[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] [PATCH v4 RFC 9/8] nbd: Minimal structured read for client
|
From: |
Vladimir Sementsov-Ogievskiy |
|
Subject: |
[Qemu-devel] [PATCH v4 RFC 9/8] nbd: Minimal structured read for client |
|
Date: |
Tue, 17 Oct 2017 15:57:10 +0300 |
Minimal implementation: for structured error only error_report error
message.
Signed-off-by: Vladimir Sementsov-Ogievskiy <address@hidden>
---
it's an updated
[PATCH v3 13/13] nbd: Minimal structured read for client
https://lists.gnu.org/archive/html/qemu-devel/2017-10/msg02760.html
from
[PATCH v3 00/13] nbd minimal structured read
https://lists.gnu.org/archive/html/qemu-devel/2017-10/msg02755.html
and now it is based on
[PATCH v4 8/8] nbd: Move nbd_read() to common header
https://lists.gnu.org/archive/html/qemu-devel/2017-10/msg03489.html
from
[PATCH v4 0/8] nbd mimimal structured read
https://lists.gnu.org/archive/html/qemu-devel/2017-10/msg03305.html
v4:
- add Error messages and errp to most of new functions
(I try to return save first occurred error from structured reply,
but rewrite it if fatal error occurred)
- add comment to NBDExportInfo
- move nbd_reply_type_lookup prot. to public .h (I think it should be
merged into some previous patch of v4)
- remove ''#include "qemu/error-report.h"'
- rewrite payload_advance* helpers
- add comments about qemu_vfree for *payload
- _SREP_ -> _REPLY_
- move checking NBD_REPLY_FLAG_DONE for NBD_REPLY_TYPE_NONE
- do not negotiate structured reply from qemu-nbd client thread
(make client->info.structured_reply in-out)
RFC: I don't like resulting error handling: it is rather confusing,
requires a lot of definitions of "local_err" which looks bad. Propagating,
setting error and then drop it for errp=NULL case - all it looks not
very effective. Any ideas?
include/block/nbd.h | 12 ++
nbd/nbd-internal.h | 1 -
block/nbd-client.c | 483 ++++++++++++++++++++++++++++++++++++++++++++++++----
nbd/client.c | 10 ++
4 files changed, 474 insertions(+), 32 deletions(-)
diff --git a/include/block/nbd.h b/include/block/nbd.h
index da6e305dd5..92d1723d7c 100644
--- a/include/block/nbd.h
+++ b/include/block/nbd.h
@@ -197,6 +197,11 @@ enum {
#define NBD_REPLY_TYPE_ERROR NBD_REPLY_ERR(1)
#define NBD_REPLY_TYPE_ERROR_OFFSET NBD_REPLY_ERR(2)
+static inline bool nbd_reply_type_is_error(int type)
+{
+ return type & (1 << 15);
+}
+
/* NBD errors are based on errno numbers, so there is a 1:1 mapping,
* but only a limited set of errno values is specified in the protocol.
* Everything else is squashed to EINVAL.
@@ -214,6 +219,11 @@ enum {
struct NBDExportInfo {
/* Set by client before nbd_receive_negotiate() */
bool request_sizes;
+
+ /* In-out fields, set by client before nbd_receive_negotiate() and
+ * updated by server results during nbd_receive_negotiate() */
+ bool structured_reply;
+
/* Set by server results during nbd_receive_negotiate() */
uint64_t size;
uint16_t flags;
@@ -284,4 +294,6 @@ static inline bool nbd_reply_is_structured(NBDReply *reply)
return reply->magic == NBD_STRUCTURED_REPLY_MAGIC;
}
+const char *nbd_reply_type_lookup(uint16_t type);
+
#endif
diff --git a/nbd/nbd-internal.h b/nbd/nbd-internal.h
index b64eb1cc9b..eeff78d3c9 100644
--- a/nbd/nbd-internal.h
+++ b/nbd/nbd-internal.h
@@ -104,7 +104,6 @@ const char *nbd_opt_lookup(uint32_t opt);
const char *nbd_rep_lookup(uint32_t rep);
const char *nbd_info_lookup(uint16_t info);
const char *nbd_cmd_lookup(uint16_t info);
-const char *nbd_reply_type_lookup(uint16_t type);
const char *nbd_err_lookup(int err);
int nbd_drop(QIOChannel *ioc, size_t size, Error **errp);
diff --git a/block/nbd-client.c b/block/nbd-client.c
index 58493b7ac4..9d9e75643e 100644
--- a/block/nbd-client.c
+++ b/block/nbd-client.c
@@ -93,7 +93,7 @@ static coroutine_fn void nbd_read_reply_entry(void *opaque)
if (i >= MAX_NBD_REQUESTS ||
!s->requests[i].coroutine ||
!s->requests[i].receiving ||
- nbd_reply_is_structured(&s->reply))
+ (nbd_reply_is_structured(&s->reply) && !s->info.structured_reply))
{
break;
}
@@ -181,75 +181,486 @@ err:
return rc;
}
-static int nbd_co_receive_reply(NBDClientSession *s,
- uint64_t handle,
- QEMUIOVector *qiov)
+static inline uint16_t payload_advance16(uint8_t **payload)
+{
+ *payload += 2;
+ return lduw_be_p(*payload - 2);
+}
+
+static inline uint32_t payload_advance32(uint8_t **payload)
+{
+ *payload += 4;
+ return lduw_be_p(*payload - 4);
+}
+
+static inline uint64_t payload_advance64(uint8_t **payload)
+{
+ *payload += 8;
+ return lduw_be_p(*payload - 8);
+}
+
+static int nbd_parse_offset_hole_payload(NBDStructuredReplyChunk *chunk,
+ uint8_t *payload, QEMUIOVector *qiov,
+ Error **errp)
+{
+ uint64_t offset;
+ uint32_t hole_size;
+
+ if (chunk->length != sizeof(offset) + sizeof(hole_size)) {
+ error_setg(errp, "Protocol error: invalid payload for "
+ "NBD_REPLY_TYPE_OFFSET_HOLE");
+ return -EINVAL;
+ }
+
+ offset = payload_advance64(&payload);
+ hole_size = payload_advance32(&payload);
+
+ if (offset + hole_size > qiov->size) {
+ error_setg(errp, "Protocol error: server sent chunk exceeding
requested"
+ " region");
+ return -EINVAL;
+ }
+
+ qemu_iovec_memset(qiov, offset, 0, hole_size);
+
+ return 0;
+}
+
+/* nbd_parse_error_payload
+ * on success @errp contains message describing nbd error reply
+ */
+static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk,
+ uint8_t *payload, int *request_ret,
+ Error **errp)
+{
+ uint32_t error;
+ uint16_t message_size;
+
+ assert(chunk->type & (1 << 15));
+
+ if (chunk->length < sizeof(error) + sizeof(message_size)) {
+ error_setg(errp,
+ "Protocol error: invalid payload for structured error");
+ return -EINVAL;
+ }
+
+ error = nbd_errno_to_system_errno(payload_advance32(&payload));
+ if (error == 0) {
+ error_setg(errp, "Protocol error: server sent structured error chunk"
+ "with error = 0");
+ return -EINVAL;
+ }
+
+ *request_ret = error;
+ message_size = payload_advance16(&payload);
+ error_setg_errno(errp, error, "%.*s", message_size, payload);
+
+ /* TODO handle ERROR_OFFSET */
+
+ return 0;
+}
+
+static int nbd_co_receive_offset_data_payload(NBDClientSession *s,
+ QEMUIOVector *qiov, Error **errp)
+{
+ QEMUIOVector sub_qiov;
+ uint64_t offset;
+ size_t data_size;
+ int ret;
+ NBDStructuredReplyChunk *chunk = &s->reply.structured;
+
+ assert(nbd_reply_is_structured(&s->reply));
+
+ if (chunk->length < sizeof(offset)) {
+ error_setg(errp, "Protocol error: invalid payload for "
+ "NBD_REPLY_TYPE_OFFSET_DATA");
+ return -EINVAL;
+ }
+
+ if (nbd_read(s->ioc, &offset, sizeof(offset), errp) < 0) {
+ return -EIO;
+ }
+ be64_to_cpus(&offset);
+
+ data_size = chunk->length - sizeof(offset);
+ if (offset + data_size > qiov->size) {
+ error_setg(errp, "Protocol error: server sent chunk exceeding
requested"
+ " region");
+ return -EINVAL;
+ }
+
+ qemu_iovec_init(&sub_qiov, qiov->niov);
+ qemu_iovec_concat(&sub_qiov, qiov, offset, data_size);
+ ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp);
+ qemu_iovec_destroy(&sub_qiov);
+
+ return ret < 0 ? -EIO : 0;
+}
+
+#define NBD_MAX_MALLOC_PAYLOAD 1000
+/* nbd_co_receive_structured_payload
+ * The resulting pointer stored in @payload requires qemu_vfree() to free it.
+ */
+static coroutine_fn int nbd_co_receive_structured_payload(
+ NBDClientSession *s, void **payload, Error **errp)
+{
+ int ret;
+ uint32_t len;
+
+ assert(nbd_reply_is_structured(&s->reply));
+
+ len = s->reply.structured.length;
+
+ if (len == 0) {
+ return 0;
+ }
+
+ if (payload == NULL) {
+ error_setg(errp, "Unexpected structured payload");
+ return -EINVAL;
+ }
+
+ if (len > NBD_MAX_MALLOC_PAYLOAD) {
+ error_setg(errp, "Too large payload");
+ return -EINVAL;
+ }
+
+ *payload = qemu_memalign(8, len);
+ ret = nbd_read(s->ioc, *payload, len, errp);
+ if (ret < 0) {
+ qemu_vfree(*payload);
+ *payload = NULL;
+ return ret;
+ }
+
+ return 0;
+}
+
+/* nbd_co_do_receive_one_chunk
+ * for simple reply:
+ * set request_ret to received reply error
+ * if qiov is not NULL: read payload to @qiov
+ * for structured reply chunk:
+ * if error chunk: read payload, set @request_ret, do not set @payload
+ * else if offset_data chunk: read payload data to @qiov, do not set @payload
+ * else: read payload to @payload
+ *
+ * The pointer stored in @payload requires qemu_vfree() to free it.
+ * If function fails @errp containts corresponding error message. If function
+ * doesn't fail but reply contains error @request_ret is set to corresponding
+ * return code and errp is set to corresponding error message.
+ */
+static coroutine_fn int nbd_co_do_receive_one_chunk(
+ NBDClientSession *s, uint64_t handle, bool only_structured,
+ int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp)
{
int ret;
int i = HANDLE_TO_INDEX(s, handle);
+ void *local_payload = NULL;
+ NBDStructuredReplyChunk *chunk;
+
+ if (payload) {
+ *payload = NULL;
+ }
+ *request_ret = 0;
/* Wait until we're woken up by nbd_read_reply_entry. */
s->requests[i].receiving = true;
qemu_coroutine_yield();
s->requests[i].receiving = false;
if (!s->ioc || s->quit) {
- ret = -EIO;
- } else {
- assert(s->reply.handle == handle);
- ret = -nbd_errno_to_system_errno(s->reply.simple.error);
- if (qiov && ret == 0) {
- if (qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
- NULL) < 0) {
- ret = -EIO;
- s->quit = true;
+ error_setg(errp, "Connection closed");
+ return -EIO;
+ }
+
+ assert(s->reply.handle == handle);
+
+ if (nbd_reply_is_simple(&s->reply)) {
+ if (only_structured) {
+ error_setg(errp, "Protocol error: simple reply when structured"
+ "reply chunk was expected");
+ return -EINVAL;
+ }
+
+ *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error);
+ if (*request_ret < 0 || !qiov) {
+ if (*request_ret < 0) {
+ error_setg_errno(errp, -*request_ret, "Request failed");
}
+ return 0;
}
- /* Tell the read handler to read another header. */
- s->reply.handle = 0;
+ return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
+ errp) < 0 ? -EIO : 0;
}
- s->requests[i].coroutine = NULL;
+ /* handle structured reply chunk */
+ assert(s->info.structured_reply);
+ chunk = &s->reply.structured;
+
+ if (chunk->type == NBD_REPLY_TYPE_NONE) {
+ if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) {
+ error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk
without"
+ "NBD_REPLY_FLAG_DONE flag set");
+ return -EINVAL;
+ }
+ return 0;
+ }
+
+ if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) {
+ if (!qiov) {
+ error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
+ return -EINVAL;
+ }
+
+ return nbd_co_receive_offset_data_payload(s, qiov, errp);
+ }
+
+ if (nbd_reply_type_is_error(chunk->type)) {
+ payload = &local_payload;
+ }
+
+ ret = nbd_co_receive_structured_payload(s, payload, errp);
+ if (ret < 0) {
+ return ret;
+ }
+
+ if (nbd_reply_type_is_error(chunk->type)) {
+ ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp);
+ qemu_vfree(local_payload);
+ return ret;
+ }
+
+ return 0;
+}
+
+/* nbd_co_receive_one_chunk
+ * Read reply, wake up read_reply_co and set s->quit if needed.
+ * Return value is a fatal error code or normal nbd reply error code
+ *
+ * The pointer stored in @payload requires qemu_vfree() to free it.
+ */
+static coroutine_fn int nbd_co_receive_one_chunk(
+ NBDClientSession *s, uint64_t handle, bool only_structured,
+ QEMUIOVector *qiov, NBDReply *reply, void **payload, Error **errp)
+{
+ int request_ret;
+ int ret = nbd_co_do_receive_one_chunk(s, handle, only_structured,
+ &request_ret, qiov, payload, errp);
+
+ if (ret < 0) {
+ s->quit = true;
+ } else {
+ /* For assert at loop start in nbd_read_reply_entry */
+ if (reply) {
+ *reply = s->reply;
+ }
+ s->reply.handle = 0;
+ ret = request_ret;
+ }
- /* Kick the read_reply_co to get the next reply. */
if (s->read_reply_co) {
aio_co_wake(s->read_reply_co);
}
+ return ret;
+}
+
+typedef struct NBDReplyChunkIter {
+ int ret;
+ Error *err;
+ bool done, only_structured;
+} NBDReplyChunkIter;
+
+static void nbd_iter_error(NBDReplyChunkIter *iter, bool fatal,
+ int ret, Error **local_err)
+{
+ assert(ret < 0);
+
+ if (fatal || iter->ret == 0) {
+ if (iter->ret != 0) {
+ error_free(iter->err);
+ iter->err = NULL;
+ }
+ iter->ret = ret;
+ error_propagate(&iter->err, *local_err);
+ } else {
+ error_free(*local_err);
+ }
+
+ *local_err = NULL;
+}
+
+/* NBD_FOREACH_REPLY_CHUNK
+ * The pointer stored in @payload requires qemu_vfree() to free it.
+ */
+#define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
+ qiov, reply, payload) \
+ for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
+ nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
+
+/* nbd_reply_chunk_iter_receive
+ * The pointer stored in @payload requires qemu_vfree() to free it.
+ */
+static bool nbd_reply_chunk_iter_receive(NBDClientSession *s,
+ NBDReplyChunkIter *iter,
+ uint64_t handle,
+ QEMUIOVector *qiov, NBDReply *reply,
+ void **payload)
+{
+ int ret;
+ NBDReply local_reply;
+ NBDStructuredReplyChunk *chunk;
+ Error *local_err = NULL;
+ if (s->quit) {
+ error_setg(&local_err, "Connection closed");
+ nbd_iter_error(iter, true, -EIO, &local_err);
+ goto break_loop;
+ }
+
+ if (iter->done) {
+ /* Previous iteration was last. */
+ goto break_loop;
+ }
+
+ if (reply == NULL) {
+ reply = &local_reply;
+ }
+
+ ret = nbd_co_receive_one_chunk(s, handle, iter->only_structured,
+ qiov, reply, payload, &local_err);
+ if (ret < 0) {
+ /* If it is a fatal error s->quit is set by nbd_co_receive_one_chunk */
+ nbd_iter_error(iter, s->quit, ret, &local_err);
+ }
+
+ /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
+ if (nbd_reply_is_simple(&s->reply) || s->quit) {
+ goto break_loop;
+ }
+
+ chunk = &reply->structured;
+ iter->only_structured = true;
+
+ if (chunk->type == NBD_REPLY_TYPE_NONE) {
+ /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk
*/
+ assert(chunk->flags & NBD_REPLY_FLAG_DONE);
+ goto break_loop;
+ }
+
+ if (chunk->flags & NBD_REPLY_FLAG_DONE) {
+ /* This iteration is last. */
+ iter->done = true;
+ }
+
+ /* Execute the loop body */
+ return true;
+
+break_loop:
+ s->requests[HANDLE_TO_INDEX(s, handle)].coroutine = NULL;
+
qemu_co_mutex_lock(&s->send_mutex);
s->in_flight--;
qemu_co_queue_next(&s->free_sema);
qemu_co_mutex_unlock(&s->send_mutex);
- return ret;
+ return false;
}
-static int nbd_co_request(BlockDriverState *bs,
- NBDRequest *request,
- QEMUIOVector *qiov)
+static int nbd_co_receive_return_code(NBDClientSession *s, uint64_t handle,
+ Error **errp)
+{
+ NBDReplyChunkIter iter;
+
+ NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, NULL, NULL) {
+ /* nbd_reply_chunk_iter_receive does all the work */
+ ;
+ }
+
+ error_propagate(errp, iter.err);
+ return iter.ret;
+}
+
+static int nbd_co_receive_cmdread_reply(NBDClientSession *s, uint64_t handle,
+ QEMUIOVector *qiov, Error **errp)
+{
+ NBDReplyChunkIter iter;
+ NBDReply reply;
+ void *payload = NULL;
+ Error *local_err = NULL;
+
+ NBD_FOREACH_REPLY_CHUNK(s, iter, handle, s->info.structured_reply,
+ qiov, &reply, &payload)
+ {
+ int ret;
+ NBDStructuredReplyChunk *chunk = &reply.structured;
+
+ assert(nbd_reply_is_structured(&reply));
+
+ switch (chunk->type) {
+ case NBD_REPLY_TYPE_OFFSET_DATA:
+ /* special cased in nbd_co_receive_one_chunk, data is already
+ * in qiov */
+ break;
+ case NBD_REPLY_TYPE_OFFSET_HOLE:
+ ret = nbd_parse_offset_hole_payload(&reply.structured, payload,
+ qiov, &local_err);
+ if (ret < 0) {
+ s->quit = true;
+ nbd_iter_error(&iter, true, ret, &local_err);
+ }
+ break;
+ default:
+ if (!nbd_reply_type_is_error(chunk->type)) {
+ /* not allowed reply type */
+ s->quit = true;
+ error_setg(&local_err,
+ "Unexpected reply type: %d (%s) for CMD_READ",
+ chunk->type, nbd_reply_type_lookup(chunk->type));
+ nbd_iter_error(&iter, true, -EINVAL, &local_err);
+ }
+ }
+
+ qemu_vfree(payload);
+ payload = NULL;
+ }
+
+ error_propagate(errp, iter.err);
+ return iter.ret;
+}
+
+static int nbd_co_request(BlockDriverState *bs, NBDRequest *request,
+ QEMUIOVector *write_qiov)
{
- NBDClientSession *client = nbd_get_client_session(bs);
int ret;
+ Error *local_err = NULL;
+ NBDClientSession *client = nbd_get_client_session(bs);
- if (qiov) {
- assert(request->type == NBD_CMD_WRITE || request->type ==
NBD_CMD_READ);
- assert(request->len == iov_size(qiov->iov, qiov->niov));
+ assert(request->type != NBD_CMD_READ);
+ if (write_qiov) {
+ assert(request->type == NBD_CMD_WRITE);
+ assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
} else {
- assert(request->type != NBD_CMD_WRITE && request->type !=
NBD_CMD_READ);
+ assert(request->type != NBD_CMD_WRITE);
}
- ret = nbd_co_send_request(bs, request,
- request->type == NBD_CMD_WRITE ? qiov : NULL);
+ ret = nbd_co_send_request(bs, request, write_qiov);
if (ret < 0) {
return ret;
}
- return nbd_co_receive_reply(client, request->handle,
- request->type == NBD_CMD_READ ? qiov : NULL);
+ ret = nbd_co_receive_return_code(client, request->handle, &local_err);
+ if (ret < 0) {
+ error_report_err(local_err);
+ }
+ return ret;
}
int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
uint64_t bytes, QEMUIOVector *qiov, int flags)
{
+ int ret;
+ Error *local_err = NULL;
+ NBDClientSession *client = nbd_get_client_session(bs);
NBDRequest request = {
.type = NBD_CMD_READ,
.from = offset,
@@ -259,7 +670,16 @@ int nbd_client_co_preadv(BlockDriverState *bs, uint64_t
offset,
assert(bytes <= NBD_MAX_BUFFER_SIZE);
assert(!flags);
- return nbd_co_request(bs, &request, qiov);
+ ret = nbd_co_send_request(bs, &request, NULL);
+ if (ret < 0) {
+ return ret;
+ }
+
+ ret = nbd_co_receive_cmdread_reply(client, request.handle, qiov,
&local_err);
+ if (ret < 0) {
+ error_report_err(local_err);
+ }
+ return ret;
}
int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset,
@@ -381,6 +801,7 @@ int nbd_client_init(BlockDriverState *bs,
qio_channel_set_blocking(QIO_CHANNEL(sioc), true, NULL);
client->info.request_sizes = true;
+ client->info.structured_reply = true;
ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), export,
tlscreds, hostname,
&client->ioc, &client->info, errp);
diff --git a/nbd/client.c b/nbd/client.c
index eeb23ae934..0297484264 100644
--- a/nbd/client.c
+++ b/nbd/client.c
@@ -684,6 +684,16 @@ int nbd_receive_negotiate(QIOChannel *ioc, const char
*name,
if (fixedNewStyle) {
int result;
+ if (info->structured_reply) {
+ result = nbd_request_simple_option(ioc,
+ NBD_OPT_STRUCTURED_REPLY,
+ errp);
+ if (result < 0) {
+ goto fail;
+ }
+ info->structured_reply = result == 1;
+ }
+
/* Try NBD_OPT_GO first - if it works, we are done (it
* also gives us a good message if the server requires
* TLS). If it is not available, fall back to
--
2.11.1
- [Qemu-devel] [PATCH v4 8/8] nbd: Move nbd_read() to common header, (continued)