On Thu, 19 Dec 2024 12:13:53 +0100 Laurent Vivier wrote:

vhost-user protocol specification has been updated with feature flags and commands we will need to implement migration.

Signed-off-by: Laurent Vivier --- vhost_user.c | 5 +++++ vhost_user.h | 46 ++++++++++++++++++++++++++++++++++++++++++---- 2 files changed, 47 insertions(+), 4 deletions(-)

diff --git a/vhost_user.c b/vhost_user.c index 4b8558fa851f..48226a8b7686 100644 --- a/vhost_user.c +++ b/vhost_user.c @@ -110,6 +110,11 @@ static const char *vu_request_to_string(unsigned int req) REQ(VHOST_USER_GET_MAX_MEM_SLOTS), REQ(VHOST_USER_ADD_MEM_REG), REQ(VHOST_USER_REM_MEM_REG), + REQ(VHOST_USER_SET_STATUS), + REQ(VHOST_USER_GET_STATUS), + REQ(VHOST_USER_GET_SHARED_OBJECT), + REQ(VHOST_USER_SET_DEVICE_STATE_FD), + REQ(VHOST_USER_CHECK_DEVICE_STATE), }; #undef REQ return vu_request_str[req]; diff --git a/vhost_user.h b/vhost_user.h index 464ba21e962f..fbacb5560755 100644 --- a/vhost_user.h +++ b/vhost_user.h @@ -37,6 +37,10 @@ enum vhost_user_protocol_feature { VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD = 12, VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS = 14, VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS = 15, + VHOST_USER_PROTOCOL_F_STATUS = 16, + /* Feature 17 reserved for VHOST_USER_PROTOCOL_F_XEN_MMAP. */ + VHOST_USER_PROTOCOL_F_SHARED_OBJECT = 18, + VHOST_USER_PROTOCOL_F_DEVICE_STATE = 19,

VHOST_USER_PROTOCOL_F_MAX }; @@ -83,6 +87,11 @@ enum vhost_user_request { VHOST_USER_GET_MAX_MEM_SLOTS = 36, VHOST_USER_ADD_MEM_REG = 37, VHOST_USER_REM_MEM_REG = 38, + VHOST_USER_SET_STATUS = 39, + VHOST_USER_GET_STATUS = 40, + VHOST_USER_GET_SHARED_OBJECT = 41, + VHOST_USER_SET_DEVICE_STATE_FD = 42, + VHOST_USER_CHECK_DEVICE_STATE = 43, VHOST_USER_MAX };

@@ -128,12 +137,39 @@ struct vhost_user_memory { struct vhost_user_memory_region regions[VHOST_MEMORY_BASELINE_NREGIONS]; };

+/** + * struct vhost_user_log - Address and size of the shared memory region used + * to log page update + * @mmap_size: Size of the shared memory region + * @mmap_offset: Offset of the shared memory region + */ +struct vhost_user_log { + uint64_t mmap_size; + uint64_t mmap_offset; +}; + +/** + * struct vhost_user_transfer_device_state - Set the direction and phase + * of the backend device state fd + * @direction: Device state transfer direction (save or load) + * @phase: Migration phase (only stopped is supported) + */ +struct vhost_user_transfer_device_state { + uint32_t direction; +#define VHOST_USER_TRANSFER_STATE_DIRECTION_SAVE 0 +#define VHOST_USER_TRANSFER_STATE_DIRECTION_LOAD 1 + uint32_t phase; +#define VHOST_USER_TRANSFER_STATE_PHASE_STOPPED 0 +}; + /** * union vhost_user_payload - vhost-user message payload - * @u64: 64-bit payload - * @state: vring state payload - * @addr: vring addresses payload - * vhost_user_memory: Memory regions information payload + * @u64: 64-bit payload + * @state: vring state payload + * @addr: vring addresses payload + * @vhost_user_memory: Memory regions information payload + * @vhost_user_log: Memory logging payload + * @vhost_user_transfer_device_state: Device state payload

Nit: these are called 'memory', 'log', and 'transfer_state'.

*/ union vhost_user_payload { #define VHOST_USER_VRING_IDX_MASK 0xff @@ -142,6 +178,8 @@ union vhost_user_payload { struct vhost_vring_state state; struct vhost_vring_addr addr; struct vhost_user_memory memory; + struct vhost_user_log log; + struct vhost_user_transfer_device_state transfer_state; };

/**

-- Stefano

On Thu, 19 Dec 2024 12:13:54 +0100 Laurent Vivier wrote:

VHOST_USER_SET_LOG_FD is an optional message with an eventfd in ancillary data, it may be used to inform the front-end that the log has been modified.

Signed-off-by: Laurent Vivier --- vhost_user.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++++++ vhost_user.h | 1 + virtio.h | 2 ++ 3 files changed, 59 insertions(+)

diff --git a/vhost_user.c b/vhost_user.c index 48226a8b7686..ce4373d9eeca 100644 --- a/vhost_user.c +++ b/vhost_user.c @@ -504,6 +504,57 @@ static bool vu_set_mem_table_exec(struct vu_dev *vdev, return false; }

+/** + * vu_close_log() - Close the logging file descriptor + * @vdev: vhost-user device + */ +static void vu_close_log(struct vu_dev *vdev) +{ + if (vdev->log_call_fd != -1) { + close(vdev->log_call_fd); + vdev->log_call_fd = -1; + } +} + +/** + * vu_log_kick() - Inform the front-end that the log has been modified + * @vdev: vhost-user device + */ +/* cppcheck-suppress unusedFunction */ +void vu_log_kick(const struct vu_dev *vdev) +{ + if (vdev->log_call_fd != -1) { + int rc; + + rc = eventfd_write(vdev->log_call_fd, 1); + if (rc == -1) + die_perror("vhost-user kick eventfd_write()"); + } +} + +/** + * vu_set_log_fd_exec() -- Set the eventfd used to report logging update

Nit: single '-' between function name and comment. -- Stefano

On 17/01/2025 19:05, Stefano Brivio wrote:

On Thu, 19 Dec 2024 12:13:56 +0100 Laurent Vivier wrote:

...

Sets logging shared memory space.

When the back-end has VHOST_USER_PROTOCOL_F_LOG_SHMFD protocol feature, the log memory fd is provided in the ancillary data of VHOST_USER_SET_LOG_BASE message, the size and offset of shared memory area provided in the message.

Signed-off-by: Laurent Vivier --- util.h | 3 ++ vhost_user.c | 87 +++++++++++++++++++++++++++++++++++++++++++++++++++- vhost_user.h | 3 ++ virtio.c | 74 ++++++++++++++++++++++++++++++++++++++++++-- virtio.h | 4 +++ 5 files changed, 168 insertions(+), 3 deletions(-)

diff --git a/util.h b/util.h index 3fa1d12544a0..d02333d5a88d 100644 --- a/util.h +++ b/util.h @@ -152,6 +152,9 @@ static inline void barrier(void) { __asm__ __volatile__("" ::: "memory"); } #define smp_wmb() smp_mb_release() #define smp_rmb() smp_mb_acquire()

+#define qatomic_or(ptr, n) \ + ((void) __atomic_fetch_or(ptr, n, __ATOMIC_SEQ_CST)) + #define NS_FN_STACK_SIZE (1024 * 1024) /* 1MiB */

int do_clone(int (*fn)(void *), char *stack_area, size_t stack_size, int flags, diff --git a/vhost_user.c b/vhost_user.c index ce4373d9eeca..c2fac58badf1 100644 --- a/vhost_user.c +++ b/vhost_user.c @@ -510,6 +510,12 @@ static bool vu_set_mem_table_exec(struct vu_dev *vdev, */ static void vu_close_log(struct vu_dev *vdev) { + if (vdev->log_table) { + if (munmap(vdev->log_table, vdev->log_size) != 0) + die_perror("close log munmap() error"); + vdev->log_table = NULL; + } + if (vdev->log_call_fd != -1) { close(vdev->log_call_fd); vdev->log_call_fd = -1; @@ -520,7 +526,6 @@ static void vu_close_log(struct vu_dev *vdev) * vu_log_kick() - Inform the front-end that the log has been modified * @vdev: vhost-user device */ -/* cppcheck-suppress unusedFunction */ void vu_log_kick(const struct vu_dev *vdev) { if (vdev->log_call_fd != -1) { @@ -532,6 +537,84 @@ void vu_log_kick(const struct vu_dev *vdev) } }

+ +/**

Excess newline.

...

+ * vu_log_page() -- Update logging table

Single '-' between function name and comment.

...

+ * @log_table: Base address of the logging table + * @page: Page number that has been updated + */ +/* NOLINTNEXTLINE(readability-non-const-parameter) */ +static void vu_log_page(uint8_t *log_table, uint64_t page) +{ + qatomic_or(&log_table[page / 8], 1 << (page % 8)); +} + +/** + * vu_log_write() -- Log memory write

Single '-' between function name and comment.

...

+ * @dev: Vhost-user device

vhost-user

...

+ * @address: Memory address + * @length: Memory size + */ +void vu_log_write(const struct vu_dev *vdev, uint64_t address, uint64_t length) +{ + uint64_t page; + + if (!vdev->log_table || !length || + !vu_has_feature(vdev, VHOST_F_LOG_ALL)) + return; + + page = address / VHOST_LOG_PAGE; + while (page * VHOST_LOG_PAGE < address + length) { + vu_log_page(vdev->log_table, page); + page++; + } + vu_log_kick(vdev); +} + +/** + * vu_set_log_base_exec() - Set the memory log base + * @vdev: vhost-user device + * @vmsg: vhost-user message + * + * Return: False as no reply is requested + * + * #syscalls:vu mmap munmap

I wonder: will there be a way around this the day that we want to disable mmap() for vhost-user mode too?

I don't think we can bypass the use of mmap.

...

+ */ +static bool vu_set_log_base_exec(struct vu_dev *vdev, + struct vhost_user_msg *msg) +{ + uint64_t log_mmap_size, log_mmap_offset; + void *base; + int fd; + + if (msg->fd_num != 1 || msg->hdr.size != sizeof(msg->payload.log)) + die("Invalid log_base message");

Maybe prefix this with "vhost-user:", otherwise it's not really clear where it's coming from.

...

+ + fd = msg->fds[0]; + log_mmap_offset = msg->payload.log.mmap_offset; + log_mmap_size = msg->payload.log.mmap_size; + + debug("Log mmap_offset: %"PRId64, log_mmap_offset); + debug("Log mmap_size: %"PRId64, log_mmap_size); + + base = mmap(0, log_mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, + log_mmap_offset); + close(fd); + if (base == MAP_FAILED) + die("log mmap error");

Same here.

...

+ + if (vdev->log_table) + munmap(vdev->log_table, vdev->log_size); + + vdev->log_table = base; + vdev->log_size = log_mmap_size; + + msg->hdr.size = sizeof(msg->payload.u64); + msg->fd_num = 0; + + return true; +} + /** * vu_set_log_fd_exec() -- Set the eventfd used to report logging update * @vdev: vhost-user device @@ -915,6 +998,7 @@ void vu_init(struct ctx *c) .notification = true, }; } + c->vdev->log_table = NULL; c->vdev->log_call_fd = -1; }

@@ -984,6 +1068,7 @@ static bool (*vu_handle[VHOST_USER_MAX])(struct vu_dev *vdev, [VHOST_USER_GET_QUEUE_NUM] = vu_get_queue_num_exec, [VHOST_USER_SET_OWNER] = vu_set_owner_exec, [VHOST_USER_SET_MEM_TABLE] = vu_set_mem_table_exec, + [VHOST_USER_SET_LOG_BASE] = vu_set_log_base_exec, [VHOST_USER_SET_LOG_FD] = vu_set_log_fd_exec, [VHOST_USER_SET_VRING_NUM] = vu_set_vring_num_exec, [VHOST_USER_SET_VRING_ADDR] = vu_set_vring_addr_exec, diff --git a/vhost_user.h b/vhost_user.h index 2fc0342ff5ba..22a5d059073f 100644 --- a/vhost_user.h +++ b/vhost_user.h @@ -15,6 +15,7 @@ #include "iov.h"

#define VHOST_USER_F_PROTOCOL_FEATURES 30 +#define VHOST_LOG_PAGE 4096

Does this need to be 65536 on ppc64 and ppc64le? In case, we have PAGE_SIZE exported by the Makefile in (it uses 'getconf' so it's not cross-build-safe, we should find a better way eventually).

VHOST_LOG_PAGE is defined as 0x1000 and does not depend on the architecture type. https://qemu-project.gitlab.io/qemu/interop/vhost-user.html#migration

...

#define VHOST_MEMORY_BASELINE_NREGIONS 8

@@ -241,5 +242,7 @@ void vu_print_capabilities(void); void vu_init(struct ctx *c); void vu_cleanup(struct vu_dev *vdev); void vu_log_kick(const struct vu_dev *vdev); +void vu_log_write(const struct vu_dev *vdev, uint64_t address, + uint64_t length); void vu_control_handler(struct vu_dev *vdev, int fd, uint32_t events); #endif /* VHOST_USER_H */ diff --git a/virtio.c b/virtio.c index 52d5a4d4be52..13838586ad1a 100644 --- a/virtio.c +++ b/virtio.c @@ -81,6 +81,7 @@

#include "util.h" #include "virtio.h" +#include "vhost_user.h"

#define VIRTQUEUE_MAX_SIZE 1024

@@ -592,7 +593,72 @@ static inline void vring_used_write(const struct vu_dev *vdev, struct vring_used *used = vq->vring.used;

used->ring[i] = *uelem; - (void)vdev; + vu_log_write(vdev, vq->vring.log_guest_addr + + offsetof(struct vring_used, ring[i]), + sizeof(used->ring[i])); +} + +/** + * vu_log_queue_fill() -- Log virtqueue memory update

Single '-' between function name and comment.

...

+ * @dev: Vhost-user device

vhost-user

...

+ * @vq: Virtqueue + * @index: Descriptor ring index + * @len: Size of the element + */ +static void vu_log_queue_fill(const struct vu_dev *vdev, struct vu_virtq *vq, + unsigned int index, unsigned int len) +{ + struct vring_desc desc_buf[VIRTQUEUE_MAX_SIZE]; + struct vring_desc *desc = vq->vring.desc; + unsigned int max, min; + unsigned num_bufs = 0; + uint64_t read_len; + + if (!vdev->log_table || !len || !vu_has_feature(vdev, VHOST_F_LOG_ALL)) + return; + + max = vq->vring.num; + + if (le16toh(desc[index].flags) & VRING_DESC_F_INDIRECT) { + unsigned int desc_len; + uint64_t desc_addr; + + if (le32toh(desc[index].len) % sizeof(struct vring_desc)) + die("Invalid size for indirect buffer table"); + + /* loop over the indirect descriptor table */ + desc_addr = le64toh(desc[index].addr); + desc_len = le32toh(desc[index].len); + max = desc_len / sizeof(struct vring_desc); + read_len = desc_len; + desc = vu_gpa_to_va(vdev, &read_len, desc_addr); + if (desc && read_len != desc_len) { + /* Failed to use zero copy */

Follow-up on the question above: could we skip mmap() if we used only this path?

We need to acces guest memory, so no.

...

+ desc = NULL; + if (!virtqueue_read_indirect_desc(vdev, desc_buf, + desc_addr, + desc_len)) + desc = desc_buf; + } + + if (!desc) + die("Invalid indirect buffer table"); + + index = 0; + } + + do { + if (++num_bufs > max) + die("Looped descriptor"); + + if (le16toh(desc[index].flags) & VRING_DESC_F_WRITE) { + min = MIN(le32toh(desc[index].len), len); + vu_log_write(vdev, le64toh(desc[index].addr), min); + len -= min; + } + } while (len > 0 && + (virtqueue_read_next_desc(desc, index, max, &index) == + VIRTQUEUE_READ_DESC_MORE));

It's a bit weird that we could get a negative length because of the do { } while. That is:

while (len > 0) { ... if (virtqueue_read_next_desc(desc, index, max, &index) != VIRTQUEUE_READ_DESC_MORE)) break; }

would have looked more natural/safer to me. But perhaps there's a reason for that.

In fact, it's copied from QEMU, I didn't want to change the code.

...

}

@@ -614,6 +680,8 @@ void vu_queue_fill_by_index(const struct vu_dev *vdev, struct vu_virtq *vq, if (!vq->vring.avail) return;

+ vu_log_queue_fill(vdev, vq, index, len); + idx = (idx + vq->used_idx) % vq->vring.num;

uelem.id = htole32(index); @@ -646,7 +714,9 @@ static inline void vring_used_idx_set(const struct vu_dev *vdev, struct vu_virtq *vq, uint16_t val) { vq->vring.used->idx = htole16(val); - (void)vdev; + vu_log_write(vdev, vq->vring.log_guest_addr + + offsetof(struct vring_used, idx), + sizeof(vq->vring.used->idx));

vq->used_idx = val; } diff --git a/virtio.h b/virtio.h index d95bb07bb913..f572341a0034 100644 --- a/virtio.h +++ b/virtio.h @@ -104,6 +104,8 @@ struct vu_dev_region { * @features: Vhost-user features * @protocol_features: Vhost-user protocol features * @log_call_fd: Eventfd to report logging update + * @log_size: Size of the logging memory region + * @log_table: Base of the logging memory region */ struct vu_dev { struct ctx *context; @@ -113,6 +115,8 @@ struct vu_dev { uint64_t features; uint64_t protocol_features; int log_call_fd; + uint64_t log_size; + uint8_t *log_table; };

/**

Thanks, Laurent

On Thu, 19 Dec 2024 12:13:59 +0100 Laurent Vivier wrote:

Set the file descriptor to use to transfer the backend device state during migration.

Signed-off-by: Laurent Vivier --- epoll_type.h | 2 ++ passt.c | 4 +++ vhost_user.c | 81 +++++++++++++++++++++++++++++++++++++++++++++++++++- virtio.h | 2 ++ vu_common.c | 49 +++++++++++++++++++++++++++++++ vu_common.h | 1 + 6 files changed, 138 insertions(+), 1 deletion(-)

diff --git a/epoll_type.h b/epoll_type.h index f3ef41584757..fd9eac392f77 100644 --- a/epoll_type.h +++ b/epoll_type.h @@ -40,6 +40,8 @@ enum epoll_type { EPOLL_TYPE_VHOST_CMD, /* vhost-user kick event socket */ EPOLL_TYPE_VHOST_KICK, + /* vhost-user migration socket */ + EPOLL_TYPE_VHOST_MIGRATION,

EPOLL_NUM_TYPES, }; diff --git a/passt.c b/passt.c index 957f3d0f4ddc..25d9823739cf 100644 --- a/passt.c +++ b/passt.c @@ -75,6 +75,7 @@ char *epoll_type_str[] = { [EPOLL_TYPE_TAP_LISTEN] = "listening qemu socket", [EPOLL_TYPE_VHOST_CMD] = "vhost-user command socket", [EPOLL_TYPE_VHOST_KICK] = "vhost-user kick socket", + [EPOLL_TYPE_VHOST_MIGRATION] = "vhost-user migration socket", }; static_assert(ARRAY_SIZE(epoll_type_str) == EPOLL_NUM_TYPES, "epoll_type_str[] doesn't match enum epoll_type"); @@ -356,6 +357,9 @@ loop: case EPOLL_TYPE_VHOST_KICK: vu_kick_cb(c.vdev, ref, &now); break; + case EPOLL_TYPE_VHOST_MIGRATION: + vu_migrate(c.vdev, eventmask); + break; default: /* Can't happen */ ASSERT(0); diff --git a/vhost_user.c b/vhost_user.c index 90c46d5b89fd..11b0b447850d 100644 --- a/vhost_user.c +++ b/vhost_user.c @@ -981,6 +981,78 @@ static bool vu_set_vring_enable_exec(struct vu_dev *vdev, return false; }

+/** + * vu_set_migration_watch() -- Add the migration file descriptor to

Single '-' between function name and comment.

+ * to the passt epoll file descriptor + * @vdev: vhost-user device + * @fd: File descriptor to add + * @direction: Direction of the migration (save or load backend state) + */ +static void vu_set_migration_watch(const struct vu_dev *vdev, int fd, + int direction)

Shouldn't direction be uint32?

+{ + union epoll_ref ref = { + .type = EPOLL_TYPE_VHOST_MIGRATION, + .fd = fd, + }; + struct epoll_event ev = { 0 }; + + ev.data.u64 = ref.u64; + switch (direction) { + case VHOST_USER_TRANSFER_STATE_DIRECTION_SAVE: + ev.events = EPOLLOUT; + break; + case VHOST_USER_TRANSFER_STATE_DIRECTION_LOAD: + ev.events = EPOLLIN; + break; + default: + ASSERT(0); + } + + epoll_ctl(vdev->context->epollfd, EPOLL_CTL_ADD, ref.fd, &ev); +} + +/** + * vu_set_device_state_fd_exec() -- Set the device state migration channel

Single '-' between function name and comment.

+ * @vdev: vhost-user device + * @vmsg: vhost-user message + * + * Return: True as the reply contains 0 to indicate success + * and set bit 8 as we don't provide our own fd. + */ +static bool vu_set_device_state_fd_exec(struct vu_dev *vdev, + struct vhost_user_msg *msg) +{ + unsigned int direction = msg->payload.transfer_state.direction; + unsigned int phase = msg->payload.transfer_state.phase; + + if (msg->fd_num != 1) + die("Invalid device_state_fd message"); + + if (phase != VHOST_USER_TRANSFER_STATE_PHASE_STOPPED) + die("Invalid device_state_fd phase: %d", phase); + + if (direction != VHOST_USER_TRANSFER_STATE_DIRECTION_SAVE && + direction != VHOST_USER_TRANSFER_STATE_DIRECTION_LOAD) + die("Invalide device_state_fd direction: %d", direction); + + if (vdev->device_state_fd != -1) { + vu_remove_watch(vdev, vdev->device_state_fd); + close(vdev->device_state_fd); + } + + vdev->device_state_fd = msg->fds[0]; + vdev->device_state_result = -1; + vu_set_migration_watch(vdev, vdev->device_state_fd, direction); + + debug("Got device_state_fd: %d", vdev->device_state_fd); + + /* We don't provide a new fd for the data transfer */ + vmsg_set_reply_u64(msg, VHOST_USER_VRING_NOFD_MASK); + + return true; +} + /** * vu_check_device_state_exec() -- Return device state migration result

Single '-' between function name and comment.

* @vdev: vhost-user device @@ -1019,6 +1091,7 @@ void vu_init(struct ctx *c) } c->vdev->log_table = NULL; c->vdev->log_call_fd = -1; + c->vdev->device_state_fd = -1; c->vdev->device_state_result = -1; }

@@ -1069,7 +1142,12 @@ void vu_cleanup(struct vu_dev *vdev)

vu_close_log(vdev);

- vdev->device_state_result = -1; + if (vdev->device_state_fd != -1) { + vu_remove_watch(vdev, vdev->device_state_fd); + close(vdev->device_state_fd); + vdev->device_state_fd = -1; + vdev->device_state_result = -1; + } }

/** @@ -1100,6 +1178,7 @@ static bool (*vu_handle[VHOST_USER_MAX])(struct vu_dev *vdev, [VHOST_USER_SET_VRING_CALL] = vu_set_vring_call_exec, [VHOST_USER_SET_VRING_ERR] = vu_set_vring_err_exec, [VHOST_USER_SET_VRING_ENABLE] = vu_set_vring_enable_exec, + [VHOST_USER_SET_DEVICE_STATE_FD] = vu_set_device_state_fd_exec, [VHOST_USER_CHECK_DEVICE_STATE] = vu_check_device_state_exec, };

diff --git a/virtio.h b/virtio.h index 512ec1bedcd3..7bef2d274acd 100644 --- a/virtio.h +++ b/virtio.h @@ -106,6 +106,7 @@ struct vu_dev_region { * @log_call_fd: Eventfd to report logging update * @log_size: Size of the logging memory region * @log_table: Base of the logging memory region + * @device_state_fd: Device state migration channel * @device_state_result: Device state migration result */ struct vu_dev { @@ -118,6 +119,7 @@ struct vu_dev { int log_call_fd; uint64_t log_size; uint8_t *log_table; + int device_state_fd; int device_state_result; };

diff --git a/vu_common.c b/vu_common.c index 16e7e76a07f3..3142b585c29f 100644 --- a/vu_common.c +++ b/vu_common.c @@ -281,3 +281,52 @@ err:

return -1; } + +/** + * vu_migrate() -- Send/receive passt insternal state to/from QEMU

Single '-' between function name and comment.

+ * @vdev: vhost-user device + * @events: epoll events + */ +void vu_migrate(struct vu_dev *vdev, uint32_t events) +{ + int ret; + + /* TODO: collect/set passt internal state + * and use vdev->device_state_fd to send/receive it + */

Second and third line are indented with spaces instead of tabs.

+ debug("vu_migrate fd %d events %x", vdev->device_state_fd, events); + if (events & EPOLLOUT) { + debug("Saving backend state"); + + /* send some stuff */ + ret = write(vdev->device_state_fd, "PASST", 6);

So, yeah, I still have my open questions/concerns here (essentially: "what if write() returns 5?"), but they can very well fit under the TODO above. We might need to refactor this anyway, perhaps even use writev(). So I think it's totally fine by now.

+ /* value to be returned by VHOST_USER_CHECK_DEVICE_STATE */ + vdev->device_state_result = ret == -1 ? -1 : 0;

Shouldn't we err() on error? Even right now for development purposes?

+ /* Closing the file descriptor signals the end of transfer */ + epoll_ctl(vdev->context->epollfd, EPOLL_CTL_DEL, + vdev->device_state_fd, NULL); + close(vdev->device_state_fd); + vdev->device_state_fd = -1; + } else if (events & EPOLLIN) { + char buf[6]; + + debug("Loading backend state"); + /* read some stuff */ + ret = read(vdev->device_state_fd, buf, sizeof(buf)); + /* value to be returned by VHOST_USER_CHECK_DEVICE_STATE */ + if (ret != sizeof(buf)) { + vdev->device_state_result = -1;

Same here.

+ } else { + ret = strncmp(buf, "PASST", sizeof(buf)); + vdev->device_state_result = ret == 0 ? 0 : -1; + } + } else if (events & EPOLLHUP) { + debug("Closing migration channel"); + + /* The end of file signals the end of the transfer. */ + epoll_ctl(vdev->context->epollfd, EPOLL_CTL_DEL, + vdev->device_state_fd, NULL); + close(vdev->device_state_fd); + vdev->device_state_fd = -1; + } +} diff --git a/vu_common.h b/vu_common.h index bd70faf3e226..d56c021ab0f9 100644 --- a/vu_common.h +++ b/vu_common.h @@ -57,4 +57,5 @@ void vu_flush(const struct vu_dev *vdev, struct vu_virtq *vq, void vu_kick_cb(struct vu_dev *vdev, union epoll_ref ref, const struct timespec *now); int vu_send_single(const struct ctx *c, const void *buf, size_t size); +void vu_migrate(struct vu_dev *vdev, uint32_t events); #endif /* VU_COMMON_H */

The rest of the series looks good to me. I can also fix up all the formal things on merge, but I guess you want to respin (at least for the "fake RARP" thing) anyway? -- Stefano