xref: /openbmc/qemu/hw/virtio/vhost-user.c (revision 6c1ebe75)
1 /*
2  * vhost-user
3  *
4  * Copyright (c) 2013 Virtual Open Systems Sarl.
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
7  * See the COPYING file in the top-level directory.
8  *
9  */
10 
11 #include "qemu/osdep.h"
12 #include "qapi/error.h"
13 #include "hw/virtio/vhost.h"
14 #include "hw/virtio/vhost-user.h"
15 #include "hw/virtio/vhost-backend.h"
16 #include "hw/virtio/virtio.h"
17 #include "hw/virtio/virtio-net.h"
18 #include "chardev/char-fe.h"
19 #include "io/channel-socket.h"
20 #include "sysemu/kvm.h"
21 #include "qemu/error-report.h"
22 #include "qemu/main-loop.h"
23 #include "qemu/sockets.h"
24 #include "sysemu/cryptodev.h"
25 #include "migration/migration.h"
26 #include "migration/postcopy-ram.h"
27 #include "trace.h"
28 
29 #include <sys/ioctl.h>
30 #include <sys/socket.h>
31 #include <sys/un.h>
32 
33 #include "standard-headers/linux/vhost_types.h"
34 
35 #ifdef CONFIG_LINUX
36 #include <linux/userfaultfd.h>
37 #endif
38 
39 #define VHOST_MEMORY_BASELINE_NREGIONS    8
40 #define VHOST_USER_F_PROTOCOL_FEATURES 30
41 #define VHOST_USER_SLAVE_MAX_FDS     8
42 
43 /*
44  * Set maximum number of RAM slots supported to
45  * the maximum number supported by the target
46  * hardware plaform.
47  */
48 #if defined(TARGET_X86) || defined(TARGET_X86_64) || \
49     defined(TARGET_ARM) || defined(TARGET_ARM_64)
50 #include "hw/acpi/acpi.h"
51 #define VHOST_USER_MAX_RAM_SLOTS ACPI_MAX_RAM_SLOTS
52 
53 #elif defined(TARGET_PPC) || defined(TARGET_PPC_64)
54 #include "hw/ppc/spapr.h"
55 #define VHOST_USER_MAX_RAM_SLOTS SPAPR_MAX_RAM_SLOTS
56 
57 #else
58 #define VHOST_USER_MAX_RAM_SLOTS 512
59 #endif
60 
61 /*
62  * Maximum size of virtio device config space
63  */
64 #define VHOST_USER_MAX_CONFIG_SIZE 256
65 
66 enum VhostUserProtocolFeature {
67     VHOST_USER_PROTOCOL_F_MQ = 0,
68     VHOST_USER_PROTOCOL_F_LOG_SHMFD = 1,
69     VHOST_USER_PROTOCOL_F_RARP = 2,
70     VHOST_USER_PROTOCOL_F_REPLY_ACK = 3,
71     VHOST_USER_PROTOCOL_F_NET_MTU = 4,
72     VHOST_USER_PROTOCOL_F_SLAVE_REQ = 5,
73     VHOST_USER_PROTOCOL_F_CROSS_ENDIAN = 6,
74     VHOST_USER_PROTOCOL_F_CRYPTO_SESSION = 7,
75     VHOST_USER_PROTOCOL_F_PAGEFAULT = 8,
76     VHOST_USER_PROTOCOL_F_CONFIG = 9,
77     VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD = 10,
78     VHOST_USER_PROTOCOL_F_HOST_NOTIFIER = 11,
79     VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD = 12,
80     VHOST_USER_PROTOCOL_F_RESET_DEVICE = 13,
81     /* Feature 14 reserved for VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS. */
82     VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS = 15,
83     VHOST_USER_PROTOCOL_F_MAX
84 };
85 
86 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
87 
88 typedef enum VhostUserRequest {
89     VHOST_USER_NONE = 0,
90     VHOST_USER_GET_FEATURES = 1,
91     VHOST_USER_SET_FEATURES = 2,
92     VHOST_USER_SET_OWNER = 3,
93     VHOST_USER_RESET_OWNER = 4,
94     VHOST_USER_SET_MEM_TABLE = 5,
95     VHOST_USER_SET_LOG_BASE = 6,
96     VHOST_USER_SET_LOG_FD = 7,
97     VHOST_USER_SET_VRING_NUM = 8,
98     VHOST_USER_SET_VRING_ADDR = 9,
99     VHOST_USER_SET_VRING_BASE = 10,
100     VHOST_USER_GET_VRING_BASE = 11,
101     VHOST_USER_SET_VRING_KICK = 12,
102     VHOST_USER_SET_VRING_CALL = 13,
103     VHOST_USER_SET_VRING_ERR = 14,
104     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
105     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
106     VHOST_USER_GET_QUEUE_NUM = 17,
107     VHOST_USER_SET_VRING_ENABLE = 18,
108     VHOST_USER_SEND_RARP = 19,
109     VHOST_USER_NET_SET_MTU = 20,
110     VHOST_USER_SET_SLAVE_REQ_FD = 21,
111     VHOST_USER_IOTLB_MSG = 22,
112     VHOST_USER_SET_VRING_ENDIAN = 23,
113     VHOST_USER_GET_CONFIG = 24,
114     VHOST_USER_SET_CONFIG = 25,
115     VHOST_USER_CREATE_CRYPTO_SESSION = 26,
116     VHOST_USER_CLOSE_CRYPTO_SESSION = 27,
117     VHOST_USER_POSTCOPY_ADVISE  = 28,
118     VHOST_USER_POSTCOPY_LISTEN  = 29,
119     VHOST_USER_POSTCOPY_END     = 30,
120     VHOST_USER_GET_INFLIGHT_FD = 31,
121     VHOST_USER_SET_INFLIGHT_FD = 32,
122     VHOST_USER_GPU_SET_SOCKET = 33,
123     VHOST_USER_RESET_DEVICE = 34,
124     /* Message number 35 reserved for VHOST_USER_VRING_KICK. */
125     VHOST_USER_GET_MAX_MEM_SLOTS = 36,
126     VHOST_USER_ADD_MEM_REG = 37,
127     VHOST_USER_REM_MEM_REG = 38,
128     VHOST_USER_MAX
129 } VhostUserRequest;
130 
131 typedef enum VhostUserSlaveRequest {
132     VHOST_USER_SLAVE_NONE = 0,
133     VHOST_USER_SLAVE_IOTLB_MSG = 1,
134     VHOST_USER_SLAVE_CONFIG_CHANGE_MSG = 2,
135     VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG = 3,
136     VHOST_USER_SLAVE_MAX
137 }  VhostUserSlaveRequest;
138 
139 typedef struct VhostUserMemoryRegion {
140     uint64_t guest_phys_addr;
141     uint64_t memory_size;
142     uint64_t userspace_addr;
143     uint64_t mmap_offset;
144 } VhostUserMemoryRegion;
145 
146 typedef struct VhostUserMemory {
147     uint32_t nregions;
148     uint32_t padding;
149     VhostUserMemoryRegion regions[VHOST_MEMORY_BASELINE_NREGIONS];
150 } VhostUserMemory;
151 
152 typedef struct VhostUserMemRegMsg {
153     uint64_t padding;
154     VhostUserMemoryRegion region;
155 } VhostUserMemRegMsg;
156 
157 typedef struct VhostUserLog {
158     uint64_t mmap_size;
159     uint64_t mmap_offset;
160 } VhostUserLog;
161 
162 typedef struct VhostUserConfig {
163     uint32_t offset;
164     uint32_t size;
165     uint32_t flags;
166     uint8_t region[VHOST_USER_MAX_CONFIG_SIZE];
167 } VhostUserConfig;
168 
169 #define VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN    512
170 #define VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN  64
171 
172 typedef struct VhostUserCryptoSession {
173     /* session id for success, -1 on errors */
174     int64_t session_id;
175     CryptoDevBackendSymSessionInfo session_setup_data;
176     uint8_t key[VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN];
177     uint8_t auth_key[VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN];
178 } VhostUserCryptoSession;
179 
180 static VhostUserConfig c __attribute__ ((unused));
181 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
182                                    + sizeof(c.size) \
183                                    + sizeof(c.flags))
184 
185 typedef struct VhostUserVringArea {
186     uint64_t u64;
187     uint64_t size;
188     uint64_t offset;
189 } VhostUserVringArea;
190 
191 typedef struct VhostUserInflight {
192     uint64_t mmap_size;
193     uint64_t mmap_offset;
194     uint16_t num_queues;
195     uint16_t queue_size;
196 } VhostUserInflight;
197 
198 typedef struct {
199     VhostUserRequest request;
200 
201 #define VHOST_USER_VERSION_MASK     (0x3)
202 #define VHOST_USER_REPLY_MASK       (0x1<<2)
203 #define VHOST_USER_NEED_REPLY_MASK  (0x1 << 3)
204     uint32_t flags;
205     uint32_t size; /* the following payload size */
206 } QEMU_PACKED VhostUserHeader;
207 
208 typedef union {
209 #define VHOST_USER_VRING_IDX_MASK   (0xff)
210 #define VHOST_USER_VRING_NOFD_MASK  (0x1<<8)
211         uint64_t u64;
212         struct vhost_vring_state state;
213         struct vhost_vring_addr addr;
214         VhostUserMemory memory;
215         VhostUserMemRegMsg mem_reg;
216         VhostUserLog log;
217         struct vhost_iotlb_msg iotlb;
218         VhostUserConfig config;
219         VhostUserCryptoSession session;
220         VhostUserVringArea area;
221         VhostUserInflight inflight;
222 } VhostUserPayload;
223 
224 typedef struct VhostUserMsg {
225     VhostUserHeader hdr;
226     VhostUserPayload payload;
227 } QEMU_PACKED VhostUserMsg;
228 
229 static VhostUserMsg m __attribute__ ((unused));
230 #define VHOST_USER_HDR_SIZE (sizeof(VhostUserHeader))
231 
232 #define VHOST_USER_PAYLOAD_SIZE (sizeof(VhostUserPayload))
233 
234 /* The version of the protocol we support */
235 #define VHOST_USER_VERSION    (0x1)
236 
237 struct vhost_user {
238     struct vhost_dev *dev;
239     /* Shared between vhost devs of the same virtio device */
240     VhostUserState *user;
241     QIOChannel *slave_ioc;
242     GSource *slave_src;
243     NotifierWithReturn postcopy_notifier;
244     struct PostCopyFD  postcopy_fd;
245     uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
246     /* Length of the region_rb and region_rb_offset arrays */
247     size_t             region_rb_len;
248     /* RAMBlock associated with a given region */
249     RAMBlock         **region_rb;
250     /* The offset from the start of the RAMBlock to the start of the
251      * vhost region.
252      */
253     ram_addr_t        *region_rb_offset;
254 
255     /* True once we've entered postcopy_listen */
256     bool               postcopy_listen;
257 
258     /* Our current regions */
259     int num_shadow_regions;
260     struct vhost_memory_region shadow_regions[VHOST_USER_MAX_RAM_SLOTS];
261 };
262 
263 struct scrub_regions {
264     struct vhost_memory_region *region;
265     int reg_idx;
266     int fd_idx;
267 };
268 
269 static bool ioeventfd_enabled(void)
270 {
271     return !kvm_enabled() || kvm_eventfds_enabled();
272 }
273 
274 static int vhost_user_read_header(struct vhost_dev *dev, VhostUserMsg *msg)
275 {
276     struct vhost_user *u = dev->opaque;
277     CharBackend *chr = u->user->chr;
278     uint8_t *p = (uint8_t *) msg;
279     int r, size = VHOST_USER_HDR_SIZE;
280 
281     r = qemu_chr_fe_read_all(chr, p, size);
282     if (r != size) {
283         error_report("Failed to read msg header. Read %d instead of %d."
284                      " Original request %d.", r, size, msg->hdr.request);
285         return -1;
286     }
287 
288     /* validate received flags */
289     if (msg->hdr.flags != (VHOST_USER_REPLY_MASK | VHOST_USER_VERSION)) {
290         error_report("Failed to read msg header."
291                 " Flags 0x%x instead of 0x%x.", msg->hdr.flags,
292                 VHOST_USER_REPLY_MASK | VHOST_USER_VERSION);
293         return -1;
294     }
295 
296     return 0;
297 }
298 
299 struct vhost_user_read_cb_data {
300     struct vhost_dev *dev;
301     VhostUserMsg *msg;
302     GMainLoop *loop;
303     int ret;
304 };
305 
306 static gboolean vhost_user_read_cb(void *do_not_use, GIOCondition condition,
307                                    gpointer opaque)
308 {
309     struct vhost_user_read_cb_data *data = opaque;
310     struct vhost_dev *dev = data->dev;
311     VhostUserMsg *msg = data->msg;
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     if (vhost_user_read_header(dev, msg) < 0) {
318         data->ret = -1;
319         goto end;
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         data->ret = -1;
328         goto end;
329     }
330 
331     if (msg->hdr.size) {
332         p += VHOST_USER_HDR_SIZE;
333         size = msg->hdr.size;
334         r = qemu_chr_fe_read_all(chr, p, size);
335         if (r != size) {
336             error_report("Failed to read msg payload."
337                          " Read %d instead of %d.", r, msg->hdr.size);
338             data->ret = -1;
339             goto end;
340         }
341     }
342 
343 end:
344     g_main_loop_quit(data->loop);
345     return G_SOURCE_REMOVE;
346 }
347 
348 static gboolean slave_read(QIOChannel *ioc, GIOCondition condition,
349                            gpointer opaque);
350 
351 /*
352  * This updates the read handler to use a new event loop context.
353  * Event sources are removed from the previous context : this ensures
354  * that events detected in the previous context are purged. They will
355  * be re-detected and processed in the new context.
356  */
357 static void slave_update_read_handler(struct vhost_dev *dev,
358                                       GMainContext *ctxt)
359 {
360     struct vhost_user *u = dev->opaque;
361 
362     if (!u->slave_ioc) {
363         return;
364     }
365 
366     if (u->slave_src) {
367         g_source_destroy(u->slave_src);
368         g_source_unref(u->slave_src);
369     }
370 
371     u->slave_src = qio_channel_add_watch_source(u->slave_ioc,
372                                                 G_IO_IN | G_IO_HUP,
373                                                 slave_read, dev, NULL,
374                                                 ctxt);
375 }
376 
377 static int vhost_user_read(struct vhost_dev *dev, VhostUserMsg *msg)
378 {
379     struct vhost_user *u = dev->opaque;
380     CharBackend *chr = u->user->chr;
381     GMainContext *prev_ctxt = chr->chr->gcontext;
382     GMainContext *ctxt = g_main_context_new();
383     GMainLoop *loop = g_main_loop_new(ctxt, FALSE);
384     struct vhost_user_read_cb_data data = {
385         .dev = dev,
386         .loop = loop,
387         .msg = msg,
388         .ret = 0
389     };
390 
391     /*
392      * We want to be able to monitor the slave channel fd while waiting
393      * for chr I/O. This requires an event loop, but we can't nest the
394      * one to which chr is currently attached : its fd handlers might not
395      * be prepared for re-entrancy. So we create a new one and switch chr
396      * to use it.
397      */
398     slave_update_read_handler(dev, ctxt);
399     qemu_chr_be_update_read_handlers(chr->chr, ctxt);
400     qemu_chr_fe_add_watch(chr, G_IO_IN | G_IO_HUP, vhost_user_read_cb, &data);
401 
402     g_main_loop_run(loop);
403 
404     /*
405      * Restore the previous event loop context. This also destroys/recreates
406      * event sources : this guarantees that all pending events in the original
407      * context that have been processed by the nested loop are purged.
408      */
409     qemu_chr_be_update_read_handlers(chr->chr, prev_ctxt);
410     slave_update_read_handler(dev, NULL);
411 
412     g_main_loop_unref(loop);
413     g_main_context_unref(ctxt);
414 
415     return data.ret;
416 }
417 
418 static int process_message_reply(struct vhost_dev *dev,
419                                  const VhostUserMsg *msg)
420 {
421     VhostUserMsg msg_reply;
422 
423     if ((msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) == 0) {
424         return 0;
425     }
426 
427     if (vhost_user_read(dev, &msg_reply) < 0) {
428         return -1;
429     }
430 
431     if (msg_reply.hdr.request != msg->hdr.request) {
432         error_report("Received unexpected msg type. "
433                      "Expected %d received %d",
434                      msg->hdr.request, msg_reply.hdr.request);
435         return -1;
436     }
437 
438     return msg_reply.payload.u64 ? -1 : 0;
439 }
440 
441 static bool vhost_user_one_time_request(VhostUserRequest request)
442 {
443     switch (request) {
444     case VHOST_USER_SET_OWNER:
445     case VHOST_USER_RESET_OWNER:
446     case VHOST_USER_SET_MEM_TABLE:
447     case VHOST_USER_GET_QUEUE_NUM:
448     case VHOST_USER_NET_SET_MTU:
449         return true;
450     default:
451         return false;
452     }
453 }
454 
455 /* most non-init callers ignore the error */
456 static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
457                             int *fds, int fd_num)
458 {
459     struct vhost_user *u = dev->opaque;
460     CharBackend *chr = u->user->chr;
461     int ret, size = VHOST_USER_HDR_SIZE + msg->hdr.size;
462 
463     /*
464      * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
465      * we just need send it once in the first time. For later such
466      * request, we just ignore it.
467      */
468     if (vhost_user_one_time_request(msg->hdr.request) && dev->vq_index != 0) {
469         msg->hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
470         return 0;
471     }
472 
473     if (qemu_chr_fe_set_msgfds(chr, fds, fd_num) < 0) {
474         error_report("Failed to set msg fds.");
475         return -1;
476     }
477 
478     ret = qemu_chr_fe_write_all(chr, (const uint8_t *) msg, size);
479     if (ret != size) {
480         error_report("Failed to write msg."
481                      " Wrote %d instead of %d.", ret, size);
482         return -1;
483     }
484 
485     return 0;
486 }
487 
488 int vhost_user_gpu_set_socket(struct vhost_dev *dev, int fd)
489 {
490     VhostUserMsg msg = {
491         .hdr.request = VHOST_USER_GPU_SET_SOCKET,
492         .hdr.flags = VHOST_USER_VERSION,
493     };
494 
495     return vhost_user_write(dev, &msg, &fd, 1);
496 }
497 
498 static int vhost_user_set_log_base(struct vhost_dev *dev, uint64_t base,
499                                    struct vhost_log *log)
500 {
501     int fds[VHOST_USER_MAX_RAM_SLOTS];
502     size_t fd_num = 0;
503     bool shmfd = virtio_has_feature(dev->protocol_features,
504                                     VHOST_USER_PROTOCOL_F_LOG_SHMFD);
505     VhostUserMsg msg = {
506         .hdr.request = VHOST_USER_SET_LOG_BASE,
507         .hdr.flags = VHOST_USER_VERSION,
508         .payload.log.mmap_size = log->size * sizeof(*(log->log)),
509         .payload.log.mmap_offset = 0,
510         .hdr.size = sizeof(msg.payload.log),
511     };
512 
513     if (shmfd && log->fd != -1) {
514         fds[fd_num++] = log->fd;
515     }
516 
517     if (vhost_user_write(dev, &msg, fds, fd_num) < 0) {
518         return -1;
519     }
520 
521     if (shmfd) {
522         msg.hdr.size = 0;
523         if (vhost_user_read(dev, &msg) < 0) {
524             return -1;
525         }
526 
527         if (msg.hdr.request != VHOST_USER_SET_LOG_BASE) {
528             error_report("Received unexpected msg type. "
529                          "Expected %d received %d",
530                          VHOST_USER_SET_LOG_BASE, msg.hdr.request);
531             return -1;
532         }
533     }
534 
535     return 0;
536 }
537 
538 static MemoryRegion *vhost_user_get_mr_data(uint64_t addr, ram_addr_t *offset,
539                                             int *fd)
540 {
541     MemoryRegion *mr;
542 
543     assert((uintptr_t)addr == addr);
544     mr = memory_region_from_host((void *)(uintptr_t)addr, offset);
545     *fd = memory_region_get_fd(mr);
546 
547     return mr;
548 }
549 
550 static void vhost_user_fill_msg_region(VhostUserMemoryRegion *dst,
551                                        struct vhost_memory_region *src,
552                                        uint64_t mmap_offset)
553 {
554     assert(src != NULL && dst != NULL);
555     dst->userspace_addr = src->userspace_addr;
556     dst->memory_size = src->memory_size;
557     dst->guest_phys_addr = src->guest_phys_addr;
558     dst->mmap_offset = mmap_offset;
559 }
560 
561 static int vhost_user_fill_set_mem_table_msg(struct vhost_user *u,
562                                              struct vhost_dev *dev,
563                                              VhostUserMsg *msg,
564                                              int *fds, size_t *fd_num,
565                                              bool track_ramblocks)
566 {
567     int i, fd;
568     ram_addr_t offset;
569     MemoryRegion *mr;
570     struct vhost_memory_region *reg;
571     VhostUserMemoryRegion region_buffer;
572 
573     msg->hdr.request = VHOST_USER_SET_MEM_TABLE;
574 
575     for (i = 0; i < dev->mem->nregions; ++i) {
576         reg = dev->mem->regions + i;
577 
578         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
579         if (fd > 0) {
580             if (track_ramblocks) {
581                 assert(*fd_num < VHOST_MEMORY_BASELINE_NREGIONS);
582                 trace_vhost_user_set_mem_table_withfd(*fd_num, mr->name,
583                                                       reg->memory_size,
584                                                       reg->guest_phys_addr,
585                                                       reg->userspace_addr,
586                                                       offset);
587                 u->region_rb_offset[i] = offset;
588                 u->region_rb[i] = mr->ram_block;
589             } else if (*fd_num == VHOST_MEMORY_BASELINE_NREGIONS) {
590                 error_report("Failed preparing vhost-user memory table msg");
591                 return -1;
592             }
593             vhost_user_fill_msg_region(&region_buffer, reg, offset);
594             msg->payload.memory.regions[*fd_num] = region_buffer;
595             fds[(*fd_num)++] = fd;
596         } else if (track_ramblocks) {
597             u->region_rb_offset[i] = 0;
598             u->region_rb[i] = NULL;
599         }
600     }
601 
602     msg->payload.memory.nregions = *fd_num;
603 
604     if (!*fd_num) {
605         error_report("Failed initializing vhost-user memory map, "
606                      "consider using -object memory-backend-file share=on");
607         return -1;
608     }
609 
610     msg->hdr.size = sizeof(msg->payload.memory.nregions);
611     msg->hdr.size += sizeof(msg->payload.memory.padding);
612     msg->hdr.size += *fd_num * sizeof(VhostUserMemoryRegion);
613 
614     return 1;
615 }
616 
617 static inline bool reg_equal(struct vhost_memory_region *shadow_reg,
618                              struct vhost_memory_region *vdev_reg)
619 {
620     return shadow_reg->guest_phys_addr == vdev_reg->guest_phys_addr &&
621         shadow_reg->userspace_addr == vdev_reg->userspace_addr &&
622         shadow_reg->memory_size == vdev_reg->memory_size;
623 }
624 
625 static void scrub_shadow_regions(struct vhost_dev *dev,
626                                  struct scrub_regions *add_reg,
627                                  int *nr_add_reg,
628                                  struct scrub_regions *rem_reg,
629                                  int *nr_rem_reg, uint64_t *shadow_pcb,
630                                  bool track_ramblocks)
631 {
632     struct vhost_user *u = dev->opaque;
633     bool found[VHOST_USER_MAX_RAM_SLOTS] = {};
634     struct vhost_memory_region *reg, *shadow_reg;
635     int i, j, fd, add_idx = 0, rm_idx = 0, fd_num = 0;
636     ram_addr_t offset;
637     MemoryRegion *mr;
638     bool matching;
639 
640     /*
641      * Find memory regions present in our shadow state which are not in
642      * the device's current memory state.
643      *
644      * Mark regions in both the shadow and device state as "found".
645      */
646     for (i = 0; i < u->num_shadow_regions; i++) {
647         shadow_reg = &u->shadow_regions[i];
648         matching = false;
649 
650         for (j = 0; j < dev->mem->nregions; j++) {
651             reg = &dev->mem->regions[j];
652 
653             mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
654 
655             if (reg_equal(shadow_reg, reg)) {
656                 matching = true;
657                 found[j] = true;
658                 if (track_ramblocks) {
659                     /*
660                      * Reset postcopy client bases, region_rb, and
661                      * region_rb_offset in case regions are removed.
662                      */
663                     if (fd > 0) {
664                         u->region_rb_offset[j] = offset;
665                         u->region_rb[j] = mr->ram_block;
666                         shadow_pcb[j] = u->postcopy_client_bases[i];
667                     } else {
668                         u->region_rb_offset[j] = 0;
669                         u->region_rb[j] = NULL;
670                     }
671                 }
672                 break;
673             }
674         }
675 
676         /*
677          * If the region was not found in the current device memory state
678          * create an entry for it in the removed list.
679          */
680         if (!matching) {
681             rem_reg[rm_idx].region = shadow_reg;
682             rem_reg[rm_idx++].reg_idx = i;
683         }
684     }
685 
686     /*
687      * For regions not marked "found", create entries in the added list.
688      *
689      * Note their indexes in the device memory state and the indexes of their
690      * file descriptors.
691      */
692     for (i = 0; i < dev->mem->nregions; i++) {
693         reg = &dev->mem->regions[i];
694         vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
695         if (fd > 0) {
696             ++fd_num;
697         }
698 
699         /*
700          * If the region was in both the shadow and device state we don't
701          * need to send a VHOST_USER_ADD_MEM_REG message for it.
702          */
703         if (found[i]) {
704             continue;
705         }
706 
707         add_reg[add_idx].region = reg;
708         add_reg[add_idx].reg_idx = i;
709         add_reg[add_idx++].fd_idx = fd_num;
710     }
711     *nr_rem_reg = rm_idx;
712     *nr_add_reg = add_idx;
713 
714     return;
715 }
716 
717 static int send_remove_regions(struct vhost_dev *dev,
718                                struct scrub_regions *remove_reg,
719                                int nr_rem_reg, VhostUserMsg *msg,
720                                bool reply_supported)
721 {
722     struct vhost_user *u = dev->opaque;
723     struct vhost_memory_region *shadow_reg;
724     int i, fd, shadow_reg_idx, ret;
725     ram_addr_t offset;
726     VhostUserMemoryRegion region_buffer;
727 
728     /*
729      * The regions in remove_reg appear in the same order they do in the
730      * shadow table. Therefore we can minimize memory copies by iterating
731      * through remove_reg backwards.
732      */
733     for (i = nr_rem_reg - 1; i >= 0; i--) {
734         shadow_reg = remove_reg[i].region;
735         shadow_reg_idx = remove_reg[i].reg_idx;
736 
737         vhost_user_get_mr_data(shadow_reg->userspace_addr, &offset, &fd);
738 
739         if (fd > 0) {
740             msg->hdr.request = VHOST_USER_REM_MEM_REG;
741             vhost_user_fill_msg_region(&region_buffer, shadow_reg, 0);
742             msg->payload.mem_reg.region = region_buffer;
743 
744             if (vhost_user_write(dev, msg, &fd, 1) < 0) {
745                 return -1;
746             }
747 
748             if (reply_supported) {
749                 ret = process_message_reply(dev, msg);
750                 if (ret) {
751                     return ret;
752                 }
753             }
754         }
755 
756         /*
757          * At this point we know the backend has unmapped the region. It is now
758          * safe to remove it from the shadow table.
759          */
760         memmove(&u->shadow_regions[shadow_reg_idx],
761                 &u->shadow_regions[shadow_reg_idx + 1],
762                 sizeof(struct vhost_memory_region) *
763                 (u->num_shadow_regions - shadow_reg_idx - 1));
764         u->num_shadow_regions--;
765     }
766 
767     return 0;
768 }
769 
770 static int send_add_regions(struct vhost_dev *dev,
771                             struct scrub_regions *add_reg, int nr_add_reg,
772                             VhostUserMsg *msg, uint64_t *shadow_pcb,
773                             bool reply_supported, bool track_ramblocks)
774 {
775     struct vhost_user *u = dev->opaque;
776     int i, fd, ret, reg_idx, reg_fd_idx;
777     struct vhost_memory_region *reg;
778     MemoryRegion *mr;
779     ram_addr_t offset;
780     VhostUserMsg msg_reply;
781     VhostUserMemoryRegion region_buffer;
782 
783     for (i = 0; i < nr_add_reg; i++) {
784         reg = add_reg[i].region;
785         reg_idx = add_reg[i].reg_idx;
786         reg_fd_idx = add_reg[i].fd_idx;
787 
788         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
789 
790         if (fd > 0) {
791             if (track_ramblocks) {
792                 trace_vhost_user_set_mem_table_withfd(reg_fd_idx, mr->name,
793                                                       reg->memory_size,
794                                                       reg->guest_phys_addr,
795                                                       reg->userspace_addr,
796                                                       offset);
797                 u->region_rb_offset[reg_idx] = offset;
798                 u->region_rb[reg_idx] = mr->ram_block;
799             }
800             msg->hdr.request = VHOST_USER_ADD_MEM_REG;
801             vhost_user_fill_msg_region(&region_buffer, reg, offset);
802             msg->payload.mem_reg.region = region_buffer;
803 
804             if (vhost_user_write(dev, msg, &fd, 1) < 0) {
805                 return -1;
806             }
807 
808             if (track_ramblocks) {
809                 uint64_t reply_gpa;
810 
811                 if (vhost_user_read(dev, &msg_reply) < 0) {
812                     return -1;
813                 }
814 
815                 reply_gpa = msg_reply.payload.mem_reg.region.guest_phys_addr;
816 
817                 if (msg_reply.hdr.request != VHOST_USER_ADD_MEM_REG) {
818                     error_report("%s: Received unexpected msg type."
819                                  "Expected %d received %d", __func__,
820                                  VHOST_USER_ADD_MEM_REG,
821                                  msg_reply.hdr.request);
822                     return -1;
823                 }
824 
825                 /*
826                  * We're using the same structure, just reusing one of the
827                  * fields, so it should be the same size.
828                  */
829                 if (msg_reply.hdr.size != msg->hdr.size) {
830                     error_report("%s: Unexpected size for postcopy reply "
831                                  "%d vs %d", __func__, msg_reply.hdr.size,
832                                  msg->hdr.size);
833                     return -1;
834                 }
835 
836                 /* Get the postcopy client base from the backend's reply. */
837                 if (reply_gpa == dev->mem->regions[reg_idx].guest_phys_addr) {
838                     shadow_pcb[reg_idx] =
839                         msg_reply.payload.mem_reg.region.userspace_addr;
840                     trace_vhost_user_set_mem_table_postcopy(
841                         msg_reply.payload.mem_reg.region.userspace_addr,
842                         msg->payload.mem_reg.region.userspace_addr,
843                         reg_fd_idx, reg_idx);
844                 } else {
845                     error_report("%s: invalid postcopy reply for region. "
846                                  "Got guest physical address %" PRIX64 ", expected "
847                                  "%" PRIX64, __func__, reply_gpa,
848                                  dev->mem->regions[reg_idx].guest_phys_addr);
849                     return -1;
850                 }
851             } else if (reply_supported) {
852                 ret = process_message_reply(dev, msg);
853                 if (ret) {
854                     return ret;
855                 }
856             }
857         } else if (track_ramblocks) {
858             u->region_rb_offset[reg_idx] = 0;
859             u->region_rb[reg_idx] = NULL;
860         }
861 
862         /*
863          * At this point, we know the backend has mapped in the new
864          * region, if the region has a valid file descriptor.
865          *
866          * The region should now be added to the shadow table.
867          */
868         u->shadow_regions[u->num_shadow_regions].guest_phys_addr =
869             reg->guest_phys_addr;
870         u->shadow_regions[u->num_shadow_regions].userspace_addr =
871             reg->userspace_addr;
872         u->shadow_regions[u->num_shadow_regions].memory_size =
873             reg->memory_size;
874         u->num_shadow_regions++;
875     }
876 
877     return 0;
878 }
879 
880 static int vhost_user_add_remove_regions(struct vhost_dev *dev,
881                                          VhostUserMsg *msg,
882                                          bool reply_supported,
883                                          bool track_ramblocks)
884 {
885     struct vhost_user *u = dev->opaque;
886     struct scrub_regions add_reg[VHOST_USER_MAX_RAM_SLOTS];
887     struct scrub_regions rem_reg[VHOST_USER_MAX_RAM_SLOTS];
888     uint64_t shadow_pcb[VHOST_USER_MAX_RAM_SLOTS] = {};
889     int nr_add_reg, nr_rem_reg;
890 
891     msg->hdr.size = sizeof(msg->payload.mem_reg);
892 
893     /* Find the regions which need to be removed or added. */
894     scrub_shadow_regions(dev, add_reg, &nr_add_reg, rem_reg, &nr_rem_reg,
895                          shadow_pcb, track_ramblocks);
896 
897     if (nr_rem_reg && send_remove_regions(dev, rem_reg, nr_rem_reg, msg,
898                 reply_supported) < 0)
899     {
900         goto err;
901     }
902 
903     if (nr_add_reg && send_add_regions(dev, add_reg, nr_add_reg, msg,
904                 shadow_pcb, reply_supported, track_ramblocks) < 0)
905     {
906         goto err;
907     }
908 
909     if (track_ramblocks) {
910         memcpy(u->postcopy_client_bases, shadow_pcb,
911                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
912         /*
913          * Now we've registered this with the postcopy code, we ack to the
914          * client, because now we're in the position to be able to deal with
915          * any faults it generates.
916          */
917         /* TODO: Use this for failure cases as well with a bad value. */
918         msg->hdr.size = sizeof(msg->payload.u64);
919         msg->payload.u64 = 0; /* OK */
920 
921         if (vhost_user_write(dev, msg, NULL, 0) < 0) {
922             return -1;
923         }
924     }
925 
926     return 0;
927 
928 err:
929     if (track_ramblocks) {
930         memcpy(u->postcopy_client_bases, shadow_pcb,
931                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
932     }
933 
934     return -1;
935 }
936 
937 static int vhost_user_set_mem_table_postcopy(struct vhost_dev *dev,
938                                              struct vhost_memory *mem,
939                                              bool reply_supported,
940                                              bool config_mem_slots)
941 {
942     struct vhost_user *u = dev->opaque;
943     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
944     size_t fd_num = 0;
945     VhostUserMsg msg_reply;
946     int region_i, msg_i;
947 
948     VhostUserMsg msg = {
949         .hdr.flags = VHOST_USER_VERSION,
950     };
951 
952     if (u->region_rb_len < dev->mem->nregions) {
953         u->region_rb = g_renew(RAMBlock*, u->region_rb, dev->mem->nregions);
954         u->region_rb_offset = g_renew(ram_addr_t, u->region_rb_offset,
955                                       dev->mem->nregions);
956         memset(&(u->region_rb[u->region_rb_len]), '\0',
957                sizeof(RAMBlock *) * (dev->mem->nregions - u->region_rb_len));
958         memset(&(u->region_rb_offset[u->region_rb_len]), '\0',
959                sizeof(ram_addr_t) * (dev->mem->nregions - u->region_rb_len));
960         u->region_rb_len = dev->mem->nregions;
961     }
962 
963     if (config_mem_slots) {
964         if (vhost_user_add_remove_regions(dev, &msg, reply_supported,
965                                           true) < 0) {
966             return -1;
967         }
968     } else {
969         if (vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
970                                               true) < 0) {
971             return -1;
972         }
973 
974         if (vhost_user_write(dev, &msg, fds, fd_num) < 0) {
975             return -1;
976         }
977 
978         if (vhost_user_read(dev, &msg_reply) < 0) {
979             return -1;
980         }
981 
982         if (msg_reply.hdr.request != VHOST_USER_SET_MEM_TABLE) {
983             error_report("%s: Received unexpected msg type."
984                          "Expected %d received %d", __func__,
985                          VHOST_USER_SET_MEM_TABLE, msg_reply.hdr.request);
986             return -1;
987         }
988 
989         /*
990          * We're using the same structure, just reusing one of the
991          * fields, so it should be the same size.
992          */
993         if (msg_reply.hdr.size != msg.hdr.size) {
994             error_report("%s: Unexpected size for postcopy reply "
995                          "%d vs %d", __func__, msg_reply.hdr.size,
996                          msg.hdr.size);
997             return -1;
998         }
999 
1000         memset(u->postcopy_client_bases, 0,
1001                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
1002 
1003         /*
1004          * They're in the same order as the regions that were sent
1005          * but some of the regions were skipped (above) if they
1006          * didn't have fd's
1007          */
1008         for (msg_i = 0, region_i = 0;
1009              region_i < dev->mem->nregions;
1010              region_i++) {
1011             if (msg_i < fd_num &&
1012                 msg_reply.payload.memory.regions[msg_i].guest_phys_addr ==
1013                 dev->mem->regions[region_i].guest_phys_addr) {
1014                 u->postcopy_client_bases[region_i] =
1015                     msg_reply.payload.memory.regions[msg_i].userspace_addr;
1016                 trace_vhost_user_set_mem_table_postcopy(
1017                     msg_reply.payload.memory.regions[msg_i].userspace_addr,
1018                     msg.payload.memory.regions[msg_i].userspace_addr,
1019                     msg_i, region_i);
1020                 msg_i++;
1021             }
1022         }
1023         if (msg_i != fd_num) {
1024             error_report("%s: postcopy reply not fully consumed "
1025                          "%d vs %zd",
1026                          __func__, msg_i, fd_num);
1027             return -1;
1028         }
1029 
1030         /*
1031          * Now we've registered this with the postcopy code, we ack to the
1032          * client, because now we're in the position to be able to deal
1033          * with any faults it generates.
1034          */
1035         /* TODO: Use this for failure cases as well with a bad value. */
1036         msg.hdr.size = sizeof(msg.payload.u64);
1037         msg.payload.u64 = 0; /* OK */
1038         if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1039             return -1;
1040         }
1041     }
1042 
1043     return 0;
1044 }
1045 
1046 static int vhost_user_set_mem_table(struct vhost_dev *dev,
1047                                     struct vhost_memory *mem)
1048 {
1049     struct vhost_user *u = dev->opaque;
1050     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
1051     size_t fd_num = 0;
1052     bool do_postcopy = u->postcopy_listen && u->postcopy_fd.handler;
1053     bool reply_supported = virtio_has_feature(dev->protocol_features,
1054                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1055     bool config_mem_slots =
1056         virtio_has_feature(dev->protocol_features,
1057                            VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS);
1058 
1059     if (do_postcopy) {
1060         /*
1061          * Postcopy has enough differences that it's best done in it's own
1062          * version
1063          */
1064         return vhost_user_set_mem_table_postcopy(dev, mem, reply_supported,
1065                                                  config_mem_slots);
1066     }
1067 
1068     VhostUserMsg msg = {
1069         .hdr.flags = VHOST_USER_VERSION,
1070     };
1071 
1072     if (reply_supported) {
1073         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1074     }
1075 
1076     if (config_mem_slots) {
1077         if (vhost_user_add_remove_regions(dev, &msg, reply_supported,
1078                                           false) < 0) {
1079             return -1;
1080         }
1081     } else {
1082         if (vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
1083                                               false) < 0) {
1084             return -1;
1085         }
1086         if (vhost_user_write(dev, &msg, fds, fd_num) < 0) {
1087             return -1;
1088         }
1089 
1090         if (reply_supported) {
1091             return process_message_reply(dev, &msg);
1092         }
1093     }
1094 
1095     return 0;
1096 }
1097 
1098 static int vhost_user_set_vring_endian(struct vhost_dev *dev,
1099                                        struct vhost_vring_state *ring)
1100 {
1101     bool cross_endian = virtio_has_feature(dev->protocol_features,
1102                                            VHOST_USER_PROTOCOL_F_CROSS_ENDIAN);
1103     VhostUserMsg msg = {
1104         .hdr.request = VHOST_USER_SET_VRING_ENDIAN,
1105         .hdr.flags = VHOST_USER_VERSION,
1106         .payload.state = *ring,
1107         .hdr.size = sizeof(msg.payload.state),
1108     };
1109 
1110     if (!cross_endian) {
1111         error_report("vhost-user trying to send unhandled ioctl");
1112         return -1;
1113     }
1114 
1115     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1116         return -1;
1117     }
1118 
1119     return 0;
1120 }
1121 
1122 static int vhost_set_vring(struct vhost_dev *dev,
1123                            unsigned long int request,
1124                            struct vhost_vring_state *ring)
1125 {
1126     VhostUserMsg msg = {
1127         .hdr.request = request,
1128         .hdr.flags = VHOST_USER_VERSION,
1129         .payload.state = *ring,
1130         .hdr.size = sizeof(msg.payload.state),
1131     };
1132 
1133     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1134         return -1;
1135     }
1136 
1137     return 0;
1138 }
1139 
1140 static int vhost_user_set_vring_num(struct vhost_dev *dev,
1141                                     struct vhost_vring_state *ring)
1142 {
1143     return vhost_set_vring(dev, VHOST_USER_SET_VRING_NUM, ring);
1144 }
1145 
1146 static void vhost_user_host_notifier_restore(struct vhost_dev *dev,
1147                                              int queue_idx)
1148 {
1149     struct vhost_user *u = dev->opaque;
1150     VhostUserHostNotifier *n = &u->user->notifier[queue_idx];
1151     VirtIODevice *vdev = dev->vdev;
1152 
1153     if (n->addr && !n->set) {
1154         virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, true);
1155         n->set = true;
1156     }
1157 }
1158 
1159 static void vhost_user_host_notifier_remove(struct vhost_dev *dev,
1160                                             int queue_idx)
1161 {
1162     struct vhost_user *u = dev->opaque;
1163     VhostUserHostNotifier *n = &u->user->notifier[queue_idx];
1164     VirtIODevice *vdev = dev->vdev;
1165 
1166     if (n->addr && n->set) {
1167         virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, false);
1168         n->set = false;
1169     }
1170 }
1171 
1172 static int vhost_user_set_vring_base(struct vhost_dev *dev,
1173                                      struct vhost_vring_state *ring)
1174 {
1175     vhost_user_host_notifier_restore(dev, ring->index);
1176 
1177     return vhost_set_vring(dev, VHOST_USER_SET_VRING_BASE, ring);
1178 }
1179 
1180 static int vhost_user_set_vring_enable(struct vhost_dev *dev, int enable)
1181 {
1182     int i;
1183 
1184     if (!virtio_has_feature(dev->features, VHOST_USER_F_PROTOCOL_FEATURES)) {
1185         return -1;
1186     }
1187 
1188     for (i = 0; i < dev->nvqs; ++i) {
1189         struct vhost_vring_state state = {
1190             .index = dev->vq_index + i,
1191             .num   = enable,
1192         };
1193 
1194         vhost_set_vring(dev, VHOST_USER_SET_VRING_ENABLE, &state);
1195     }
1196 
1197     return 0;
1198 }
1199 
1200 static int vhost_user_get_vring_base(struct vhost_dev *dev,
1201                                      struct vhost_vring_state *ring)
1202 {
1203     VhostUserMsg msg = {
1204         .hdr.request = VHOST_USER_GET_VRING_BASE,
1205         .hdr.flags = VHOST_USER_VERSION,
1206         .payload.state = *ring,
1207         .hdr.size = sizeof(msg.payload.state),
1208     };
1209 
1210     vhost_user_host_notifier_remove(dev, ring->index);
1211 
1212     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1213         return -1;
1214     }
1215 
1216     if (vhost_user_read(dev, &msg) < 0) {
1217         return -1;
1218     }
1219 
1220     if (msg.hdr.request != VHOST_USER_GET_VRING_BASE) {
1221         error_report("Received unexpected msg type. Expected %d received %d",
1222                      VHOST_USER_GET_VRING_BASE, msg.hdr.request);
1223         return -1;
1224     }
1225 
1226     if (msg.hdr.size != sizeof(msg.payload.state)) {
1227         error_report("Received bad msg size.");
1228         return -1;
1229     }
1230 
1231     *ring = msg.payload.state;
1232 
1233     return 0;
1234 }
1235 
1236 static int vhost_set_vring_file(struct vhost_dev *dev,
1237                                 VhostUserRequest request,
1238                                 struct vhost_vring_file *file)
1239 {
1240     int fds[VHOST_USER_MAX_RAM_SLOTS];
1241     size_t fd_num = 0;
1242     VhostUserMsg msg = {
1243         .hdr.request = request,
1244         .hdr.flags = VHOST_USER_VERSION,
1245         .payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK,
1246         .hdr.size = sizeof(msg.payload.u64),
1247     };
1248 
1249     if (ioeventfd_enabled() && file->fd > 0) {
1250         fds[fd_num++] = file->fd;
1251     } else {
1252         msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
1253     }
1254 
1255     if (vhost_user_write(dev, &msg, fds, fd_num) < 0) {
1256         return -1;
1257     }
1258 
1259     return 0;
1260 }
1261 
1262 static int vhost_user_set_vring_kick(struct vhost_dev *dev,
1263                                      struct vhost_vring_file *file)
1264 {
1265     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_KICK, file);
1266 }
1267 
1268 static int vhost_user_set_vring_call(struct vhost_dev *dev,
1269                                      struct vhost_vring_file *file)
1270 {
1271     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_CALL, file);
1272 }
1273 
1274 
1275 static int vhost_user_get_u64(struct vhost_dev *dev, int request, uint64_t *u64)
1276 {
1277     VhostUserMsg msg = {
1278         .hdr.request = request,
1279         .hdr.flags = VHOST_USER_VERSION,
1280     };
1281 
1282     if (vhost_user_one_time_request(request) && dev->vq_index != 0) {
1283         return 0;
1284     }
1285 
1286     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1287         return -1;
1288     }
1289 
1290     if (vhost_user_read(dev, &msg) < 0) {
1291         return -1;
1292     }
1293 
1294     if (msg.hdr.request != request) {
1295         error_report("Received unexpected msg type. Expected %d received %d",
1296                      request, msg.hdr.request);
1297         return -1;
1298     }
1299 
1300     if (msg.hdr.size != sizeof(msg.payload.u64)) {
1301         error_report("Received bad msg size.");
1302         return -1;
1303     }
1304 
1305     *u64 = msg.payload.u64;
1306 
1307     return 0;
1308 }
1309 
1310 static int vhost_user_get_features(struct vhost_dev *dev, uint64_t *features)
1311 {
1312     if (vhost_user_get_u64(dev, VHOST_USER_GET_FEATURES, features) < 0) {
1313         return -EPROTO;
1314     }
1315 
1316     return 0;
1317 }
1318 
1319 static int enforce_reply(struct vhost_dev *dev,
1320                          const VhostUserMsg *msg)
1321 {
1322     uint64_t dummy;
1323 
1324     if (msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1325         return process_message_reply(dev, msg);
1326     }
1327 
1328    /*
1329     * We need to wait for a reply but the backend does not
1330     * support replies for the command we just sent.
1331     * Send VHOST_USER_GET_FEATURES which makes all backends
1332     * send a reply.
1333     */
1334     return vhost_user_get_features(dev, &dummy);
1335 }
1336 
1337 static int vhost_user_set_vring_addr(struct vhost_dev *dev,
1338                                      struct vhost_vring_addr *addr)
1339 {
1340     VhostUserMsg msg = {
1341         .hdr.request = VHOST_USER_SET_VRING_ADDR,
1342         .hdr.flags = VHOST_USER_VERSION,
1343         .payload.addr = *addr,
1344         .hdr.size = sizeof(msg.payload.addr),
1345     };
1346 
1347     bool reply_supported = virtio_has_feature(dev->protocol_features,
1348                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1349 
1350     /*
1351      * wait for a reply if logging is enabled to make sure
1352      * backend is actually logging changes
1353      */
1354     bool wait_for_reply = addr->flags & (1 << VHOST_VRING_F_LOG);
1355 
1356     if (reply_supported && wait_for_reply) {
1357         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1358     }
1359 
1360     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1361         return -1;
1362     }
1363 
1364     if (wait_for_reply) {
1365         return enforce_reply(dev, &msg);
1366     }
1367 
1368     return 0;
1369 }
1370 
1371 static int vhost_user_set_u64(struct vhost_dev *dev, int request, uint64_t u64,
1372                               bool wait_for_reply)
1373 {
1374     VhostUserMsg msg = {
1375         .hdr.request = request,
1376         .hdr.flags = VHOST_USER_VERSION,
1377         .payload.u64 = u64,
1378         .hdr.size = sizeof(msg.payload.u64),
1379     };
1380 
1381     if (wait_for_reply) {
1382         bool reply_supported = virtio_has_feature(dev->protocol_features,
1383                                           VHOST_USER_PROTOCOL_F_REPLY_ACK);
1384         if (reply_supported) {
1385             msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1386         }
1387     }
1388 
1389     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1390         return -1;
1391     }
1392 
1393     if (wait_for_reply) {
1394         return enforce_reply(dev, &msg);
1395     }
1396 
1397     return 0;
1398 }
1399 
1400 static int vhost_user_set_features(struct vhost_dev *dev,
1401                                    uint64_t features)
1402 {
1403     /*
1404      * wait for a reply if logging is enabled to make sure
1405      * backend is actually logging changes
1406      */
1407     bool log_enabled = features & (0x1ULL << VHOST_F_LOG_ALL);
1408 
1409     return vhost_user_set_u64(dev, VHOST_USER_SET_FEATURES, features,
1410                               log_enabled);
1411 }
1412 
1413 static int vhost_user_set_protocol_features(struct vhost_dev *dev,
1414                                             uint64_t features)
1415 {
1416     return vhost_user_set_u64(dev, VHOST_USER_SET_PROTOCOL_FEATURES, features,
1417                               false);
1418 }
1419 
1420 static int vhost_user_set_owner(struct vhost_dev *dev)
1421 {
1422     VhostUserMsg msg = {
1423         .hdr.request = VHOST_USER_SET_OWNER,
1424         .hdr.flags = VHOST_USER_VERSION,
1425     };
1426 
1427     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1428         return -EPROTO;
1429     }
1430 
1431     return 0;
1432 }
1433 
1434 static int vhost_user_get_max_memslots(struct vhost_dev *dev,
1435                                        uint64_t *max_memslots)
1436 {
1437     uint64_t backend_max_memslots;
1438     int err;
1439 
1440     err = vhost_user_get_u64(dev, VHOST_USER_GET_MAX_MEM_SLOTS,
1441                              &backend_max_memslots);
1442     if (err < 0) {
1443         return err;
1444     }
1445 
1446     *max_memslots = backend_max_memslots;
1447 
1448     return 0;
1449 }
1450 
1451 static int vhost_user_reset_device(struct vhost_dev *dev)
1452 {
1453     VhostUserMsg msg = {
1454         .hdr.flags = VHOST_USER_VERSION,
1455     };
1456 
1457     msg.hdr.request = virtio_has_feature(dev->protocol_features,
1458                                          VHOST_USER_PROTOCOL_F_RESET_DEVICE)
1459         ? VHOST_USER_RESET_DEVICE
1460         : VHOST_USER_RESET_OWNER;
1461 
1462     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1463         return -1;
1464     }
1465 
1466     return 0;
1467 }
1468 
1469 static int vhost_user_slave_handle_config_change(struct vhost_dev *dev)
1470 {
1471     int ret = -1;
1472 
1473     if (!dev->config_ops) {
1474         return -1;
1475     }
1476 
1477     if (dev->config_ops->vhost_dev_config_notifier) {
1478         ret = dev->config_ops->vhost_dev_config_notifier(dev);
1479     }
1480 
1481     return ret;
1482 }
1483 
1484 static int vhost_user_slave_handle_vring_host_notifier(struct vhost_dev *dev,
1485                                                        VhostUserVringArea *area,
1486                                                        int fd)
1487 {
1488     int queue_idx = area->u64 & VHOST_USER_VRING_IDX_MASK;
1489     size_t page_size = qemu_real_host_page_size;
1490     struct vhost_user *u = dev->opaque;
1491     VhostUserState *user = u->user;
1492     VirtIODevice *vdev = dev->vdev;
1493     VhostUserHostNotifier *n;
1494     void *addr;
1495     char *name;
1496 
1497     if (!virtio_has_feature(dev->protocol_features,
1498                             VHOST_USER_PROTOCOL_F_HOST_NOTIFIER) ||
1499         vdev == NULL || queue_idx >= virtio_get_num_queues(vdev)) {
1500         return -1;
1501     }
1502 
1503     n = &user->notifier[queue_idx];
1504 
1505     if (n->addr) {
1506         virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, false);
1507         object_unparent(OBJECT(&n->mr));
1508         munmap(n->addr, page_size);
1509         n->addr = NULL;
1510     }
1511 
1512     if (area->u64 & VHOST_USER_VRING_NOFD_MASK) {
1513         return 0;
1514     }
1515 
1516     /* Sanity check. */
1517     if (area->size != page_size) {
1518         return -1;
1519     }
1520 
1521     addr = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1522                 fd, area->offset);
1523     if (addr == MAP_FAILED) {
1524         return -1;
1525     }
1526 
1527     name = g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
1528                            user, queue_idx);
1529     memory_region_init_ram_device_ptr(&n->mr, OBJECT(vdev), name,
1530                                       page_size, addr);
1531     g_free(name);
1532 
1533     if (virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, true)) {
1534         object_unparent(OBJECT(&n->mr));
1535         munmap(addr, page_size);
1536         return -1;
1537     }
1538 
1539     n->addr = addr;
1540     n->set = true;
1541 
1542     return 0;
1543 }
1544 
1545 static void close_slave_channel(struct vhost_user *u)
1546 {
1547     g_source_destroy(u->slave_src);
1548     g_source_unref(u->slave_src);
1549     u->slave_src = NULL;
1550     object_unref(OBJECT(u->slave_ioc));
1551     u->slave_ioc = NULL;
1552 }
1553 
1554 static gboolean slave_read(QIOChannel *ioc, GIOCondition condition,
1555                            gpointer opaque)
1556 {
1557     struct vhost_dev *dev = opaque;
1558     struct vhost_user *u = dev->opaque;
1559     VhostUserHeader hdr = { 0, };
1560     VhostUserPayload payload = { 0, };
1561     Error *local_err = NULL;
1562     gboolean rc = G_SOURCE_CONTINUE;
1563     int ret = 0;
1564     struct iovec iov;
1565     g_autofree int *fd = NULL;
1566     size_t fdsize = 0;
1567     int i;
1568 
1569     /* Read header */
1570     iov.iov_base = &hdr;
1571     iov.iov_len = VHOST_USER_HDR_SIZE;
1572 
1573     if (qio_channel_readv_full_all(ioc, &iov, 1, &fd, &fdsize, &local_err)) {
1574         error_report_err(local_err);
1575         goto err;
1576     }
1577 
1578     if (hdr.size > VHOST_USER_PAYLOAD_SIZE) {
1579         error_report("Failed to read msg header."
1580                 " Size %d exceeds the maximum %zu.", hdr.size,
1581                 VHOST_USER_PAYLOAD_SIZE);
1582         goto err;
1583     }
1584 
1585     /* Read payload */
1586     if (qio_channel_read_all(ioc, (char *) &payload, hdr.size, &local_err)) {
1587         error_report_err(local_err);
1588         goto err;
1589     }
1590 
1591     switch (hdr.request) {
1592     case VHOST_USER_SLAVE_IOTLB_MSG:
1593         ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
1594         break;
1595     case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
1596         ret = vhost_user_slave_handle_config_change(dev);
1597         break;
1598     case VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG:
1599         ret = vhost_user_slave_handle_vring_host_notifier(dev, &payload.area,
1600                                                           fd ? fd[0] : -1);
1601         break;
1602     default:
1603         error_report("Received unexpected msg type: %d.", hdr.request);
1604         ret = -EINVAL;
1605     }
1606 
1607     /*
1608      * REPLY_ACK feature handling. Other reply types has to be managed
1609      * directly in their request handlers.
1610      */
1611     if (hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1612         struct iovec iovec[2];
1613 
1614 
1615         hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
1616         hdr.flags |= VHOST_USER_REPLY_MASK;
1617 
1618         payload.u64 = !!ret;
1619         hdr.size = sizeof(payload.u64);
1620 
1621         iovec[0].iov_base = &hdr;
1622         iovec[0].iov_len = VHOST_USER_HDR_SIZE;
1623         iovec[1].iov_base = &payload;
1624         iovec[1].iov_len = hdr.size;
1625 
1626         if (qio_channel_writev_all(ioc, iovec, ARRAY_SIZE(iovec), &local_err)) {
1627             error_report_err(local_err);
1628             goto err;
1629         }
1630     }
1631 
1632     goto fdcleanup;
1633 
1634 err:
1635     close_slave_channel(u);
1636     rc = G_SOURCE_REMOVE;
1637 
1638 fdcleanup:
1639     if (fd) {
1640         for (i = 0; i < fdsize; i++) {
1641             close(fd[i]);
1642         }
1643     }
1644     return rc;
1645 }
1646 
1647 static int vhost_setup_slave_channel(struct vhost_dev *dev)
1648 {
1649     VhostUserMsg msg = {
1650         .hdr.request = VHOST_USER_SET_SLAVE_REQ_FD,
1651         .hdr.flags = VHOST_USER_VERSION,
1652     };
1653     struct vhost_user *u = dev->opaque;
1654     int sv[2], ret = 0;
1655     bool reply_supported = virtio_has_feature(dev->protocol_features,
1656                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1657     Error *local_err = NULL;
1658     QIOChannel *ioc;
1659 
1660     if (!virtio_has_feature(dev->protocol_features,
1661                             VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
1662         return 0;
1663     }
1664 
1665     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
1666         error_report("socketpair() failed");
1667         return -1;
1668     }
1669 
1670     ioc = QIO_CHANNEL(qio_channel_socket_new_fd(sv[0], &local_err));
1671     if (!ioc) {
1672         error_report_err(local_err);
1673         return -1;
1674     }
1675     u->slave_ioc = ioc;
1676     slave_update_read_handler(dev, NULL);
1677 
1678     if (reply_supported) {
1679         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1680     }
1681 
1682     ret = vhost_user_write(dev, &msg, &sv[1], 1);
1683     if (ret) {
1684         goto out;
1685     }
1686 
1687     if (reply_supported) {
1688         ret = process_message_reply(dev, &msg);
1689     }
1690 
1691 out:
1692     close(sv[1]);
1693     if (ret) {
1694         close_slave_channel(u);
1695     }
1696 
1697     return ret;
1698 }
1699 
1700 #ifdef CONFIG_LINUX
1701 /*
1702  * Called back from the postcopy fault thread when a fault is received on our
1703  * ufd.
1704  * TODO: This is Linux specific
1705  */
1706 static int vhost_user_postcopy_fault_handler(struct PostCopyFD *pcfd,
1707                                              void *ufd)
1708 {
1709     struct vhost_dev *dev = pcfd->data;
1710     struct vhost_user *u = dev->opaque;
1711     struct uffd_msg *msg = ufd;
1712     uint64_t faultaddr = msg->arg.pagefault.address;
1713     RAMBlock *rb = NULL;
1714     uint64_t rb_offset;
1715     int i;
1716 
1717     trace_vhost_user_postcopy_fault_handler(pcfd->idstr, faultaddr,
1718                                             dev->mem->nregions);
1719     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1720         trace_vhost_user_postcopy_fault_handler_loop(i,
1721                 u->postcopy_client_bases[i], dev->mem->regions[i].memory_size);
1722         if (faultaddr >= u->postcopy_client_bases[i]) {
1723             /* Ofset of the fault address in the vhost region */
1724             uint64_t region_offset = faultaddr - u->postcopy_client_bases[i];
1725             if (region_offset < dev->mem->regions[i].memory_size) {
1726                 rb_offset = region_offset + u->region_rb_offset[i];
1727                 trace_vhost_user_postcopy_fault_handler_found(i,
1728                         region_offset, rb_offset);
1729                 rb = u->region_rb[i];
1730                 return postcopy_request_shared_page(pcfd, rb, faultaddr,
1731                                                     rb_offset);
1732             }
1733         }
1734     }
1735     error_report("%s: Failed to find region for fault %" PRIx64,
1736                  __func__, faultaddr);
1737     return -1;
1738 }
1739 
1740 static int vhost_user_postcopy_waker(struct PostCopyFD *pcfd, RAMBlock *rb,
1741                                      uint64_t offset)
1742 {
1743     struct vhost_dev *dev = pcfd->data;
1744     struct vhost_user *u = dev->opaque;
1745     int i;
1746 
1747     trace_vhost_user_postcopy_waker(qemu_ram_get_idstr(rb), offset);
1748 
1749     if (!u) {
1750         return 0;
1751     }
1752     /* Translate the offset into an address in the clients address space */
1753     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1754         if (u->region_rb[i] == rb &&
1755             offset >= u->region_rb_offset[i] &&
1756             offset < (u->region_rb_offset[i] +
1757                       dev->mem->regions[i].memory_size)) {
1758             uint64_t client_addr = (offset - u->region_rb_offset[i]) +
1759                                    u->postcopy_client_bases[i];
1760             trace_vhost_user_postcopy_waker_found(client_addr);
1761             return postcopy_wake_shared(pcfd, client_addr, rb);
1762         }
1763     }
1764 
1765     trace_vhost_user_postcopy_waker_nomatch(qemu_ram_get_idstr(rb), offset);
1766     return 0;
1767 }
1768 #endif
1769 
1770 /*
1771  * Called at the start of an inbound postcopy on reception of the
1772  * 'advise' command.
1773  */
1774 static int vhost_user_postcopy_advise(struct vhost_dev *dev, Error **errp)
1775 {
1776 #ifdef CONFIG_LINUX
1777     struct vhost_user *u = dev->opaque;
1778     CharBackend *chr = u->user->chr;
1779     int ufd;
1780     VhostUserMsg msg = {
1781         .hdr.request = VHOST_USER_POSTCOPY_ADVISE,
1782         .hdr.flags = VHOST_USER_VERSION,
1783     };
1784 
1785     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1786         error_setg(errp, "Failed to send postcopy_advise to vhost");
1787         return -1;
1788     }
1789 
1790     if (vhost_user_read(dev, &msg) < 0) {
1791         error_setg(errp, "Failed to get postcopy_advise reply from vhost");
1792         return -1;
1793     }
1794 
1795     if (msg.hdr.request != VHOST_USER_POSTCOPY_ADVISE) {
1796         error_setg(errp, "Unexpected msg type. Expected %d received %d",
1797                      VHOST_USER_POSTCOPY_ADVISE, msg.hdr.request);
1798         return -1;
1799     }
1800 
1801     if (msg.hdr.size) {
1802         error_setg(errp, "Received bad msg size.");
1803         return -1;
1804     }
1805     ufd = qemu_chr_fe_get_msgfd(chr);
1806     if (ufd < 0) {
1807         error_setg(errp, "%s: Failed to get ufd", __func__);
1808         return -1;
1809     }
1810     qemu_set_nonblock(ufd);
1811 
1812     /* register ufd with userfault thread */
1813     u->postcopy_fd.fd = ufd;
1814     u->postcopy_fd.data = dev;
1815     u->postcopy_fd.handler = vhost_user_postcopy_fault_handler;
1816     u->postcopy_fd.waker = vhost_user_postcopy_waker;
1817     u->postcopy_fd.idstr = "vhost-user"; /* Need to find unique name */
1818     postcopy_register_shared_ufd(&u->postcopy_fd);
1819     return 0;
1820 #else
1821     error_setg(errp, "Postcopy not supported on non-Linux systems");
1822     return -1;
1823 #endif
1824 }
1825 
1826 /*
1827  * Called at the switch to postcopy on reception of the 'listen' command.
1828  */
1829 static int vhost_user_postcopy_listen(struct vhost_dev *dev, Error **errp)
1830 {
1831     struct vhost_user *u = dev->opaque;
1832     int ret;
1833     VhostUserMsg msg = {
1834         .hdr.request = VHOST_USER_POSTCOPY_LISTEN,
1835         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
1836     };
1837     u->postcopy_listen = true;
1838     trace_vhost_user_postcopy_listen();
1839     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1840         error_setg(errp, "Failed to send postcopy_listen to vhost");
1841         return -1;
1842     }
1843 
1844     ret = process_message_reply(dev, &msg);
1845     if (ret) {
1846         error_setg(errp, "Failed to receive reply to postcopy_listen");
1847         return ret;
1848     }
1849 
1850     return 0;
1851 }
1852 
1853 /*
1854  * Called at the end of postcopy
1855  */
1856 static int vhost_user_postcopy_end(struct vhost_dev *dev, Error **errp)
1857 {
1858     VhostUserMsg msg = {
1859         .hdr.request = VHOST_USER_POSTCOPY_END,
1860         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
1861     };
1862     int ret;
1863     struct vhost_user *u = dev->opaque;
1864 
1865     trace_vhost_user_postcopy_end_entry();
1866     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1867         error_setg(errp, "Failed to send postcopy_end to vhost");
1868         return -1;
1869     }
1870 
1871     ret = process_message_reply(dev, &msg);
1872     if (ret) {
1873         error_setg(errp, "Failed to receive reply to postcopy_end");
1874         return ret;
1875     }
1876     postcopy_unregister_shared_ufd(&u->postcopy_fd);
1877     close(u->postcopy_fd.fd);
1878     u->postcopy_fd.handler = NULL;
1879 
1880     trace_vhost_user_postcopy_end_exit();
1881 
1882     return 0;
1883 }
1884 
1885 static int vhost_user_postcopy_notifier(NotifierWithReturn *notifier,
1886                                         void *opaque)
1887 {
1888     struct PostcopyNotifyData *pnd = opaque;
1889     struct vhost_user *u = container_of(notifier, struct vhost_user,
1890                                          postcopy_notifier);
1891     struct vhost_dev *dev = u->dev;
1892 
1893     switch (pnd->reason) {
1894     case POSTCOPY_NOTIFY_PROBE:
1895         if (!virtio_has_feature(dev->protocol_features,
1896                                 VHOST_USER_PROTOCOL_F_PAGEFAULT)) {
1897             /* TODO: Get the device name into this error somehow */
1898             error_setg(pnd->errp,
1899                        "vhost-user backend not capable of postcopy");
1900             return -ENOENT;
1901         }
1902         break;
1903 
1904     case POSTCOPY_NOTIFY_INBOUND_ADVISE:
1905         return vhost_user_postcopy_advise(dev, pnd->errp);
1906 
1907     case POSTCOPY_NOTIFY_INBOUND_LISTEN:
1908         return vhost_user_postcopy_listen(dev, pnd->errp);
1909 
1910     case POSTCOPY_NOTIFY_INBOUND_END:
1911         return vhost_user_postcopy_end(dev, pnd->errp);
1912 
1913     default:
1914         /* We ignore notifications we don't know */
1915         break;
1916     }
1917 
1918     return 0;
1919 }
1920 
1921 static int vhost_user_backend_init(struct vhost_dev *dev, void *opaque,
1922                                    Error **errp)
1923 {
1924     uint64_t features, protocol_features, ram_slots;
1925     struct vhost_user *u;
1926     int err;
1927 
1928     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
1929 
1930     u = g_new0(struct vhost_user, 1);
1931     u->user = opaque;
1932     u->dev = dev;
1933     dev->opaque = u;
1934 
1935     err = vhost_user_get_features(dev, &features);
1936     if (err < 0) {
1937         error_setg_errno(errp, -err, "vhost_backend_init failed");
1938         return err;
1939     }
1940 
1941     if (virtio_has_feature(features, VHOST_USER_F_PROTOCOL_FEATURES)) {
1942         dev->backend_features |= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
1943 
1944         err = vhost_user_get_u64(dev, VHOST_USER_GET_PROTOCOL_FEATURES,
1945                                  &protocol_features);
1946         if (err < 0) {
1947             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
1948             return -EPROTO;
1949         }
1950 
1951         dev->protocol_features =
1952             protocol_features & VHOST_USER_PROTOCOL_FEATURE_MASK;
1953 
1954         if (!dev->config_ops || !dev->config_ops->vhost_dev_config_notifier) {
1955             /* Don't acknowledge CONFIG feature if device doesn't support it */
1956             dev->protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG);
1957         } else if (!(protocol_features &
1958                     (1ULL << VHOST_USER_PROTOCOL_F_CONFIG))) {
1959             error_setg(errp, "Device expects VHOST_USER_PROTOCOL_F_CONFIG "
1960                        "but backend does not support it.");
1961             return -EINVAL;
1962         }
1963 
1964         err = vhost_user_set_protocol_features(dev, dev->protocol_features);
1965         if (err < 0) {
1966             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
1967             return -EPROTO;
1968         }
1969 
1970         /* query the max queues we support if backend supports Multiple Queue */
1971         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)) {
1972             err = vhost_user_get_u64(dev, VHOST_USER_GET_QUEUE_NUM,
1973                                      &dev->max_queues);
1974             if (err < 0) {
1975                 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
1976                 return -EPROTO;
1977             }
1978         } else {
1979             dev->max_queues = 1;
1980         }
1981 
1982         if (dev->num_queues && dev->max_queues < dev->num_queues) {
1983             error_setg(errp, "The maximum number of queues supported by the "
1984                        "backend is %" PRIu64, dev->max_queues);
1985             return -EINVAL;
1986         }
1987 
1988         if (virtio_has_feature(features, VIRTIO_F_IOMMU_PLATFORM) &&
1989                 !(virtio_has_feature(dev->protocol_features,
1990                     VHOST_USER_PROTOCOL_F_SLAVE_REQ) &&
1991                  virtio_has_feature(dev->protocol_features,
1992                     VHOST_USER_PROTOCOL_F_REPLY_ACK))) {
1993             error_setg(errp, "IOMMU support requires reply-ack and "
1994                        "slave-req protocol features.");
1995             return -EINVAL;
1996         }
1997 
1998         /* get max memory regions if backend supports configurable RAM slots */
1999         if (!virtio_has_feature(dev->protocol_features,
2000                                 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS)) {
2001             u->user->memory_slots = VHOST_MEMORY_BASELINE_NREGIONS;
2002         } else {
2003             err = vhost_user_get_max_memslots(dev, &ram_slots);
2004             if (err < 0) {
2005                 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2006                 return -EPROTO;
2007             }
2008 
2009             if (ram_slots < u->user->memory_slots) {
2010                 error_setg(errp, "The backend specified a max ram slots limit "
2011                            "of %" PRIu64", when the prior validated limit was "
2012                            "%d. This limit should never decrease.", ram_slots,
2013                            u->user->memory_slots);
2014                 return -EINVAL;
2015             }
2016 
2017             u->user->memory_slots = MIN(ram_slots, VHOST_USER_MAX_RAM_SLOTS);
2018         }
2019     }
2020 
2021     if (dev->migration_blocker == NULL &&
2022         !virtio_has_feature(dev->protocol_features,
2023                             VHOST_USER_PROTOCOL_F_LOG_SHMFD)) {
2024         error_setg(&dev->migration_blocker,
2025                    "Migration disabled: vhost-user backend lacks "
2026                    "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
2027     }
2028 
2029     if (dev->vq_index == 0) {
2030         err = vhost_setup_slave_channel(dev);
2031         if (err < 0) {
2032             error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2033             return -EPROTO;
2034         }
2035     }
2036 
2037     u->postcopy_notifier.notify = vhost_user_postcopy_notifier;
2038     postcopy_add_notifier(&u->postcopy_notifier);
2039 
2040     return 0;
2041 }
2042 
2043 static int vhost_user_backend_cleanup(struct vhost_dev *dev)
2044 {
2045     struct vhost_user *u;
2046 
2047     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2048 
2049     u = dev->opaque;
2050     if (u->postcopy_notifier.notify) {
2051         postcopy_remove_notifier(&u->postcopy_notifier);
2052         u->postcopy_notifier.notify = NULL;
2053     }
2054     u->postcopy_listen = false;
2055     if (u->postcopy_fd.handler) {
2056         postcopy_unregister_shared_ufd(&u->postcopy_fd);
2057         close(u->postcopy_fd.fd);
2058         u->postcopy_fd.handler = NULL;
2059     }
2060     if (u->slave_ioc) {
2061         close_slave_channel(u);
2062     }
2063     g_free(u->region_rb);
2064     u->region_rb = NULL;
2065     g_free(u->region_rb_offset);
2066     u->region_rb_offset = NULL;
2067     u->region_rb_len = 0;
2068     g_free(u);
2069     dev->opaque = 0;
2070 
2071     return 0;
2072 }
2073 
2074 static int vhost_user_get_vq_index(struct vhost_dev *dev, int idx)
2075 {
2076     assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
2077 
2078     return idx;
2079 }
2080 
2081 static int vhost_user_memslots_limit(struct vhost_dev *dev)
2082 {
2083     struct vhost_user *u = dev->opaque;
2084 
2085     return u->user->memory_slots;
2086 }
2087 
2088 static bool vhost_user_requires_shm_log(struct vhost_dev *dev)
2089 {
2090     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2091 
2092     return virtio_has_feature(dev->protocol_features,
2093                               VHOST_USER_PROTOCOL_F_LOG_SHMFD);
2094 }
2095 
2096 static int vhost_user_migration_done(struct vhost_dev *dev, char* mac_addr)
2097 {
2098     VhostUserMsg msg = { };
2099 
2100     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2101 
2102     /* If guest supports GUEST_ANNOUNCE do nothing */
2103     if (virtio_has_feature(dev->acked_features, VIRTIO_NET_F_GUEST_ANNOUNCE)) {
2104         return 0;
2105     }
2106 
2107     /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
2108     if (virtio_has_feature(dev->protocol_features,
2109                            VHOST_USER_PROTOCOL_F_RARP)) {
2110         msg.hdr.request = VHOST_USER_SEND_RARP;
2111         msg.hdr.flags = VHOST_USER_VERSION;
2112         memcpy((char *)&msg.payload.u64, mac_addr, 6);
2113         msg.hdr.size = sizeof(msg.payload.u64);
2114 
2115         return vhost_user_write(dev, &msg, NULL, 0);
2116     }
2117     return -1;
2118 }
2119 
2120 static bool vhost_user_can_merge(struct vhost_dev *dev,
2121                                  uint64_t start1, uint64_t size1,
2122                                  uint64_t start2, uint64_t size2)
2123 {
2124     ram_addr_t offset;
2125     int mfd, rfd;
2126 
2127     (void)vhost_user_get_mr_data(start1, &offset, &mfd);
2128     (void)vhost_user_get_mr_data(start2, &offset, &rfd);
2129 
2130     return mfd == rfd;
2131 }
2132 
2133 static int vhost_user_net_set_mtu(struct vhost_dev *dev, uint16_t mtu)
2134 {
2135     VhostUserMsg msg;
2136     bool reply_supported = virtio_has_feature(dev->protocol_features,
2137                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2138 
2139     if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU))) {
2140         return 0;
2141     }
2142 
2143     msg.hdr.request = VHOST_USER_NET_SET_MTU;
2144     msg.payload.u64 = mtu;
2145     msg.hdr.size = sizeof(msg.payload.u64);
2146     msg.hdr.flags = VHOST_USER_VERSION;
2147     if (reply_supported) {
2148         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2149     }
2150 
2151     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2152         return -1;
2153     }
2154 
2155     /* If reply_ack supported, slave has to ack specified MTU is valid */
2156     if (reply_supported) {
2157         return process_message_reply(dev, &msg);
2158     }
2159 
2160     return 0;
2161 }
2162 
2163 static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
2164                                             struct vhost_iotlb_msg *imsg)
2165 {
2166     VhostUserMsg msg = {
2167         .hdr.request = VHOST_USER_IOTLB_MSG,
2168         .hdr.size = sizeof(msg.payload.iotlb),
2169         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
2170         .payload.iotlb = *imsg,
2171     };
2172 
2173     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2174         return -EFAULT;
2175     }
2176 
2177     return process_message_reply(dev, &msg);
2178 }
2179 
2180 
2181 static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
2182 {
2183     /* No-op as the receive channel is not dedicated to IOTLB messages. */
2184 }
2185 
2186 static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
2187                                  uint32_t config_len, Error **errp)
2188 {
2189     VhostUserMsg msg = {
2190         .hdr.request = VHOST_USER_GET_CONFIG,
2191         .hdr.flags = VHOST_USER_VERSION,
2192         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + config_len,
2193     };
2194 
2195     if (!virtio_has_feature(dev->protocol_features,
2196                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2197         error_setg(errp, "VHOST_USER_PROTOCOL_F_CONFIG not supported");
2198         return -EINVAL;
2199     }
2200 
2201     assert(config_len <= VHOST_USER_MAX_CONFIG_SIZE);
2202 
2203     msg.payload.config.offset = 0;
2204     msg.payload.config.size = config_len;
2205     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2206         error_setg_errno(errp, EPROTO, "vhost_get_config failed");
2207         return -EPROTO;
2208     }
2209 
2210     if (vhost_user_read(dev, &msg) < 0) {
2211         error_setg_errno(errp, EPROTO, "vhost_get_config failed");
2212         return -EPROTO;
2213     }
2214 
2215     if (msg.hdr.request != VHOST_USER_GET_CONFIG) {
2216         error_setg(errp,
2217                    "Received unexpected msg type. Expected %d received %d",
2218                    VHOST_USER_GET_CONFIG, msg.hdr.request);
2219         return -EINVAL;
2220     }
2221 
2222     if (msg.hdr.size != VHOST_USER_CONFIG_HDR_SIZE + config_len) {
2223         error_setg(errp, "Received bad msg size.");
2224         return -EINVAL;
2225     }
2226 
2227     memcpy(config, msg.payload.config.region, config_len);
2228 
2229     return 0;
2230 }
2231 
2232 static int vhost_user_set_config(struct vhost_dev *dev, const uint8_t *data,
2233                                  uint32_t offset, uint32_t size, uint32_t flags)
2234 {
2235     uint8_t *p;
2236     bool reply_supported = virtio_has_feature(dev->protocol_features,
2237                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2238 
2239     VhostUserMsg msg = {
2240         .hdr.request = VHOST_USER_SET_CONFIG,
2241         .hdr.flags = VHOST_USER_VERSION,
2242         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + size,
2243     };
2244 
2245     if (!virtio_has_feature(dev->protocol_features,
2246                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2247         return -1;
2248     }
2249 
2250     if (reply_supported) {
2251         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2252     }
2253 
2254     if (size > VHOST_USER_MAX_CONFIG_SIZE) {
2255         return -1;
2256     }
2257 
2258     msg.payload.config.offset = offset,
2259     msg.payload.config.size = size,
2260     msg.payload.config.flags = flags,
2261     p = msg.payload.config.region;
2262     memcpy(p, data, size);
2263 
2264     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2265         return -1;
2266     }
2267 
2268     if (reply_supported) {
2269         return process_message_reply(dev, &msg);
2270     }
2271 
2272     return 0;
2273 }
2274 
2275 static int vhost_user_crypto_create_session(struct vhost_dev *dev,
2276                                             void *session_info,
2277                                             uint64_t *session_id)
2278 {
2279     bool crypto_session = virtio_has_feature(dev->protocol_features,
2280                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2281     CryptoDevBackendSymSessionInfo *sess_info = session_info;
2282     VhostUserMsg msg = {
2283         .hdr.request = VHOST_USER_CREATE_CRYPTO_SESSION,
2284         .hdr.flags = VHOST_USER_VERSION,
2285         .hdr.size = sizeof(msg.payload.session),
2286     };
2287 
2288     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2289 
2290     if (!crypto_session) {
2291         error_report("vhost-user trying to send unhandled ioctl");
2292         return -1;
2293     }
2294 
2295     memcpy(&msg.payload.session.session_setup_data, sess_info,
2296               sizeof(CryptoDevBackendSymSessionInfo));
2297     if (sess_info->key_len) {
2298         memcpy(&msg.payload.session.key, sess_info->cipher_key,
2299                sess_info->key_len);
2300     }
2301     if (sess_info->auth_key_len > 0) {
2302         memcpy(&msg.payload.session.auth_key, sess_info->auth_key,
2303                sess_info->auth_key_len);
2304     }
2305     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2306         error_report("vhost_user_write() return -1, create session failed");
2307         return -1;
2308     }
2309 
2310     if (vhost_user_read(dev, &msg) < 0) {
2311         error_report("vhost_user_read() return -1, create session failed");
2312         return -1;
2313     }
2314 
2315     if (msg.hdr.request != VHOST_USER_CREATE_CRYPTO_SESSION) {
2316         error_report("Received unexpected msg type. Expected %d received %d",
2317                      VHOST_USER_CREATE_CRYPTO_SESSION, msg.hdr.request);
2318         return -1;
2319     }
2320 
2321     if (msg.hdr.size != sizeof(msg.payload.session)) {
2322         error_report("Received bad msg size.");
2323         return -1;
2324     }
2325 
2326     if (msg.payload.session.session_id < 0) {
2327         error_report("Bad session id: %" PRId64 "",
2328                               msg.payload.session.session_id);
2329         return -1;
2330     }
2331     *session_id = msg.payload.session.session_id;
2332 
2333     return 0;
2334 }
2335 
2336 static int
2337 vhost_user_crypto_close_session(struct vhost_dev *dev, uint64_t session_id)
2338 {
2339     bool crypto_session = virtio_has_feature(dev->protocol_features,
2340                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2341     VhostUserMsg msg = {
2342         .hdr.request = VHOST_USER_CLOSE_CRYPTO_SESSION,
2343         .hdr.flags = VHOST_USER_VERSION,
2344         .hdr.size = sizeof(msg.payload.u64),
2345     };
2346     msg.payload.u64 = session_id;
2347 
2348     if (!crypto_session) {
2349         error_report("vhost-user trying to send unhandled ioctl");
2350         return -1;
2351     }
2352 
2353     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2354         error_report("vhost_user_write() return -1, close session failed");
2355         return -1;
2356     }
2357 
2358     return 0;
2359 }
2360 
2361 static bool vhost_user_mem_section_filter(struct vhost_dev *dev,
2362                                           MemoryRegionSection *section)
2363 {
2364     bool result;
2365 
2366     result = memory_region_get_fd(section->mr) >= 0;
2367 
2368     return result;
2369 }
2370 
2371 static int vhost_user_get_inflight_fd(struct vhost_dev *dev,
2372                                       uint16_t queue_size,
2373                                       struct vhost_inflight *inflight)
2374 {
2375     void *addr;
2376     int fd;
2377     struct vhost_user *u = dev->opaque;
2378     CharBackend *chr = u->user->chr;
2379     VhostUserMsg msg = {
2380         .hdr.request = VHOST_USER_GET_INFLIGHT_FD,
2381         .hdr.flags = VHOST_USER_VERSION,
2382         .payload.inflight.num_queues = dev->nvqs,
2383         .payload.inflight.queue_size = queue_size,
2384         .hdr.size = sizeof(msg.payload.inflight),
2385     };
2386 
2387     if (!virtio_has_feature(dev->protocol_features,
2388                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2389         return 0;
2390     }
2391 
2392     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2393         return -1;
2394     }
2395 
2396     if (vhost_user_read(dev, &msg) < 0) {
2397         return -1;
2398     }
2399 
2400     if (msg.hdr.request != VHOST_USER_GET_INFLIGHT_FD) {
2401         error_report("Received unexpected msg type. "
2402                      "Expected %d received %d",
2403                      VHOST_USER_GET_INFLIGHT_FD, msg.hdr.request);
2404         return -1;
2405     }
2406 
2407     if (msg.hdr.size != sizeof(msg.payload.inflight)) {
2408         error_report("Received bad msg size.");
2409         return -1;
2410     }
2411 
2412     if (!msg.payload.inflight.mmap_size) {
2413         return 0;
2414     }
2415 
2416     fd = qemu_chr_fe_get_msgfd(chr);
2417     if (fd < 0) {
2418         error_report("Failed to get mem fd");
2419         return -1;
2420     }
2421 
2422     addr = mmap(0, msg.payload.inflight.mmap_size, PROT_READ | PROT_WRITE,
2423                 MAP_SHARED, fd, msg.payload.inflight.mmap_offset);
2424 
2425     if (addr == MAP_FAILED) {
2426         error_report("Failed to mmap mem fd");
2427         close(fd);
2428         return -1;
2429     }
2430 
2431     inflight->addr = addr;
2432     inflight->fd = fd;
2433     inflight->size = msg.payload.inflight.mmap_size;
2434     inflight->offset = msg.payload.inflight.mmap_offset;
2435     inflight->queue_size = queue_size;
2436 
2437     return 0;
2438 }
2439 
2440 static int vhost_user_set_inflight_fd(struct vhost_dev *dev,
2441                                       struct vhost_inflight *inflight)
2442 {
2443     VhostUserMsg msg = {
2444         .hdr.request = VHOST_USER_SET_INFLIGHT_FD,
2445         .hdr.flags = VHOST_USER_VERSION,
2446         .payload.inflight.mmap_size = inflight->size,
2447         .payload.inflight.mmap_offset = inflight->offset,
2448         .payload.inflight.num_queues = dev->nvqs,
2449         .payload.inflight.queue_size = inflight->queue_size,
2450         .hdr.size = sizeof(msg.payload.inflight),
2451     };
2452 
2453     if (!virtio_has_feature(dev->protocol_features,
2454                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2455         return 0;
2456     }
2457 
2458     if (vhost_user_write(dev, &msg, &inflight->fd, 1) < 0) {
2459         return -1;
2460     }
2461 
2462     return 0;
2463 }
2464 
2465 bool vhost_user_init(VhostUserState *user, CharBackend *chr, Error **errp)
2466 {
2467     if (user->chr) {
2468         error_setg(errp, "Cannot initialize vhost-user state");
2469         return false;
2470     }
2471     user->chr = chr;
2472     user->memory_slots = 0;
2473     return true;
2474 }
2475 
2476 void vhost_user_cleanup(VhostUserState *user)
2477 {
2478     int i;
2479 
2480     if (!user->chr) {
2481         return;
2482     }
2483 
2484     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2485         if (user->notifier[i].addr) {
2486             object_unparent(OBJECT(&user->notifier[i].mr));
2487             munmap(user->notifier[i].addr, qemu_real_host_page_size);
2488             user->notifier[i].addr = NULL;
2489         }
2490     }
2491     user->chr = NULL;
2492 }
2493 
2494 const VhostOps user_ops = {
2495         .backend_type = VHOST_BACKEND_TYPE_USER,
2496         .vhost_backend_init = vhost_user_backend_init,
2497         .vhost_backend_cleanup = vhost_user_backend_cleanup,
2498         .vhost_backend_memslots_limit = vhost_user_memslots_limit,
2499         .vhost_set_log_base = vhost_user_set_log_base,
2500         .vhost_set_mem_table = vhost_user_set_mem_table,
2501         .vhost_set_vring_addr = vhost_user_set_vring_addr,
2502         .vhost_set_vring_endian = vhost_user_set_vring_endian,
2503         .vhost_set_vring_num = vhost_user_set_vring_num,
2504         .vhost_set_vring_base = vhost_user_set_vring_base,
2505         .vhost_get_vring_base = vhost_user_get_vring_base,
2506         .vhost_set_vring_kick = vhost_user_set_vring_kick,
2507         .vhost_set_vring_call = vhost_user_set_vring_call,
2508         .vhost_set_features = vhost_user_set_features,
2509         .vhost_get_features = vhost_user_get_features,
2510         .vhost_set_owner = vhost_user_set_owner,
2511         .vhost_reset_device = vhost_user_reset_device,
2512         .vhost_get_vq_index = vhost_user_get_vq_index,
2513         .vhost_set_vring_enable = vhost_user_set_vring_enable,
2514         .vhost_requires_shm_log = vhost_user_requires_shm_log,
2515         .vhost_migration_done = vhost_user_migration_done,
2516         .vhost_backend_can_merge = vhost_user_can_merge,
2517         .vhost_net_set_mtu = vhost_user_net_set_mtu,
2518         .vhost_set_iotlb_callback = vhost_user_set_iotlb_callback,
2519         .vhost_send_device_iotlb_msg = vhost_user_send_device_iotlb_msg,
2520         .vhost_get_config = vhost_user_get_config,
2521         .vhost_set_config = vhost_user_set_config,
2522         .vhost_crypto_create_session = vhost_user_crypto_create_session,
2523         .vhost_crypto_close_session = vhost_user_crypto_close_session,
2524         .vhost_backend_mem_section_filter = vhost_user_mem_section_filter,
2525         .vhost_get_inflight_fd = vhost_user_get_inflight_fd,
2526         .vhost_set_inflight_fd = vhost_user_set_inflight_fd,
2527 };
2528