qemu-block
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-block] [PATCH v3 5/7] block: add throttle block filter driver


From: Manos Pitsidianakis
Subject: Re: [Qemu-block] [PATCH v3 5/7] block: add throttle block filter driver
Date: Thu, 3 Aug 2017 14:48:56 +0300
User-agent: NeoMutt/20170609-57-1e93be (1.8.3)

On Thu, Aug 03, 2017 at 10:07:41AM +0200, Kevin Wolf wrote:
Am 31.07.2017 um 11:54 hat Manos Pitsidianakis geschrieben:
+/* Extract ThrottleConfig options. Assumes cfg is initialized and will be
+ * checked for validity.
+ */
+static int throttle_extract_options(QemuOpts *opts, ThrottleConfig *cfg,
+                                     Error **errp)
+{
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_TOTAL)) {
+        cfg->buckets[THROTTLE_BPS_TOTAL].avg =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_TOTAL,
+                                0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_READ)) {
+        cfg->buckets[THROTTLE_BPS_READ].avg  =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_READ,
+                                0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_WRITE)) {
+        cfg->buckets[THROTTLE_BPS_WRITE].avg =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_WRITE,
+                                0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_TOTAL)) {
+        cfg->buckets[THROTTLE_OPS_TOTAL].avg =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_TOTAL,
+                                0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_READ)) {
+        cfg->buckets[THROTTLE_OPS_READ].avg =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_READ,
+                                0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_WRITE)) {
+        cfg->buckets[THROTTLE_OPS_WRITE].avg =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_WRITE,
+                                0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_TOTAL_MAX)) {
+        cfg->buckets[THROTTLE_BPS_TOTAL].max =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_BPS_TOTAL_MAX, 0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_READ_MAX)) {
+        cfg->buckets[THROTTLE_BPS_READ].max  =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_BPS_READ_MAX, 0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_WRITE_MAX)) {
+        cfg->buckets[THROTTLE_BPS_WRITE].max =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_BPS_WRITE_MAX, 0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_TOTAL_MAX)) {
+        cfg->buckets[THROTTLE_OPS_TOTAL].max =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_IOPS_TOTAL_MAX, 0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_READ_MAX)) {
+        cfg->buckets[THROTTLE_OPS_READ].max =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_IOPS_READ_MAX, 0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_WRITE_MAX)) {
+        cfg->buckets[THROTTLE_OPS_WRITE].max =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_IOPS_WRITE_MAX, 0);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_TOTAL_MAX_LENGTH)) 
{
+        if (qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_BPS_TOTAL_MAX_LENGTH, 1) > UINT_MAX) {
+            error_setg(errp, "%s value must be in the range [0, %u]",
+                       THROTTLE_OPT_PREFIX QEMU_OPT_BPS_TOTAL_MAX_LENGTH,
+                       UINT_MAX);
+            return -1;
+        }
+        cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_BPS_TOTAL_MAX_LENGTH, 1);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_READ_MAX_LENGTH)) {
+        if (qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_BPS_READ_MAX_LENGTH, 1) > UINT_MAX) {
+            error_setg(errp, "%s must be in the range [0, %u]",
+                       THROTTLE_OPT_PREFIX QEMU_OPT_BPS_READ_MAX_LENGTH,
+                       UINT_MAX);
+            return -1;
+        }
+        cfg->buckets[THROTTLE_BPS_READ].burst_length  =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_BPS_READ_MAX_LENGTH, 1);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_BPS_WRITE_MAX_LENGTH)) 
{
+        if (qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_BPS_WRITE_MAX_LENGTH, 1) > UINT_MAX) {
+            error_setg(errp, "%s must be in the range [0, %u]",
+                       THROTTLE_OPT_PREFIX QEMU_OPT_BPS_WRITE_MAX_LENGTH,
+                       UINT_MAX);
+            return -1;
+        }
+        cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_BPS_WRITE_MAX_LENGTH, 1);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX 
QEMU_OPT_IOPS_TOTAL_MAX_LENGTH)) {
+        if (qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_IOPS_TOTAL_MAX_LENGTH, 1) > UINT_MAX) 
{
+            error_setg(errp, "%s must be in the range [0, %u]",
+                       THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_TOTAL_MAX_LENGTH,
+                       UINT_MAX);
+            return -1;
+        }
+        cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_IOPS_TOTAL_MAX_LENGTH, 1);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_READ_MAX_LENGTH)) 
{
+        if (qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_IOPS_READ_MAX_LENGTH, 1) > UINT_MAX) {
+            error_setg(errp, "%s must be in the range [0, %u]",
+                       THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_READ_MAX_LENGTH,
+                       UINT_MAX);
+            return -1;
+        }
+        cfg->buckets[THROTTLE_OPS_READ].burst_length =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_IOPS_READ_MAX_LENGTH, 1);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX 
QEMU_OPT_IOPS_WRITE_MAX_LENGTH)) {
+        if (qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_IOPS_WRITE_MAX_LENGTH, 1) > UINT_MAX) 
{
+            error_setg(errp, "%s must be in the range [0, %u]",
+                       THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_WRITE_MAX_LENGTH,
+                       UINT_MAX);
+            return -1;
+        }
+        cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX
+                                QEMU_OPT_IOPS_WRITE_MAX_LENGTH, 1);
+    }
+    if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_SIZE)) {
+        cfg->op_size =
+            qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX QEMU_OPT_IOPS_SIZE,
+                                0);
+    }
+    return 0;
+}

This function is very repetitive, but each block is long enough that
you have to look closely to review whether the right constants are used
everywhere.

Maybe this could become a bit more readable with a macro or two?

How about this?


#   #define IF_OPT_SET(rvalue,opt_name) \
#       if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX opt_name)) { \
# rvalue = qemu_opt_get_number(opts, THROTTLE_OPT_PREFIX opt_name, 0); }
#       IF_OPT_SET(cfg->buckets[THROTTLE_BPS_TOTAL].avg, QEMU_OPT_BPS_TOTAL);
#       IF_OPT_SET(cfg->buckets[THROTTLE_BPS_READ].avg, QEMU_OPT_BPS_READ);
#       IF_OPT_SET(cfg->buckets[THROTTLE_BPS_WRITE].avg, QEMU_OPT_BPS_WRITE);
#       IF_OPT_SET(cfg->buckets[THROTTLE_OPS_TOTAL].avg, QEMU_OPT_IOPS_TOTAL);
#       IF_OPT_SET(cfg->buckets[THROTTLE_OPS_READ].avg, QEMU_OPT_IOPS_READ);
#       IF_OPT_SET(cfg->buckets[THROTTLE_OPS_WRITE].avg, QEMU_OPT_IOPS_WRITE);
#       IF_OPT_SET(cfg->buckets[THROTTLE_BPS_TOTAL].max, 
QEMU_OPT_BPS_TOTAL_MAX);
#       IF_OPT_SET(cfg->buckets[THROTTLE_BPS_READ].max , QEMU_OPT_BPS_READ_MAX);
#       IF_OPT_SET(cfg->buckets[THROTTLE_BPS_WRITE].max, 
QEMU_OPT_BPS_WRITE_MAX);
#       IF_OPT_SET(cfg->buckets[THROTTLE_OPS_TOTAL].max, 
QEMU_OPT_IOPS_TOTAL_MAX);
#       IF_OPT_SET(cfg->buckets[THROTTLE_OPS_READ].max, QEMU_OPT_IOPS_READ_MAX);
#       IF_OPT_SET(cfg->buckets[THROTTLE_OPS_WRITE].max, 
QEMU_OPT_IOPS_WRITE_MAX);
#       IF_OPT_SET(cfg->op_size, QEMU_OPT_IOPS_SIZE);

#   #define IF_OPT_UINT_SET(rvalue,opt_name) \
#       if (qemu_opt_get(opts, THROTTLE_OPT_PREFIX opt_name)) { \
#           if (qemu_opt_get_number(opts,  \
#                       THROTTLE_OPT_PREFIX opt_name, 1) > UINT_MAX) { \
#               error_setg(errp, "%s value must be in the range [0, %u]", \
#                          THROTTLE_OPT_PREFIX opt_name, UINT_MAX); \
#               return -1; \
#           } \
#           rvalue = qemu_opt_get_number(opts, opt_name, 1); \
#       }

#       IF_OPT_UINT_SET(cfg->buckets[THROTTLE_BPS_TOTAL].burst_length,
#                       QEMU_OPT_BPS_TOTAL_MAX_LENGTH);
#       IF_OPT_UINT_SET(cfg->buckets[THROTTLE_BPS_READ].burst_length,
#                       QEMU_OPT_BPS_READ_MAX_LENGTH);
#       IF_OPT_UINT_SET(cfg->buckets[THROTTLE_BPS_WRITE].burst_length,
#                       QEMU_OPT_BPS_WRITE_MAX_LENGTH);
#       IF_OPT_UINT_SET(cfg->buckets[THROTTLE_OPS_TOTAL].burst_length,
#                       QEMU_OPT_IOPS_TOTAL_MAX_LENGTH);
#       IF_OPT_UINT_SET(cfg->buckets[THROTTLE_OPS_READ].burst_length,
#                       QEMU_OPT_IOPS_READ_MAX_LENGTH);
#       IF_OPT_UINT_SET(cfg->buckets[THROTTLE_OPS_WRITE].burst_length,
#                       QEMU_OPT_IOPS_WRITE_MAX_LENGTH);



+static int throttle_configure_tgm(BlockDriverState *bs,
+                                  ThrottleGroupMember *tgm,
+                                  QDict *options, Error **errp)
+{
+    int ret;
+    ThrottleConfig cfg;
+    const char *group_name = NULL;
+    Error *local_err = NULL;
+    QemuOpts *opts = qemu_opts_create(&throttle_opts, NULL, 0, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return -EINVAL;
+    }

As Stefan said, qemu_opts_create() can't fail if you pass NULL for id,
so you can just use &error_abort.

+
+    qemu_opts_absorb_qdict(opts, options, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        goto err;
+    }
+
+    /* If no name is specified, an anonymous group will be created */
+    group_name = qemu_opt_get(opts, QEMU_OPT_THROTTLE_GROUP_NAME);
+
+    /* Register membership to group with name group_name */
+    throttle_group_register_tgm(tgm, group_name, bdrv_get_aio_context(bs));

The documentation of throttle_group_register_tgm() suggests that you
have to pass a string here, but group_name can be NULL if the option
wasn't given. Probably just means that the comment for that function
needs to be updated.

+    /* Copy previous configuration */
+    throttle_group_get_config(tgm, &cfg);
+
+    /* Change limits if user has specified them */
+    if (throttle_extract_options(opts, &cfg, errp) ||
+        !throttle_is_valid(&cfg, errp)) {
+        throttle_group_unregister_tgm(tgm);
+        goto err;
+    }
+    /* Update group configuration */
+    throttle_group_config(tgm, &cfg);
+
+    ret = 0;
+    goto fin;
+
+err:
+    ret = -EINVAL;
+fin:
+    qemu_opts_del(opts);
+    return ret;
+}
+
+static int throttle_open(BlockDriverState *bs, QDict *options,
+                            int flags, Error **errp)
+{
+    ThrottleGroupMember *tgm = bs->opaque;
+
+    bs->file = bdrv_open_child(NULL, options, "file",
+                                    bs, &child_file, false, errp);

Indentation is off.

+    if (!bs->file) {
+        return -EINVAL;
+    }
+
+    return throttle_configure_tgm(bs, tgm, options, errp);
+}
+
+static void throttle_close(BlockDriverState *bs)
+{
+    ThrottleGroupMember *tgm = bs->opaque;
+    throttle_group_unregister_tgm(tgm);
+}
+
+
+static int64_t throttle_getlength(BlockDriverState *bs)
+{
+    return bdrv_getlength(bs->file->bs);
+}
+
+
+static int coroutine_fn throttle_co_preadv(BlockDriverState *bs,
+                                           uint64_t offset, uint64_t bytes,
+                                           QEMUIOVector *qiov, int flags)
+{
+
+    ThrottleGroupMember *tgm = bs->opaque;
+    throttle_group_co_io_limits_intercept(tgm, bytes, false);
+
+    return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
+}
+
+static int coroutine_fn throttle_co_pwritev(BlockDriverState *bs,
+                                            uint64_t offset, uint64_t bytes,
+                                            QEMUIOVector *qiov, int flags)
+{
+    ThrottleGroupMember *tgm = bs->opaque;
+    throttle_group_co_io_limits_intercept(tgm, bytes, true);
+
+    return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
+}

I think we want to set BlockDriver.supported_write_flags so that passing
down flags is actually of any use.

Can you explain what you mean? Do you mean doing this in throttle_open():

   bs->supported_write_flags = bs->file->bs->supported_write_flags;
   bs->supported_zero_flags = bs->file->bs->supported_zero_flags;

+
+static int coroutine_fn throttle_co_pwrite_zeroes(BlockDriverState *bs,
+        int64_t offset, int bytes, BdrvRequestFlags flags)
+{
+    ThrottleGroupMember *tgm = bs->opaque;
+    throttle_group_co_io_limits_intercept(tgm, bytes, true);
+
+    return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
+}

The same is true for BlockDriver.supported_zero_flags.

+static int coroutine_fn throttle_co_pdiscard(BlockDriverState *bs,
+        int64_t offset, int bytes)
+{
+    ThrottleGroupMember *tgm = bs->opaque;
+    throttle_group_co_io_limits_intercept(tgm, bytes, true);
+
+    return bdrv_co_pdiscard(bs->file->bs, offset, bytes);
+}
+
+static int throttle_co_flush(BlockDriverState *bs)
+{
+    return bdrv_co_flush(bs->file->bs);
+}
+
+static void throttle_detach_aio_context(BlockDriverState *bs)
+{
+    ThrottleGroupMember *tgm = bs->opaque;
+    throttle_group_detach_aio_context(tgm);
+}
+
+static void throttle_attach_aio_context(BlockDriverState *bs,
+                                    AioContext *new_context)

Indentation is off here...

+{
+    ThrottleGroupMember *tgm = bs->opaque;
+    throttle_group_attach_aio_context(tgm, new_context);
+}
+
+static int throttle_reopen_prepare(BDRVReopenState *reopen_state,
+                                      BlockReopenQueue *queue, Error **errp)

...and here.

+{
+    ThrottleGroupMember *tgm = NULL;
+
+    assert(reopen_state != NULL);
+    assert(reopen_state->bs != NULL);
+
+    reopen_state->opaque = g_new0(ThrottleGroupMember, 1);
+    tgm = reopen_state->opaque;
+
+    return throttle_configure_tgm(reopen_state->bs, tgm, reopen_state->options,
+            errp);
+}
+
+static void throttle_reopen_commit(BDRVReopenState *state)
+{
+    ThrottleGroupMember *tgm = state->bs->opaque;
+
+    throttle_group_unregister_tgm(tgm);
+    g_free(state->bs->opaque);
+    state->bs->opaque = state->opaque;
+    state->opaque = NULL;
+}
+
+static void throttle_reopen_abort(BDRVReopenState *state)
+{
+    ThrottleGroupMember *tgm = state->opaque;
+
+    throttle_group_unregister_tgm(tgm);
+    g_free(state->opaque);
+    state->opaque = NULL;
+}
+
+static bool throttle_recurse_is_first_non_filter(BlockDriverState *bs,
+                                                 BlockDriverState *candidate)
+{
+    return bdrv_recurse_is_first_non_filter(bs->file->bs, candidate);
+}
+
+static BlockDriver bdrv_throttle = {
+    .format_name                        =   "throttle",
+    .protocol_name                      =   "throttle",
+    .instance_size                      =   sizeof(ThrottleGroupMember),
+
+    .bdrv_file_open                     =   throttle_open,
+    .bdrv_close                         =   throttle_close,
+    .bdrv_co_flush                      =   throttle_co_flush,
+
+    .bdrv_child_perm                    =   bdrv_filter_default_perms,
+
+    .bdrv_getlength                     =   throttle_getlength,
+
+    .bdrv_co_preadv                     =   throttle_co_preadv,
+    .bdrv_co_pwritev                    =   throttle_co_pwritev,
+
+    .bdrv_co_pwrite_zeroes              =   throttle_co_pwrite_zeroes,
+    .bdrv_co_pdiscard                   =   throttle_co_pdiscard,
+
+    .bdrv_recurse_is_first_non_filter   =   
throttle_recurse_is_first_non_filter,
+
+    .bdrv_attach_aio_context            =   throttle_attach_aio_context,
+    .bdrv_detach_aio_context            =   throttle_detach_aio_context,
+
+    .bdrv_reopen_prepare                =   throttle_reopen_prepare,
+    .bdrv_reopen_commit                 =   throttle_reopen_commit,
+    .bdrv_reopen_abort                  =   throttle_reopen_abort,
+
+    .is_filter                          =   true,
+};

What about .bdrv_co_get_block_status?

We should use the default implementation here that you merged into your 2.11 tree, thanks for noticing and for all the comments.

+static void bdrv_throttle_init(void)
+{
+    bdrv_register(&bdrv_throttle);
+}
+
+block_init(bdrv_throttle_init);
diff --git a/include/qemu/throttle-options.h b/include/qemu/throttle-options.h
index 182b7896e1..3528a8f4a2 100644
--- a/include/qemu/throttle-options.h
+++ b/include/qemu/throttle-options.h
@@ -29,6 +29,7 @@
 #define QEMU_OPT_BPS_WRITE_MAX "bps-write-max"
 #define QEMU_OPT_BPS_WRITE_MAX_LENGTH "bps-write-max-length"
 #define QEMU_OPT_IOPS_SIZE "iops-size"
+#define QEMU_OPT_THROTTLE_GROUP_NAME "throttle-group"

 #define THROTTLE_OPT_PREFIX "throttling."
 #define THROTTLE_OPTS \

Kevin

Attachment: signature.asc
Description: PGP signature


reply via email to

[Prev in Thread] Current Thread [Next in Thread]