Re: [Qemu-devel] [PATCH v2] 9pfs: add support for IO limits to 9p-local driver

[Pradeep Jagadeesh]

Hi Greg,

Thanks for looking into the patch. Please look at the replies inline.

Regards,
Pradeep

-----Original Message-----
From: Greg Kurz [mailto:address@hidden 
Sent: Friday, September 09, 2016 5:29 PM
To: Pradeep Jagadeesh
Cc: Aneesh Kumar K.V; Pradeep Jagadeesh; Alberto Garcia; address@hidden; 
Claudio Fontana; Eric Blake
Subject: Re: [PATCH v2] 9pfs: add support for IO limits to 9p-local driver

On Fri,  9 Sep 2016 05:10:27 -0400
Pradeep Jagadeesh <address@hidden> wrote:

> Uses throttling APIs to limit I/O bandwidth and number of operations 
> on the devices which use 9p-local driver.
> 
> Signed-off-by: Pradeep Jagadeesh <address@hidden>
> ---

Hi Pradeep,

Please find some remarks below. I haven't dived deep enough to see if this 
actually works. Maybe Berto can provide some feedback ?

Cheers.

--
Greg

>  fsdev/file-op-9p.h      |   3 +
>  fsdev/qemu-fsdev-opts.c |  52 +++++++++++++
>  hw/9pfs/9p-local.c      |  18 ++++-
>  hw/9pfs/9p-throttle.c   | 201 
> ++++++++++++++++++++++++++++++++++++++++++++++++
>  hw/9pfs/9p-throttle.h   |  46 +++++++++++
>  hw/9pfs/9p.c            |   7 ++
>  hw/9pfs/Makefile.objs   |   1 +
>  7 files changed, 326 insertions(+), 2 deletions(-)  create mode 
> 100644 hw/9pfs/9p-throttle.c  create mode 100644 hw/9pfs/9p-throttle.h
> 
> This adds the support for the 9p-local driver.
> For now this functionality can be enabled only through qemu cli options.
> QMP interface and support to other drivers need further extensions.
> To make it simple for other drivers, the throttle code has been put in 
> separate files.
> 
> v1 -> v2:
> 
> -Fixed FsContext redeclaration issue
> -Removed couple of function declarations from 9p-throttle.h -Fixed 
> some of the .help messages
> 
> diff --git a/fsdev/file-op-9p.h b/fsdev/file-op-9p.h index 
> 6db9fea..e86b91a 100644
> --- a/fsdev/file-op-9p.h
> +++ b/fsdev/file-op-9p.h
> @@ -17,6 +17,7 @@
>  #include <dirent.h>
>  #include <utime.h>
>  #include <sys/vfs.h>
> +#include "hw/9pfs/9p-throttle.h"
>  
>  #define SM_LOCAL_MODE_BITS    0600
>  #define SM_LOCAL_DIR_MODE_BITS    0700
> @@ -74,6 +75,7 @@ typedef struct FsDriverEntry {
>      char *path;
>      int export_flags;
>      FileOperations *ops;
> +    FsThrottle fst;
>  } FsDriverEntry;
>  
>  typedef struct FsContext
> @@ -83,6 +85,7 @@ typedef struct FsContext
>      int export_flags;
>      struct xattr_operations **xops;
>      struct extended_ops exops;
> +    FsThrottle *fst;
>      /* fs driver specific data */
>      void *private;
>  } FsContext;
> diff --git a/fsdev/qemu-fsdev-opts.c b/fsdev/qemu-fsdev-opts.c index 
> 1dd8c7a..2774855 100644
> --- a/fsdev/qemu-fsdev-opts.c
> +++ b/fsdev/qemu-fsdev-opts.c
> @@ -37,6 +37,58 @@ static QemuOptsList qemu_fsdev_opts = {
>          }, {
>              .name = "sock_fd",
>              .type = QEMU_OPT_NUMBER,
> +        }, {
> +            .name = "",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "limit total bytes per second",
> +        }, {
> +            .name = "bps_rd",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "limit read bytes per second",
> +        }, {
> +            .name = "bps_wr",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "limit write bytes per second",
> +        }, {
> +            .name = "iops",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "limit total io operations per second",
> +        }, {
> +            .name = "iops_rd",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "limit read operations per second ",
> +        }, {
> +            .name = "iops_wr",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "limit write operations per second",
> +        }, {
> +            .name = "bps_max",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "maximum bytes burst",
> +        }, {
> +            .name = "bps_rd_max",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "maximum bytes read burst",
> +        }, {
> +            .name = "bps_wr_max",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "maximum bytes write burst",
> +        }, {
> +            .name = "iops_max",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "maximum io operations burst",
> +        }, {
> +            .name = "iops_rd_max",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "maximum io operations read burst",
> +        }, {
> +            .name = "iops_wr_max",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "maximum io operations write burst",
> +        }, {
> +            .name = "iops_size",
> +            .type = QEMU_OPT_NUMBER,
> +            .help = "io iops-size",
>          },
>  

In case we end up sharing code with blockdev as suggested by Eric, maybe you 
can use the same QMP friendly naming scheme ("throttling.*") and the same help 
strings as well.

[Pradeep Jagadeesh] 
[Pradeep Jagadeesh] You mean to say I should re-name the options how they are 
done for block devices? Or mentioning
the cli options itself as throttling.*?
                                 
>          { /*End of list */ }
> diff --git a/hw/9pfs/9p-local.c b/hw/9pfs/9p-local.c index 
> 3f271fc..49c2819 100644
> --- a/hw/9pfs/9p-local.c
> +++ b/hw/9pfs/9p-local.c
> @@ -420,6 +420,8 @@ static ssize_t local_preadv(FsContext *ctx, 
> V9fsFidOpenState *fs,
>                              const struct iovec *iov,
>                              int iovcnt, off_t offset)  {
> +    throttle9p_request(ctx->fst, false, iov->iov_len);
> +
>  #ifdef CONFIG_PREADV
>      return preadv(fs->fd, iov, iovcnt, offset);  #else @@ -436,8 
> +438,10 @@ static ssize_t local_pwritev(FsContext *ctx, V9fsFidOpenState *fs,
>                               const struct iovec *iov,
>                               int iovcnt, off_t offset)  {
> -    ssize_t ret
> -;
> +    ssize_t ret;
> +
> +    throttle9p_request(ctx->fst, true, iov->iov_len);
> +
>  #ifdef CONFIG_PREADV
>      ret = pwritev(fs->fd, iov, iovcnt, offset);  #else @@ -1213,6 
> +1217,9 @@ static int local_parse_opts(QemuOpts *opts, struct FsDriverEntry 
> *fse)
>      const char *sec_model = qemu_opt_get(opts, "security_model");
>      const char *path = qemu_opt_get(opts, "path");
>  
> +    /* get the throttle structure */

Not sure this comment is helpful.
[Pradeep Jagadeesh] OK

> +    FsThrottle *fst = &fse->fst;
> +
>      if (!sec_model) {
>          error_report("Security model not specified, local fs needs security 
> model");
>          error_printf("valid options are:"
> @@ -1240,6 +1247,13 @@ static int local_parse_opts(QemuOpts *opts, struct 
> FsDriverEntry *fse)
>          error_report("fsdev: No path specified");
>          return -1;
>      }
> +
> +    throttle9p_enable_io_limits(opts, fst);
> +
> +    if (throttle9p_get_io_limits_state(fst)) {
> +        throttle9p_configure_iolimits(opts, fst);
> +    }
> +
>      fse->path = g_strdup(path);
>  
>      return 0;
> diff --git a/hw/9pfs/9p-throttle.c b/hw/9pfs/9p-throttle.c new file 
> mode 100644 index 0000000..f2a7ba5
> --- /dev/null
> +++ b/hw/9pfs/9p-throttle.c
> @@ -0,0 +1,201 @@
> +/*
> + * 9P Throttle
> + *
> + * Copyright (C) 2016 Huawei Technologies Duesseldorf GmbH
> + *
> + * Author: Pradeep Jagadeesh <address@hidden>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or
> + * (at your option) any later version.
> + *
> + * See the COPYING file in the top-level directory for details.
> + *
> + */
> +
> +#include "qemu/osdep.h"
> +#include "fsdev/qemu-fsdev.h"   /* local_ops */
> +#include "qemu/cutils.h"
> +#include "qemu/error-report.h"
> +#include <libgen.h>
> +#include <linux/fs.h>
> +#include <sys/ioctl.h>
> +#include <string.h>
> +#include "fsdev/file-op-9p.h"
> +#include "9p-throttle.h"
> +

Only "qemu/osdep.h", "qemu/error-report.h" and "9p-throttle.h" are actually 
needed.

> +void throttle9p_enable_io_limits(QemuOpts *opts, FsThrottle *fst) {
> +    const float bps = qemu_opt_get_number(opts, "bps", 0);
> +    const float iops = qemu_opt_get_number(opts, "iops", 0);
> +    const float rdbps = qemu_opt_get_number(opts, "bps_rd", 0);
> +    const float wrbps = qemu_opt_get_number(opts, "bps_wr", 0);
> +    const float rdops = qemu_opt_get_number(opts, "iops_rd", 0);
> +    const float wrops = qemu_opt_get_number(opts, "iops_wr", 0);
> +
> +    if (bps > 0 || iops > 0 || rdbps > 0 ||
> +        wrbps > 0 || rdops > 0 || wrops > 0) {
> +        fst->io_limits_enabled = true;
> +    } else {
> +        fst->io_limits_enabled = false;
> +    }
> +}
> +

This function should be named *_parse_* but I'm not even sure it is actually 
needed (see below).
[Pradeep Jagadeesh] This function is used to avoid the parsing of the options 
if not enabled.
>From my understanding of code block devices throttling code, irrespective of 
>the throttle options
are enabled or not the parse functions are called for all the devices. I am 
trying to avoid that 
by checking these variables at the before I get into parsing and setting up the 
throttle structures. 

> +static bool throttle9p_check_for_wait(FsThrottle *fst, bool is_write) 
> +{
> +    if (fst->any_timer_armed[is_write]) {
> +        return true;
> +    } else {
> +        return throttle_schedule_timer(&fst->ts, &fst->tt, is_write);
> +    }
> +}
> +
> +static void throttle9p_schedule_next_request(FsThrottle *fst, bool 
> +is_write) {
> +    bool must_wait = throttle9p_check_for_wait(fst, is_write);
> +    if (!fst->pending_reqs[is_write]) {
> +        return;
> +    }
> +    if (!must_wait) {
> +        if (qemu_in_coroutine() &&
> +            qemu_co_queue_next(&fst->throttled_reqs[is_write])) {
> +            ;
> +       } else {
> +           int64_t now = qemu_clock_get_ns(fst->tt.clock_type);
> +           timer_mod(fst->tt.timers[is_write], now + 1);
> +           fst->any_timer_armed[is_write] = true;
> +       }
> +   }
> +}
> +
> +static void throttle9p_timer_cb(FsThrottle *fst, bool is_write) {
> +    bool empty_queue;
> +    qemu_mutex_lock(&fst->lock);
> +    fst->any_timer_armed[is_write] = false;
> +    qemu_mutex_unlock(&fst->lock);
> +    empty_queue = !qemu_co_enter_next(&fst->throttled_reqs[is_write]);
> +    if (empty_queue) {
> +        qemu_mutex_lock(&fst->lock);
> +        throttle9p_schedule_next_request(fst, is_write);
> +        qemu_mutex_unlock(&fst->lock);
> +    }
> +}
> +
> +
> +bool throttle9p_get_io_limits_state(FsThrottle *fst)

The name looks a bit strange, since this helper simply returns a boolean flag.
I guess throttle9p_enabled() is enough.
[Pradeep Jagadeesh] OK

> +{
> +
> +    return fst->io_limits_enabled;
> +}
> +
> +static void throttle9p_read_timer_cb(void *opaque) {
> +    throttle9p_timer_cb(opaque, false); }
> +
> +static void throttle9p_write_timer_cb(void *opaque) {
> +    throttle9p_timer_cb(opaque, true); }
> +
> +void throttle9p_configure_iolimits(QemuOpts *opts, FsThrottle *fst) {

This function can fail, it should have a return value (0 or -1).
[Pradeep Jagadeesh] OK

> +    memset(&fst->ts, 1, sizeof(fst->ts));
> +    memset(&fst->tt, 1, sizeof(fst->tt));
> +    memset(&fst->cfg, 0, sizeof(fst->cfg));

Same remark as Claudio.
[Pradeep Jagadeesh] OK, will address.

> +    fst->aioctx = qemu_get_aio_context();
> +
> +    if (!fst->aioctx) {
> +        error_report("Failed to create AIO Context");
> +        exit(1);
> +    }
> +    throttle_init(&fst->ts);
> +    throttle_timers_init(&fst->tt,
> +                         fst->aioctx,
> +                         QEMU_CLOCK_REALTIME,
> +                         throttle9p_read_timer_cb,
> +                         throttle9p_write_timer_cb,
> +                         fst);

Shouldn't all this be done later when we know the config is valid ?
[Pradeep Jagadeesh] Yes, fixed.

> +    throttle_config_init(&fst->cfg);
> +    g_assert(throttle_is_valid(&fst->cfg, NULL));
> +

What's the point in calling throttle_is_valid() on a freshly initialized config 
?
[Pradeep Jagadeesh] Removed

> +    qemu_co_queue_init(&fst->throttled_reqs[0]);
> +    qemu_co_queue_init(&fst->throttled_reqs[1]);

Later, when the config is valid ?
[Pradeep Jagadeesh] Done

> +    fst->cfg.buckets[THROTTLE_BPS_TOTAL].avg =
> +          qemu_opt_get_number(opts, "bps", 0);
> +    fst->cfg.buckets[THROTTLE_BPS_READ].avg  =
> +          qemu_opt_get_number(opts, "bps_rd", 0);
> +    fst->cfg.buckets[THROTTLE_BPS_WRITE].avg =
> +          qemu_opt_get_number(opts, "bps_wr", 0);
> +    fst->cfg.buckets[THROTTLE_OPS_TOTAL].avg =
> +          qemu_opt_get_number(opts, "iops", 0);
> +    fst->cfg.buckets[THROTTLE_OPS_READ].avg =
> +          qemu_opt_get_number(opts, "iops_rd", 0);
> +    fst->cfg.buckets[THROTTLE_OPS_WRITE].avg =
> +          qemu_opt_get_number(opts, "iops_wr", 0);
> +
> +    fst->cfg.buckets[THROTTLE_BPS_TOTAL].max =
> +          qemu_opt_get_number(opts, "bps_max", 0);
> +    fst->cfg.buckets[THROTTLE_BPS_READ].max  =
> +          qemu_opt_get_number(opts, "bps_rd_max", 0);
> +    fst->cfg.buckets[THROTTLE_BPS_WRITE].max =
> +          qemu_opt_get_number(opts, "bps_wr_max", 0);
> +    fst->cfg.buckets[THROTTLE_OPS_TOTAL].max =
> +          qemu_opt_get_number(opts, "iops_max", 0);
> +    fst->cfg.buckets[THROTTLE_OPS_READ].max =
> +          qemu_opt_get_number(opts, "iops_rd_max", 0);
> +    fst->cfg.buckets[THROTTLE_OPS_WRITE].max =
> +          qemu_opt_get_number(opts, "iops_wr_max", 0);
> +
> +    fst->cfg.buckets[THROTTLE_BPS_TOTAL].burst_length =
> +          qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
> +    fst->cfg.buckets[THROTTLE_BPS_READ].burst_length  =
> +          qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
> +    fst->cfg.buckets[THROTTLE_BPS_WRITE].burst_length =
> +          qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
> +    fst->cfg.buckets[THROTTLE_OPS_TOTAL].burst_length =
> +          qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
> +    fst->cfg.buckets[THROTTLE_OPS_READ].burst_length =
> +          qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
> +    fst->cfg.buckets[THROTTLE_OPS_WRITE].burst_length =
> +          qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
> +    fst->cfg.op_size =
> +          qemu_opt_get_number(opts, "iops_size", 0);
> +
> +    throttle_config(&fst->ts, &fst->tt, &fst->cfg);

Let's set the config later, when we we it is valid.
[Pradeep Jagadeesh] OK

> +    if (!throttle_is_valid(&fst->cfg, NULL)) {

You should pass an Error * to throttle_is_valid() to be able to report the 
misconfiguration to the user. I guess it is okay to print it here using
error_repport_err() (see include/qapi/error.h) and return -1.
[Pradeep Jagadeesh] OK

> +        return;
> +    }
> +
> +    g_assert(fst->tt.timers[0]);
> +    g_assert(fst->tt.timers[1]);

These are really not needed since, timers are set with:

    QEMUTimer *ts = g_malloc0(sizeof(QEMUTimer));

and g_malloc0() never returns NULL when passed a non-nul size. It calls
g_assert() internally instead.
[Pradeep Jagadeesh] OK

> +    fst->pending_reqs[0] = 0;
> +    fst->pending_reqs[1] = 0;
> +    fst->any_timer_armed[0] = false;
> +    fst->any_timter_armed[1] = false;
> +    qemu_mutex_init(&fst->lock);

And there you may set the enabled flag.
[Pradeep Jagadeesh] Explained before.

> +}
> +
> +void throttle9p_request(FsThrottle *fst, bool is_write, ssize_t 
> +bytes) {
> +    if (fst->io_limits_enabled) {

throttle9p_enabled(fst)
[Pradeep Jagadeesh] OK

> +        qemu_mutex_lock(&fst->lock);
> +        bool must_wait = throttle9p_check_for_wait(fst, is_write);
> +        if (must_wait || fst->pending_reqs[is_write]) {
> +            fst->pending_reqs[is_write]++;
> +            qemu_mutex_unlock(&fst->lock);
> +            qemu_co_queue_wait(&fst->throttled_reqs[is_write]);
> +            qemu_mutex_lock(&fst->lock);
> +            fst->pending_reqs[is_write]--;
> +       }
> +       throttle_account(&fst->ts, is_write, bytes);
> +       throttle9p_schedule_next_request(fst, is_write);
> +       qemu_mutex_unlock(&fst->lock);
> +    }
> +}
> +
> +void throttle9p_cleanup(FsThrottle *fst) {
> +    throttle_timers_destroy(&fst->tt);
> +    qemu_mutex_destroy(&fst->lock);
> +}
> diff --git a/hw/9pfs/9p-throttle.h b/hw/9pfs/9p-throttle.h new file 
> mode 100644 index 0000000..0f7551d
> --- /dev/null
> +++ b/hw/9pfs/9p-throttle.h
> @@ -0,0 +1,46 @@
> +/*
> + * 9P Throttle
> + *
> + * Copyright (C) 2016 Huawei Technologies Duesseldorf GmbH
> + *
> + * Author: Pradeep Jagadeesh <address@hidden>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or
> + * (at your option) any later version.
> + *
> + * See the COPYING file in the top-level directory for details.
> + *
> + */
> +
> +#ifndef _9P_THROTTLE_H
> +#define _9P_THROTTLE_H
> +
> +#include <stdbool.h>
> +#include <stdint.h>

These includes are forbidden. They are already handled by "qemu/osdep.h" which 
is supposed to be included by all .c files.
[Pradeep Jagadeesh] OK

> +#include "block/aio.h"
> +#include "qemu/main-loop.h"
> +#include "qemu/coroutine.h"
> +#include "qemu/throttle.h"
> +
> +typedef struct FsThrottle {
> +    ThrottleState ts;
> +    ThrottleTimers tt;
> +    AioContext   *aioctx;
> +    ThrottleConfig cfg;
> +    bool io_limits_enabled;

Let's simply call this enabled.
[Pradeep Jagadeesh] OK

> +    CoQueue      throttled_reqs[2];
> +    unsigned     pending_reqs[2];
> +    bool any_timer_armed[2];
> +    QemuMutex lock;
> +} FsThrottle;
> +
> +void throttle9p_enable_io_limits(QemuOpts *, FsThrottle *);
> +
> +bool throttle9p_get_io_limits_state(FsThrottle *);
> +
> +void throttle9p_configure_iolimits(QemuOpts *, FsThrottle *);
> +
> +void throttle9p_request(FsThrottle *, bool , ssize_t);
> +
> +void throttle9p_cleanup(FsThrottle *); #endif /* _9P_THROTTLE_H */
> diff --git a/hw/9pfs/9p.c b/hw/9pfs/9p.c index dfe293d..7be926a 100644
> --- a/hw/9pfs/9p.c
> +++ b/hw/9pfs/9p.c
> @@ -3490,6 +3490,10 @@ int v9fs_device_realize_common(V9fsState *s, Error 
> **errp)
>          error_setg(errp, "share path %s is not a directory", fse->path);
>          goto out;
>      }
> +
> +    /* Throttle structure initialization */

Not sure this comment is helpful.
[Pradeep Jagadeesh] Just to give a hint, where the throttle structures are 
getting
Populated. May be some other comment?

> +    s->ctx.fst = &fse->fst;
> +
>      v9fs_path_free(&path);
>  
>      rc = 0;
> @@ -3504,6 +3508,9 @@ out:
>  
>  void v9fs_device_unrealize_common(V9fsState *s, Error **errp)  {
> +    if (throttle9p_get_io_limits_state(s->ctx.fst)) {
> +        throttle9p_cleanup(s->ctx.fst);
> +    }
>      g_free(s->ctx.fs_root);
>      g_free(s->tag);
>  }
> diff --git a/hw/9pfs/Makefile.objs b/hw/9pfs/Makefile.objs index 
> da0ae0c..07523f1 100644
> --- a/hw/9pfs/Makefile.objs
> +++ b/hw/9pfs/Makefile.objs
> @@ -5,5 +5,6 @@ common-obj-y += coth.o cofs.o codir.o cofile.o  
> common-obj-y += coxattr.o 9p-synth.o
>  common-obj-$(CONFIG_OPEN_BY_HANDLE) +=  9p-handle.o  common-obj-y += 
> 9p-proxy.o
> +common-obj-y += 9p-throttle.o
>  
>  obj-y += virtio-9p-device.o