO;
2017         }
2018 
2019         /*
2020          * We will use all the protocol features we support - although
2021          * we suppress F_CONFIG if we know QEMUs internal code can not support
2022          * it.
2023          */
2024         protocol_features &= VHOST_USER_PROTOCOL_FEATURE_MASK;
2025 
2026         if (supports_f_config) {
2027             if (!virtio_has_feature(protocol_features,
2028                                     VHOST_USER_PROTOCOL_F_CONFIG)) {
2029                 error_setg(errp, "vhost-user device expecting "
2030                            "VHOST_USER_PROTOCOL_F_CONFIG but the vhost-user backend does "
2031                            "not support it.");
2032                 return -EPROTO;
2033             }
2034         } else {
2035             if (virtio_has_feature(protocol_features,
2036                                    VHOST_USER_PROTOCOL_F_CONFIG)) {
2037                 warn_report("vhost-user backend supports "
2038                             "VHOST_USER_PROTOCOL_F_CONFIG but QEMU does not.");
2039                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG);
2040             }
2041         }
2042 
2043         /* final set of protocol features */
2044         dev->protocol_features = protocol_features;
2045         err = vhost_user_set_protocol_features(dev, dev->protocol_features);
2046         if (err < 0) {
2047             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2048             return -EPROTO;
2049         }
2050 
2051         /* query the max queues we support if backend supports Multiple Queue */
2052         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)) {
2053             err = vhost_user_get_u64(dev, VHOST_USER_GET_QUEUE_NUM,
2054                                      &dev->max_queues);
2055             if (err < 0) {
2056                 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2057                 return -EPROTO;
2058             }
2059         } else {
2060             dev->max_queues = 1;
2061         }
2062 
2063         if (dev->num_queues && dev->max_queues < dev->num_queues) {
2064             error_setg(errp, "The maximum number of queues supported by the "
2065                        "backend is %" PRIu64, dev->max_queues);
2066             return -EINVAL;
2067         }
2068 
2069         if (virtio_has_feature(features, VIRTIO_F_IOMMU_PLATFORM) &&
2070                 !(virtio_has_feature(dev->protocol_features,
2071                     VHOST_USER_PROTOCOL_F_BACKEND_REQ) &&
2072                  virtio_has_feature(dev->protocol_features,
2073                     VHOST_USER_PROTOCOL_F_REPLY_ACK))) {
2074             error_setg(errp, "IOMMU support requires reply-ack and "
2075                        "slave-req protocol features.");
2076             return -EINVAL;
2077         }
2078 
2079         /* get max memory regions if backend supports configurable RAM slots */
2080         if (!virtio_has_feature(dev->protocol_features,
2081                                 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS)) {
2082             u->user->memory_slots = VHOST_MEMORY_BASELINE_NREGIONS;
2083         } else {
2084             err = vhost_user_get_max_memslots(dev, &ram_slots);
2085             if (err < 0) {
2086                 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2087                 return -EPROTO;
2088             }
2089 
2090             if (ram_slots < u->user->memory_slots) {
2091                 error_setg(errp, "The backend specified a max ram slots limit "
2092                            "of %" PRIu64", when the prior validated limit was "
2093                            "%d. This limit should never decrease.", ram_slots,
2094                            u->user->memory_slots);
2095                 return -EINVAL;
2096             }
2097 
2098             u->user->memory_slots = MIN(ram_slots, VHOST_USER_MAX_RAM_SLOTS);
2099         }
2100     }
2101 
2102     if (dev->migration_blocker == NULL &&
2103         !virtio_has_feature(dev->protocol_features,
2104                             VHOST_USER_PROTOCOL_F_LOG_SHMFD)) {
2105         error_setg(&dev->migration_blocker,
2106                    "Migration disabled: vhost-user backend lacks "
2107                    "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
2108     }
2109 
2110     if (dev->vq_index == 0) {
2111         err = vhost_setup_slave_channel(dev);
2112         if (err < 0) {
2113             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2114             return -EPROTO;
2115         }
2116     }
2117 
2118     u->postcopy_notifier.notify = vhost_user_postcopy_notifier;
2119     postcopy_add_notifier(&u->postcopy_notifier);
2120 
2121     return 0;
2122 }
2123 
2124 static int vhost_user_backend_cleanup(struct vhost_dev *dev)
2125 {
2126     struct vhost_user *u;
2127 
2128     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2129 
2130     u = dev->opaque;
2131     if (u->postcopy_notifier.notify) {
2132         postcopy_remove_notifier(&u->postcopy_notifier);
2133         u->postcopy_notifier.notify = NULL;
2134     }
2135     u->postcopy_listen = false;
2136     if (u->postcopy_fd.handler) {
2137         postcopy_unregister_shared_ufd(&u->postcopy_fd);
2138         close(u->postcopy_fd.fd);
2139         u->postcopy_fd.handler = NULL;
2140     }
2141     if (u->slave_ioc) {
2142         close_slave_channel(u);
2143     }
2144     g_free(u->region_rb);
2145     u->region_rb = NULL;
2146     g_free(u->region_rb_offset);
2147     u->region_rb_offset = NULL;
2148     u->region_rb_len = 0;
2149     g_free(u);
2150     dev->opaque = 0;
2151 
2152     return 0;
2153 }
2154 
2155 static int vhost_user_get_vq_index(struct vhost_dev *dev, int idx)
2156 {
2157     assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
2158 
2159     return idx;
2160 }
2161 
2162 static int vhost_user_memslots_limit(struct vhost_dev *dev)
2163 {
2164     struct vhost_user *u = dev->opaque;
2165 
2166     return u->user->memory_slots;
2167 }
2168 
2169 static bool vhost_user_requires_shm_log(struct vhost_dev *dev)
2170 {
2171     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2172 
2173     return virtio_has_feature(dev->protocol_features,
2174                               VHOST_USER_PROTOCOL_F_LOG_SHMFD);
2175 }
2176 
2177 static int vhost_user_migration_done(struct vhost_dev *dev, char* mac_addr)
2178 {
2179     VhostUserMsg msg = { };
2180 
2181     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2182 
2183     /* If guest supports GUEST_ANNOUNCE do nothing */
2184     if (virtio_has_feature(dev->acked_features, VIRTIO_NET_F_GUEST_ANNOUNCE)) {
2185         return 0;
2186     }
2187 
2188     /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
2189     if (virtio_has_feature(dev->protocol_features,
2190                            VHOST_USER_PROTOCOL_F_RARP)) {
2191         msg.hdr.request = VHOST_USER_SEND_RARP;
2192         msg.hdr.flags = VHOST_USER_VERSION;
2193         memcpy((char *)&msg.payload.u64, mac_addr, 6);
2194         msg.hdr.size = sizeof(msg.payload.u64);
2195 
2196         return vhost_user_write(dev, &msg, NULL, 0);
2197     }
2198     return -ENOTSUP;
2199 }
2200 
2201 static bool vhost_user_can_merge(struct vhost_dev *dev,
2202                                  uint64_t start1, uint64_t size1,
2203                                  uint64_t start2, uint64_t size2)
2204 {
2205     ram_addr_t offset;
2206     int mfd, rfd;
2207 
2208     (void)vhost_user_get_mr_data(start1, &offset, &mfd);
2209     (void)vhost_user_get_mr_data(start2, &offset, &rfd);
2210 
2211     return mfd == rfd;
2212 }
2213 
2214 static int vhost_user_net_set_mtu(struct vhost_dev *dev, uint16_t mtu)
2215 {
2216     VhostUserMsg msg;
2217     bool reply_supported = virtio_has_feature(dev->protocol_features,
2218                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2219     int ret;
2220 
2221     if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU))) {
2222         return 0;
2223     }
2224 
2225     msg.hdr.request = VHOST_USER_NET_SET_MTU;
2226     msg.payload.u64 = mtu;
2227     msg.hdr.size = sizeof(msg.payload.u64);
2228     msg.hdr.flags = VHOST_USER_VERSION;
2229     if (reply_supported) {
2230         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2231     }
2232 
2233     ret = vhost_user_write(dev, &msg, NULL, 0);
2234     if (ret < 0) {
2235         return ret;
2236     }
2237 
2238     /* If reply_ack supported, slave has to ack specified MTU is valid */
2239     if (reply_supported) {
2240         return process_message_reply(dev, &msg);
2241     }
2242 
2243     return 0;
2244 }
2245 
2246 static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
2247                                             struct vhost_iotlb_msg *imsg)
2248 {
2249     int ret;
2250     VhostUserMsg msg = {
2251         .hdr.request = VHOST_USER_IOTLB_MSG,
2252         .hdr.size = sizeof(msg.payload.iotlb),
2253         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
2254         .payload.iotlb = *imsg,
2255     };
2256 
2257     ret = vhost_user_write(dev, &msg, NULL, 0);
2258     if (ret < 0) {
2259         return ret;
2260     }
2261 
2262     return process_message_reply(dev, &msg);
2263 }
2264 
2265 
2266 static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
2267 {
2268     /* No-op as the receive channel is not dedicated to IOTLB messages. */
2269 }
2270 
2271 static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
2272                                  uint32_t config_len, Error **errp)
2273 {
2274     int ret;
2275     VhostUserMsg msg = {
2276         .hdr.request = VHOST_USER_GET_CONFIG,
2277         .hdr.flags = VHOST_USER_VERSION,
2278         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + config_len,
2279     };
2280 
2281     if (!virtio_has_feature(dev->protocol_features,
2282                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2283         error_setg(errp, "VHOST_USER_PROTOCOL_F_CONFIG not supported");
2284         return -EINVAL;
2285     }
2286 
2287     assert(config_len <= VHOST_USER_MAX_CONFIG_SIZE);
2288 
2289     msg.payload.config.offset = 0;
2290     msg.payload.config.size = config_len;
2291     ret = vhost_user_write(dev, &msg, NULL, 0);
2292     if (ret < 0) {
2293         error_setg_errno(errp, -ret, "vhost_get_config failed");
2294         return ret;
2295     }
2296 
2297     ret = vhost_user_read(dev, &msg);
2298     if (ret < 0) {
2299         error_setg_errno(errp, -ret, "vhost_get_config failed");
2300         return ret;
2301     }
2302 
2303     if (msg.hdr.request != VHOST_USER_GET_CONFIG) {
2304         error_setg(errp,
2305                    "Received unexpected msg type. Expected %d received %d",
2306                    VHOST_USER_GET_CONFIG, msg.hdr.request);
2307         return -EPROTO;
2308     }
2309 
2310     if (msg.hdr.size != VHOST_USER_CONFIG_HDR_SIZE + config_len) {
2311         error_setg(errp, "Received bad msg size.");
2312         return -EPROTO;
2313     }
2314 
2315     memcpy(config, msg.payload.config.region, config_len);
2316 
2317     return 0;
2318 }
2319 
2320 static int vhost_user_set_config(struct vhost_dev *dev, const uint8_t *data,
2321                                  uint32_t offset, uint32_t size, uint32_t flags)
2322 {
2323     int ret;
2324     uint8_t *p;
2325     bool reply_supported = virtio_has_feature(dev->protocol_features,
2326                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2327 
2328     VhostUserMsg msg = {
2329         .hdr.request = VHOST_USER_SET_CONFIG,
2330         .hdr.flags = VHOST_USER_VERSION,
2331         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + size,
2332     };
2333 
2334     if (!virtio_has_feature(dev->protocol_features,
2335                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2336         return -ENOTSUP;
2337     }
2338 
2339     if (reply_supported) {
2340         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2341     }
2342 
2343     if (size > VHOST_USER_MAX_CONFIG_SIZE) {
2344         return -EINVAL;
2345     }
2346 
2347     msg.payload.config.offset = offset,
2348     msg.payload.config.size = size,
2349     msg.payload.config.flags = flags,
2350     p = msg.payload.config.region;
2351     memcpy(p, data, size);
2352 
2353     ret = vhost_user_write(dev, &msg, NULL, 0);
2354     if (ret < 0) {
2355         return ret;
2356     }
2357 
2358     if (reply_supported) {
2359         return process_message_reply(dev, &msg);
2360     }
2361 
2362     return 0;
2363 }
2364 
2365 static int vhost_user_crypto_create_session(struct vhost_dev *dev,
2366                                             void *session_info,
2367                                             uint64_t *session_id)
2368 {
2369     int ret;
2370     bool crypto_session = virtio_has_feature(dev->protocol_features,
2371                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2372     CryptoDevBackendSessionInfo *backend_info = session_info;
2373     VhostUserMsg msg = {
2374         .hdr.request = VHOST_USER_CREATE_CRYPTO_SESSION,
2375         .hdr.flags = VHOST_USER_VERSION,
2376         .hdr.size = sizeof(msg.payload.session),
2377     };
2378 
2379     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2380 
2381     if (!crypto_session) {
2382         error_report("vhost-user trying to send unhandled ioctl");
2383         return -ENOTSUP;
2384     }
2385 
2386     if (backend_info->op_code == VIRTIO_CRYPTO_AKCIPHER_CREATE_SESSION) {
2387         CryptoDevBackendAsymSessionInfo *sess = &backend_info->u.asym_sess_info;
2388         size_t keylen;
2389 
2390         memcpy(&msg.payload.session.u.asym.session_setup_data, sess,
2391                sizeof(CryptoDevBackendAsymSessionInfo));
2392         if (sess->keylen) {
2393             keylen = sizeof(msg.payload.session.u.asym.key);
2394             if (sess->keylen > keylen) {
2395                 error_report("Unsupported asymmetric key size");
2396                 return -ENOTSUP;
2397             }
2398 
2399             memcpy(&msg.payload.session.u.asym.key, sess->key,
2400                    sess->keylen);
2401         }
2402     } else {
2403         CryptoDevBackendSymSessionInfo *sess = &backend_info->u.sym_sess_info;
2404         size_t keylen;
2405 
2406         memcpy(&msg.payload.session.u.sym.session_setup_data, sess,
2407                sizeof(CryptoDevBackendSymSessionInfo));
2408         if (sess->key_len) {
2409             keylen = sizeof(msg.payload.session.u.sym.key);
2410             if (sess->key_len > keylen) {
2411                 error_report("Unsupported cipher key size");
2412                 return -ENOTSUP;
2413             }
2414 
2415             memcpy(&msg.payload.session.u.sym.key, sess->cipher_key,
2416                    sess->key_len);
2417         }
2418 
2419         if (sess->auth_key_len > 0) {
2420             keylen = sizeof(msg.payload.session.u.sym.auth_key);
2421             if (sess->auth_key_len > keylen) {
2422                 error_report("Unsupported auth key size");
2423                 return -ENOTSUP;
2424             }
2425 
2426             memcpy(&msg.payload.session.u.sym.auth_key, sess->auth_key,
2427                    sess->auth_key_len);
2428         }
2429     }
2430 
2431     msg.payload.session.op_code = backend_info->op_code;
2432     msg.payload.session.session_id = backend_info->session_id;
2433     ret = vhost_user_write(dev, &msg, NULL, 0);
2434     if (ret < 0) {
2435         error_report("vhost_user_write() return %d, create session failed",
2436                      ret);
2437         return ret;
2438     }
2439 
2440     ret = vhost_user_read(dev, &msg);
2441     if (ret < 0) {
2442         error_report("vhost_user_read() return %d, create session failed",
2443                      ret);
2444         return ret;
2445     }
2446 
2447     if (msg.hdr.request != VHOST_USER_CREATE_CRYPTO_SESSION) {
2448         error_report("Received unexpected msg type. Expected %d received %d",
2449                      VHOST_USER_CREATE_CRYPTO_SESSION, msg.hdr.request);
2450         return -EPROTO;
2451     }
2452 
2453     if (msg.hdr.size != sizeof(msg.payload.session)) {
2454         error_report("Received bad msg size.");
2455         return -EPROTO;
2456     }
2457 
2458     if (msg.payload.session.session_id < 0) {
2459         error_report("Bad session id: %" PRId64 "",
2460                               msg.payload.session.session_id);
2461         return -EINVAL;
2462     }
2463     *session_id = msg.payload.session.session_id;
2464 
2465     return 0;
2466 }
2467 
2468 static int
2469 vhost_user_crypto_close_session(struct vhost_dev *dev, uint64_t session_id)
2470 {
2471     int ret;
2472     bool crypto_session = virtio_has_feature(dev->protocol_features,
2473                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2474     VhostUserMsg msg = {
2475         .hdr.request = VHOST_USER_CLOSE_CRYPTO_SESSION,
2476         .hdr.flags = VHOST_USER_VERSION,
2477         .hdr.size = sizeof(msg.payload.u64),
2478     };
2479     msg.payload.u64 = session_id;
2480 
2481     if (!crypto_session) {
2482         error_report("vhost-user trying to send unhandled ioctl");
2483         return -ENOTSUP;
2484     }
2485 
2486     ret = vhost_user_write(dev, &msg, NULL, 0);
2487     if (ret < 0) {
2488         error_report("vhost_user_write() return %d, close session failed",
2489                      ret);
2490         return ret;
2491     }
2492 
2493     return 0;
2494 }
2495 
2496 static bool vhost_user_mem_section_filter(struct vhost_dev *dev,
2497                                           MemoryRegionSection *section)
2498 {
2499     return memory_region_get_fd(section->mr) >= 0;
2500 }
2501 
2502 static int vhost_user_get_inflight_fd(struct vhost_dev *dev,
2503                                       uint16_t queue_size,
2504                                       struct vhost_inflight *inflight)
2505 {
2506     void *addr;
2507     int fd;
2508     int ret;
2509     struct vhost_user *u = dev->opaque;
2510     CharBackend *chr = u->user->chr;
2511     VhostUserMsg msg = {
2512         .hdr.request = VHOST_USER_GET_INFLIGHT_FD,
2513         .hdr.flags = VHOST_USER_VERSION,
2514         .payload.inflight.num_queues = dev->nvqs,
2515         .payload.inflight.queue_size = queue_size,
2516         .hdr.size = sizeof(msg.payload.inflight),
2517     };
2518 
2519     if (!virtio_has_feature(dev->protocol_features,
2520                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2521         return 0;
2522     }
2523 
2524     ret = vhost_user_write(dev, &msg, NULL, 0);
2525     if (ret < 0) {
2526         return ret;
2527     }
2528 
2529     ret = vhost_user_read(dev, &msg);
2530     if (ret < 0) {
2531         return ret;
2532     }
2533 
2534     if (msg.hdr.request != VHOST_USER_GET_INFLIGHT_FD) {
2535         error_report("Received unexpected msg type. "
2536                      "Expected %d received %d",
2537                      VHOST_USER_GET_INFLIGHT_FD, msg.hdr.request);
2538         return -EPROTO;
2539     }
2540 
2541     if (msg.hdr.size != sizeof(msg.payload.inflight)) {
2542         error_report("Received bad msg size.");
2543         return -EPROTO;
2544     }
2545 
2546     if (!msg.payload.inflight.mmap_size) {
2547         return 0;
2548     }
2549 
2550     fd = qemu_chr_fe_get_msgfd(chr);
2551     if (fd < 0) {
2552         error_report("Failed to get mem fd");
2553         return -EIO;
2554     }
2555 
2556     addr = mmap(0, msg.payload.inflight.mmap_size, PROT_READ | PROT_WRITE,
2557                 MAP_SHARED, fd, msg.payload.inflight.mmap_offset);
2558 
2559     if (addr == MAP_FAILED) {
2560         error_report("Failed to mmap mem fd");
2561         close(fd);
2562         return -EFAULT;
2563     }
2564 
2565     inflight->addr = addr;
2566     inflight->fd = fd;
2567     inflight->size = msg.payload.inflight.mmap_size;
2568     inflight->offset = msg.payload.inflight.mmap_offset;
2569     inflight->queue_size = queue_size;
2570 
2571     return 0;
2572 }
2573 
2574 static int vhost_user_set_inflight_fd(struct vhost_dev *dev,
2575                                       struct vhost_inflight *inflight)
2576 {
2577     VhostUserMsg msg = {
2578         .hdr.request = VHOST_USER_SET_INFLIGHT_FD,
2579         .hdr.flags = VHOST_USER_VERSION,
2580         .payload.inflight.mmap_size = inflight->size,
2581         .payload.inflight.mmap_offset = inflight->offset,
2582         .payload.inflight.num_queues = dev->nvqs,
2583         .payload.inflight.queue_size = inflight->queue_size,
2584         .hdr.size = sizeof(msg.payload.inflight),
2585     };
2586 
2587     if (!virtio_has_feature(dev->protocol_features,
2588                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2589         return 0;
2590     }
2591 
2592     return vhost_user_write(dev, &msg, &inflight->fd, 1);
2593 }
2594 
2595 static void vhost_user_state_destroy(gpointer data)
2596 {
2597     VhostUserHostNotifier *n = (VhostUserHostNotifier *) data;
2598     if (n) {
2599         vhost_user_host_notifier_remove(n, NULL);
2600         object_unparent(OBJECT(&n->mr));
2601         /*
2602          * We can't free until vhost_user_host_notifier_remove has
2603          * done it's thing so schedule the free with RCU.
2604          */
2605         g_free_rcu(n, rcu);
2606     }
2607 }
2608 
2609 bool vhost_user_init(VhostUserState *user, CharBackend *chr, Error **errp)
2610 {
2611     if (user->chr) {
2612         error_setg(errp, "Cannot initialize vhost-user state");
2613         return false;
2614     }
2615     user->chr = chr;
2616     user->memory_slots = 0;
2617     user->notifiers = g_ptr_array_new_full(VIRTIO_QUEUE_MAX / 4,
2618                                            &vhost_user_state_destroy);
2619     return true;
2620 }
2621 
2622 void vhost_user_cleanup(VhostUserState *user)
2623 {
2624     if (!user->chr) {
2625         return;
2626     }
2627     memory_region_transaction_begin();
2628     user->notifiers = (GPtrArray *) g_ptr_array_free(user->notifiers, true);
2629     memory_region_transaction_commit();
2630     user->chr = NULL;
2631 }
2632 
2633 
2634 typedef struct {
2635     vu_async_close_fn cb;
2636     DeviceState *dev;
2637     CharBackend *cd;
2638     struct vhost_dev *vhost;
2639 } VhostAsyncCallback;
2640 
2641 static void vhost_user_async_close_bh(void *opaque)
2642 {
2643     VhostAsyncCallback *data = opaque;
2644     struct vhost_dev *vhost = data->vhost;
2645 
2646     /*
2647      * If the vhost_dev has been cleared in the meantime there is
2648      * nothing left to do as some other path has completed the
2649      * cleanup.
2650      */
2651     if (vhost->vdev) {
2652         data->cb(data->dev);
2653     }
2654 
2655     g_free(data);
2656 }
2657 
2658 /*
2659  * We only schedule the work if the machine is running. If suspended
2660  * we want to keep all the in-flight data as is for migration
2661  * purposes.
2662  */
2663 void vhost_user_async_close(DeviceState *d,
2664                             CharBackend *chardev, struct vhost_dev *vhost,
2665                             vu_async_close_fn cb)
2666 {
2667     if (!runstate_check(RUN_STATE_SHUTDOWN)) {
2668         /*
2669          * A close event may happen during a read/write, but vhost
2670          * code assumes the vhost_dev remains setup, so delay the
2671          * stop & clear.
2672          */
2673         AioContext *ctx = qemu_get_current_aio_context();
2674         VhostAsyncCallback *data = g_new0(VhostAsyncCallback, 1);
2675 
2676         /* Save data for the callback */
2677         data->cb = cb;
2678         data->dev = d;
2679         data->cd = chardev;
2680         data->vhost = vhost;
2681 
2682         /* Disable any further notifications on the chardev */
2683         qemu_chr_fe_set_handlers(chardev,
2684                                  NULL, NULL, NULL, NULL, NULL, NULL,
2685                                  false);
2686 
2687         aio_bh_schedule_oneshot(ctx, vhost_user_async_close_bh, data);
2688 
2689         /*
2690          * Move vhost device to the stopped state. The vhost-user device
2691          * will be clean up and disconnected in BH. This can be useful in
2692          * the vhost migration code. If disconnect was caught there is an
2693          * option for the general vhost code to get the dev state without
2694          * knowing its type (in this case vhost-user).
2695          *
2696          * Note if the vhost device is fully cleared by the time we
2697          * execute the bottom half we won't continue with the cleanup.
2698          */
2699         vhost->started = false;
2700     }
2701 }
2702 
2703 static int vhost_user_dev_start(struct vhost_dev *dev, bool started)
2704 {
2705     if (!virtio_has_feature(dev->protocol_features,
2706                             VHOST_USER_PROTOCOL_F_STATUS)) {
2707         return 0;
2708     }
2709 
2710     /* Set device status only for last queue pair */
2711     if (dev->vq_index + dev->nvqs != dev->vq_index_end) {
2712         return 0;
2713     }
2714 
2715     if (started) {
2716         return vhost_user_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE |
2717                                           VIRTIO_CONFIG_S_DRIVER |
2718                                           VIRTIO_CONFIG_S_DRIVER_OK);
2719     } else {
2720         return 0;
2721     }
2722 }
2723 
2724 static void vhost_user_reset_status(struct vhost_dev *dev)
2725 {
2726     /* Set device status only for last queue pair */
2727     if (dev->vq_index + dev->nvqs != dev->vq_index_end) {
2728         return;
2729     }
2730 
2731     if (virtio_has_feature(dev->protocol_features,
2732                            VHOST_USER_PROTOCOL_F_STATUS)) {
2733         vhost_user_set_status(dev, 0);
2734     }
2735 }
2736 
2737 const VhostOps user_ops = {
2738         .backend_type = VHOST_BACKEND_TYPE_USER,
2739         .vhost_backend_init = vhost_user_backend_init,
2740         .vhost_backend_cleanup = vhost_user_backend_cleanup,
2741         .vhost_backend_memslots_limit = vhost_user_memslots_limit,
2742         .vhost_set_log_base = vhost_user_set_log_base,
2743         .vhost_set_mem_table = vhost_user_set_mem_table,
2744         .vhost_set_vring_addr = vhost_user_set_vring_addr,
2745         .vhost_set_vring_endian = vhost_user_set_vring_endian,
2746         .vhost_set_vring_num = vhost_user_set_vring_num,
2747         .vhost_set_vring_base = vhost_user_set_vring_base,
2748         .vhost_get_vring_base = vhost_user_get_vring_base,
2749         .vhost_set_vring_kick = vhost_user_set_vring_kick,
2750         .vhost_set_vring_call = vhost_user_set_vring_call,
2751         .vhost_set_vring_err = vhost_user_set_vring_err,
2752         .vhost_set_features = vhost_user_set_features,
2753         .vhost_get_features = vhost_user_get_features,
2754         .vhost_set_owner = vhost_user_set_owner,
2755         .vhost_reset_device = vhost_user_reset_device,
2756         .vhost_get_vq_index = vhost_user_get_vq_index,
2757         .vhost_set_vring_enable = vhost_user_set_vring_enable,
2758         .vhost_requires_shm_log = vhost_user_requires_shm_log,
2759         .vhost_migration_done = vhost_user_migration_done,
2760         .vhost_backend_can_merge = vhost_user_can_merge,
2761         .vhost_net_set_mtu = vhost_user_net_set_mtu,
2762         .vhost_set_iotlb_callback = vhost_user_set_iotlb_callback,
2763         .vhost_send_device_iotlb_msg = vhost_user_send_device_iotlb_msg,
2764         .vhost_get_config = vhost_user_get_config,
2765         .vhost_set_config = vhost_user_set_config,
2766         .vhost_crypto_create_session = vhost_user_crypto_create_session,
2767         .vhost_crypto_close_session = vhost_user_crypto_close_session,
2768         .vhost_backend_mem_section_filter = vhost_user_mem_section_filter,
2769         .vhost_get_inflight_fd = vhost_user_get_inflight_fd,
2770         .vhost_set_inflight_fd = vhost_user_set_inflight_fd,
2771         .vhost_dev_start = vhost_user_dev_start,
2772         .vhost_reset_status = vhost_user_reset_status,
2773 };
2774