New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Async IO #223

Closed
behlendorf opened this Issue May 3, 2011 · 8 comments

Comments

Projects
None yet
7 participants
@behlendorf
Member

behlendorf commented May 3, 2011

The asynchronous IO handlers have not yet been implemented. By default ZFS handles everything asynchronously so this shouldn't be that hard but it will take some care to implement the handlers. Once they are in place the synchronous handlers can removed and replaced with the generic do_sync_read() and do_sync_write() handlers. They simply call the async hooks and then block resulting in a synchronous version.

@dechamps

This comment has been minimized.

Show comment
Hide comment
@dechamps

dechamps Jul 1, 2012

Contributor

By default ZFS handles everything asynchronously so this shouldn't be that hard

Well, not quite. The ZIO pipeline is indeed asynchronous, but there are cases when ZFS does blocking operations before getting to the ZIO pipeline:

  • Unaligned rewrites, which translates to a read-modify-write (see #361). The reading part is done synchronously as part of zfs_write();
  • Synchronous writes, which translates to a call to zfs_commit() inside zfs_write(). zfs_commit() synchronously blocks until the ZIL has been flushed.

Making these two operations asynchronous would require a significant amount of work. Of course, one could delegate all these blocking operations to worker threads, but that would defeat the purpose of async I/O in the first place.

Contributor

dechamps commented Jul 1, 2012

By default ZFS handles everything asynchronously so this shouldn't be that hard

Well, not quite. The ZIO pipeline is indeed asynchronous, but there are cases when ZFS does blocking operations before getting to the ZIO pipeline:

  • Unaligned rewrites, which translates to a read-modify-write (see #361). The reading part is done synchronously as part of zfs_write();
  • Synchronous writes, which translates to a call to zfs_commit() inside zfs_write(). zfs_commit() synchronously blocks until the ZIL has been flushed.

Making these two operations asynchronous would require a significant amount of work. Of course, one could delegate all these blocking operations to worker threads, but that would defeat the purpose of async I/O in the first place.

@cwedgwood

This comment has been minimized.

Show comment
Hide comment
@cwedgwood

cwedgwood Sep 1, 2012

Contributor

A reference implementation from another fs (btrfs):

http://lwn.net/Articles/373866/
Contributor

cwedgwood commented Sep 1, 2012

A reference implementation from another fs (btrfs):

http://lwn.net/Articles/373866/

@behlendorf behlendorf referenced this issue Oct 29, 2012

Closed

Direct IO #224

@Firstyear

This comment has been minimized.

Show comment
Hide comment
@Firstyear

Firstyear May 26, 2013

This would be good to have as certain databases IE mariadb use AIO by default on some distro's.

Firstyear commented May 26, 2013

This would be good to have as certain databases IE mariadb use AIO by default on some distro's.

@ryao

This comment has been minimized.

Show comment
Hide comment
@ryao

ryao Nov 11, 2013

Member

Fruit in #zfsonlinux on freenode reports that writev() is a factor of 10 slower for him on ZFS than on other filesystems. The code responsible for writev() is in fs/read_write.c in Linux's tree. Specifically, do_readv_writev(). A cursory examination revealed that it invokes f_op->write() in a loop when a_op->aio_write() is not available. This means that each writev() is translated into a series of non-aligned write() calls that each block on a read-modify-write and are processed sequentially.

Member

ryao commented Nov 11, 2013

Fruit in #zfsonlinux on freenode reports that writev() is a factor of 10 slower for him on ZFS than on other filesystems. The code responsible for writev() is in fs/read_write.c in Linux's tree. Specifically, do_readv_writev(). A cursory examination revealed that it invokes f_op->write() in a loop when a_op->aio_write() is not available. This means that each writev() is translated into a series of non-aligned write() calls that each block on a read-modify-write and are processed sequentially.

@behlendorf

This comment has been minimized.

Show comment
Hide comment
@behlendorf

behlendorf Nov 13, 2013

Member

That sounds about right. If we're able to cleanly pull in the asynchronous dmu work the FreeBSD guys did we should be able to resolve this cleanly. It will likely also depend on the scatter-gather improvements and generally doing things in a more Linuxy way.

http://lists.freebsd.org/pipermail/zfs-devel/2013-September/000388.html

Member

behlendorf commented Nov 13, 2013

That sounds about right. If we're able to cleanly pull in the asynchronous dmu work the FreeBSD guys did we should be able to resolve this cleanly. It will likely also depend on the scatter-gather improvements and generally doing things in a more Linuxy way.

http://lists.freebsd.org/pipermail/zfs-devel/2013-September/000388.html

@bassu

This comment has been minimized.

Show comment
Hide comment
@bassu

bassu May 5, 2014

Contributor

How long will it take to implement this? For some workloads, I believe not having enough asynchronous IO is the root cause of all the problems.

Contributor

bassu commented May 5, 2014

How long will it take to implement this? For some workloads, I believe not having enough asynchronous IO is the root cause of all the problems.

@DeHackEd

This comment has been minimized.

Show comment
Hide comment
@DeHackEd

DeHackEd May 5, 2014

Contributor

I'm pretty sure this refers to the aio(7) programmer's interface which doesn't see that much usage outside certain explicitly enabled situations. eg: qmeu/KVM supports it for file IO. I doubt this would cause the sorts of breakage you're suggesting.

Contributor

DeHackEd commented May 5, 2014

I'm pretty sure this refers to the aio(7) programmer's interface which doesn't see that much usage outside certain explicitly enabled situations. eg: qmeu/KVM supports it for file IO. I doubt this would cause the sorts of breakage you're suggesting.

@ryao

This comment has been minimized.

Show comment
Hide comment
@ryao

ryao May 5, 2014

Member

@bassu It could be implemented in a couple of days if one of us sat down to do it. The only problems are that it is not a priority and while aio_sync exists, not much software can be expected to use it correctly because Linux simply does not implement it andinstead relies on hacks to make AIO seem safe.

Member

ryao commented May 5, 2014

@bassu It could be implemented in a couple of days if one of us sat down to do it. The only problems are that it is not a priority and while aio_sync exists, not much software can be expected to use it correctly because Linux simply does not implement it andinstead relies on hacks to make AIO seem safe.

ryao added a commit to ryao/zfs that referenced this issue Aug 4, 2014

Implement AIO hooks
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to implement fops->aio_write synchronous to make
software that expects this behavior safe. However, there are several
reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not respect O_SYNC on files and assumes synchronous
behavior from do_readv_writev(), even though its fallback clearly does
not enforce it.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure. Exporting any file
system that implements AIO the way this patch does bears similar risk.
However, it seems reasonable to forgo crippling our AIO implementation
in favor of developing patches to fix this problem in Linux's nfsd for
the reasons stated earlier. In the interim, the risk will remain.
Failing to implement AIO will not change the problem that nfsd created,
so there is no reason for nfsd's mistake to block our implementation of
AIO.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Aug 4, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to implement fops->aio_write synchronous to make
software that expects this behavior safe. However, there are several
reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not respect O_SYNC on files and assumes synchronous
behavior from do_readv_writev(), even though its fallback clearly does
not enforce it.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure. Exporting any file
system that implements AIO the way this patch does bears similar risk.
However, it seems reasonable to forgo crippling our AIO implementation
in favor of developing patches to fix this problem in Linux's nfsd for
the reasons stated earlier. In the interim, the risk will remain.
Failing to implement AIO will not change the problem that nfsd created,
so there is no reason for nfsd's mistake to block our implementation of
AIO.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

@ryao ryao referenced this issue Aug 4, 2014

Closed

Linux AIO Support #2565

ryao added a commit to ryao/zfs that referenced this issue Aug 7, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

DeHackEd added a commit to DeHackEd/zfs that referenced this issue Aug 9, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Aug 10, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Aug 11, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Aug 12, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Aug 12, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Sep 2, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Sep 3, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Sep 3, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Sep 3, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Lastly, it is important to known that handling of the iovec updates differs
between Illumos and Linux in the implementation of read/write. On Linux, it is
the VFS' responsibility whle on Illumos, it is the filesystem's reponsibility.
We take the intermediate solution of copying the iovec so that the ZFS code can
update it like on Solaris while leaving the originals alone. This imposes some
overhead. We could always revisit this should profiling show that the
allocations are a problem.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

ryao added a commit to ryao/zfs that referenced this issue Sep 3, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Lastly, it is important to known that handling of the iovec updates differs
between Illumos and Linux in the implementation of read/write. On Linux, it is
the VFS' responsibility whle on Illumos, it is the filesystem's reponsibility.
We take the intermediate solution of copying the iovec so that the ZFS code can
update it like on Solaris while leaving the originals alone. This imposes some
overhead. We could always revisit this should profiling show that the
allocations are a problem.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

@behlendorf behlendorf closed this in cd3939c Sep 5, 2014

DeHackEd pushed a commit to DeHackEd/zfs that referenced this issue Sep 18, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Lastly, it is important to know that handling of the iovec updates differs
between Illumos and Linux in the implementation of read/write. On Linux,
it is the VFS' responsibility whle on Illumos, it is the filesystem's
responsibility.  We take the intermediate solution of copying the iovec
so that the ZFS code can update it like on Solaris while leaving the
originals alone. This imposes some overhead. We could always revisit
this should profiling show that the allocations are a problem.

Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #223
Closes #2373

ryao added a commit to ryao/zfs that referenced this issue Oct 8, 2014

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to make our fops->aio_write implementation
synchronous to make software that expects this behavior safe. However,
there are several reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not pass O_SYNC on files opened with it and instead
relies on a open()/write()/close() to enforce synchronous behavior when
the flush is only guarenteed on last close.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure when something else
is also accessing the file. Exporting any file system that implements
AIO the way this patch does bears similar risk. However, it seems
reasonable to forgo crippling our AIO implementation in favor of
developing patches to fix this problem in Linux's nfsd for the reasons
stated earlier. In the interim, the risk will remain. Failing to
implement AIO will not change the problem that nfsd created, so there is
no reason for nfsd's mistake to block our implementation of AIO.

It also should be noted that `aio_cancel()` will always return
`AIO_NOTCANCELED` under this implementation. It is possible to implement
aio_cancel by deferring work to taskqs and use `kiocb_set_cancel_fn()`
to set a callback function for cancelling work sent to taskqs, but the
simpler approach is allowed by the specification:

```
Which operations are cancelable is implementation-defined.
```

http://pubs.opengroup.org/onlinepubs/009695399/functions/aio_cancel.html

The only programs on my system that are capable of using `aio_cancel()`
are QEMU, beecrypt and fio use it according to a recursive grep of my
system's `/usr/src/debug`. That suggests that `aio_cancel()` users are
rare. Implementing aio_cancel() is left to a future date when it is
clear that there are consumers that benefit from its implementation to
justify the work.

Lastly, it is important to known that handling of the iovec updates differs
between Illumos and Linux in the implementation of read/write. On Linux, it is
the VFS' responsibility whle on Illumos, it is the filesystem's reponsibility.
We take the intermediate solution of copying the iovec so that the ZFS code can
update it like on Solaris while leaving the originals alone. This imposes some
overhead. We could always revisit this should profiling show that the
allocations are a problem.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>

behlendorf added a commit to behlendorf/zfs that referenced this issue Jun 9, 2015

Linux AIO Support
nfsd uses do_readv_writev() to implement fops->read and fops->write.
do_readv_writev() will attempt to read/write using fops->aio_read and
fops->aio_write, but it will fallback to fops->read and fops->write when
AIO is not available. However, the fallback will perform a call for each
individual data page. Since our default recordsize is 128KB, sequential
operations on NFS will generate 32 DMU transactions where only 1
transaction was needed. That was unnecessary overhead and we implement
fops->aio_read and fops->aio_write to eliminate it.

ZFS originated in OpenSolaris, where the AIO API is entirely implemented
in userland's libc by intelligently mapping them to VOP_WRITE, VOP_READ
and VOP_FSYNC.  Linux implements AIO inside the kernel itself. Linux
filesystems therefore must implement their own AIO logic and nearly all
of them implement fops->aio_write synchronously. Consequently, they do
not implement aio_fsync(). However, since the ZPL works by mapping
Linux's VFS calls to the functions implementing Illumos' VFS operations,
we instead implement AIO in the kernel by mapping the operations to the
VOP_READ, VOP_WRITE and VOP_FSYNC equivalents. We therefore implement
fops->aio_fsync.

One might be inclined to implement fops->aio_write synchronous to make
software that expects this behavior safe. However, there are several
reasons not to do this:

1. Other platforms do not implement aio_write() synchronously and since
the majority of userland software using AIO should be cross platform,
expectations of synchronous behavior should not be a problem.

2. We would hurt the performance of programs that use POSIX interfaces
properly while simultaneously encouraging the creation of more
non-compliant software.

3. The broader community concluded that userland software should be
patched to properly use POSIX interfaces instead of implementing hacks
in filesystems to cater to broken software. This concept is best
described as the O_PONIES debate.

4. Making an asynchronous write synchronous is non sequitur.

Any software dependent on synchronous aio_write behavior will suffer
data loss on ZFSOnLinux in a kernel panic / system failure of at most
zfs_txg_timeout seconds, which by default is 5 seconds. This seems like
a reasonable consequence of using non-compliant software.

It should be noted that this is also a problem in the kernel itself
where nfsd does not respect O_SYNC on files and assumes synchronous
behavior from do_readv_writev(), even though its fallback clearly does
not enforce it.

Exporting any filesystem that does not implement AIO via NFS risks data
loss in the event of a kernel panic / system failure. Exporting any file
system that implements AIO the way this patch does bears similar risk.
However, it seems reasonable to forgo crippling our AIO implementation
in favor of developing patches to fix this problem in Linux's nfsd for
the reasons stated earlier. In the interim, the risk will remain.
Failing to implement AIO will not change the problem that nfsd created,
so there is no reason for nfsd's mistake to block our implementation of
AIO.

Closes:
zfsonlinux/zfs#223
zfsonlinux/zfs#2373

Signed-off-by: Richard Yao <ryao@gentoo.org>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment