Re: [RFC] failure atomic writes for file systems and block devices
On Tue, Feb 28, 2017 at 03:22:04PM -0800, Darrick J. Wong wrote: > (Assuming there's no syncv involved here...?) No. While I think we could implement it for XFS similar how we roll transactions over multiple inodes for a few transactions, the use case is much more limited, and the potential pitfalls are much bigger. > > have to check the F_IOINFO fcntl before, which is a bit of a killer. > > Because of that I've also not implemented any other validity checks > > yet, as they might make thing even worse when an open on a not supported > > file system or device fails, but not on an old kernel. Maybe we need > > a new open version that checks arguments properly first? > > Does fcntl(F_SETFL...) suffer from this? Yes.
Re: [RFC] failure atomic writes for file systems and block devices
On Tue, Feb 28, 2017 at 03:48:16PM -0500, Chris Mason wrote: > One thing that isn't clear to me is how we're dealing with boundary bio > mappings, which will get submitted by submit_page_section() > > sdio->boundary = buffer_boundary(map_bh); The old dio code is not supported at all by this code at the moment. We'll either need the new block dev direct I/O code on block devices (and limit to BIO_MAX_PAGES, this is a bug in this patchset if people ever have devices with > 1MB atomic write support. And thanks to NVMe the failure case is silent, sigh..), or we need file system support for out of place writes. > > In btrfs I'd just chain things together and do the extent pointer swap > afterwards, but I didn't follow the XFS code well enough to see how its > handled there. But either way it feels like an error prone surprise > waiting for later, and one gap we really want to get right in the FS > support is O_ATOMIC across a fragmented extent. > > If I'm reading the XFS patches right, the code always cows for atomic. It doesn't really COW - it uses the COW infrastructure to write out of place and then commit it into the file later. Because of that we don't really care about things like boundary blocks (which XFS never used in that form anyway) - data is written first, the cache is flushed and then we swap around the extent pointers. > Are > you planning on adding an optimization to use atomic support in the device > to skip COW when possible? We could do that fairly easily for files that have a contiguous mapping for the atomic write I/O. But at this point I have a lot more trust in the fs code than the devices, especially due to the silent failure mode.
Re: [RFC] failure atomic writes for file systems and block devices
On Wed, Mar 01, 2017 at 01:21:41PM +0200, Amir Goldstein wrote: > [CC += linux-...@vger.kernel.org] for that question and for the new API We'll need to iterate over the API a few more times first I think..
Re: [RFC] failure atomic writes for file systems and block devices
On Tue, Feb 28, 2017 at 4:57 PM, Christoph Hellwig wrote: > Hi all, > > this series implements a new O_ATOMIC flag for failure atomic writes > to files. It is based on and tries to unify to earlier proposals, > the first one for block devices by Chris Mason: > > https://lwn.net/Articles/573092/ > > and the second one for regular files, published by HP Research at > Usenix FAST 2015: > > > https://www.usenix.org/conference/fast15/technical-sessions/presentation/verma > > It adds a new O_ATOMIC flag for open, which requests writes to be > failure-atomic, that is either the whole write makes it to persistent > storage, or none of it, even in case of power of other failures. > > There are two implementation various of this: on block devices O_ATOMIC > must be combined with O_(D)SYNC so that storage devices that can handle > large writes atomically can simply do that without any additional work. > This case is supported by NVMe. > > The second case is for file systems, where we simply write new blocks > out of places and then remap them into the file atomically on either > completion of an O_(D)SYNC write or when fsync is called explicitly. > > The semantics of the latter case are explained in detail at the Usenix > paper above. > > Last but not least a new fcntl is implemented to provide information > about I/O restrictions such as alignment requirements and the maximum > atomic write size. > > The implementation is simple and clean, but I'm rather unhappy about > the interface as it has too many failure modes to bullet proof. For > one old kernels ignore unknown open flags silently, so applications > have to check the F_IOINFO fcntl before, which is a bit of a killer. > Because of that I've also not implemented any other validity checks > yet, as they might make thing even worse when an open on a not supported > file system or device fails, but not on an old kernel. Maybe we need > a new open version that checks arguments properly first? > [CC += linux-...@vger.kernel.org] for that question and for the new API > Also I'm really worried about the NVMe failure modes - devices simply > advertise an atomic write size, with no way for the device to know > that the host requested a given write to be atomic, and thus no > error reporting. This is made worse by NVMe 1.2 adding per-namespace > atomic I/O parameters that devices can use to introduce additional > odd alignment quirks - while there is some language in the spec > requiring them not to weaken the per-controller guarantees it all > looks rather weak and I'm not too confident in all implementations > getting everything right. > > Last but not least this depends on a few XFS patches, so to actually > apply / run the patches please use this git tree: > > git://git.infradead.org/users/hch/vfs.git O_ATOMIC > > Gitweb: > > http://git.infradead.org/users/hch/vfs.git/shortlog/refs/heads/O_ATOMIC
Re: [RFC] failure atomic writes for file systems and block devices
On 02/28/2017 09:57 AM, Christoph Hellwig wrote: Hi all, this series implements a new O_ATOMIC flag for failure atomic writes to files. It is based on and tries to unify to earlier proposals, the first one for block devices by Chris Mason: https://urldefense.proofpoint.com/v2/url?u=https-3A__lwn.net_Articles_573092_&d=DwIBAg&c=5VD0RTtNlTh3ycd41b3MUw&r=9QPtTAxcitoznaWRKKHoEQ&m=P5byIhbDCF-kdlNpZVpxMKG3E36-cQ-lK27coqUFUng&s=rqXtuRMvf2rijHel_VAiO-KQ8AtQ5DXEI2obnCI_ljQ&e= and the second one for regular files, published by HP Research at Usenix FAST 2015: https://urldefense.proofpoint.com/v2/url?u=https-3A__www.usenix.org_conference_fast15_technical-2Dsessions_presentation_verma&d=DwIBAg&c=5VD0RTtNlTh3ycd41b3MUw&r=9QPtTAxcitoznaWRKKHoEQ&m=P5byIhbDCF-kdlNpZVpxMKG3E36-cQ-lK27coqUFUng&s=ilnrrNs8nG4_UV2xx7tc2Efm20d2Wa8PHoJE8WUTCwI&e= It adds a new O_ATOMIC flag for open, which requests writes to be failure-atomic, that is either the whole write makes it to persistent storage, or none of it, even in case of power of other failures. There are two implementation various of this: on block devices O_ATOMIC must be combined with O_(D)SYNC so that storage devices that can handle large writes atomically can simply do that without any additional work. This case is supported by NVMe. Hi Christoph, This is great, and supporting code in both dio and bio get rid of some of the warts from when I tried. The DIO_PAGES define used to be an upper limit on the max contiguous bio that would get built, but that's much better now. One thing that isn't clear to me is how we're dealing with boundary bio mappings, which will get submitted by submit_page_section() sdio->boundary = buffer_boundary(map_bh); In btrfs I'd just chain things together and do the extent pointer swap afterwards, but I didn't follow the XFS code well enough to see how its handled there. But either way it feels like an error prone surprise waiting for later, and one gap we really want to get right in the FS support is O_ATOMIC across a fragmented extent. If I'm reading the XFS patches right, the code always cows for atomic. Are you planning on adding an optimization to use atomic support in the device to skip COW when possible? To turn off mysql double buffering, we only need 16K or 64K writes, which most of the time you'd be able to pass down directly without cows. -chris
Re: [RFC] failure atomic writes for file systems and block devices
On Tue, Feb 28, 2017 at 06:57:25AM -0800, Christoph Hellwig wrote: > Hi all, > > this series implements a new O_ATOMIC flag for failure atomic writes > to files. It is based on and tries to unify to earlier proposals, > the first one for block devices by Chris Mason: > > https://lwn.net/Articles/573092/ > > and the second one for regular files, published by HP Research at > Usenix FAST 2015: > > > https://www.usenix.org/conference/fast15/technical-sessions/presentation/verma > > It adds a new O_ATOMIC flag for open, which requests writes to be > failure-atomic, that is either the whole write makes it to persistent > storage, or none of it, even in case of power of other failures. > > There are two implementation various of this: on block devices O_ATOMIC > must be combined with O_(D)SYNC so that storage devices that can handle > large writes atomically can simply do that without any additional work. > This case is supported by NVMe. > > The second case is for file systems, where we simply write new blocks > out of places and then remap them into the file atomically on either > completion of an O_(D)SYNC write or when fsync is called explicitly. > > The semantics of the latter case are explained in detail at the Usenix > paper above. (Assuming there's no syncv involved here...?) > Last but not least a new fcntl is implemented to provide information > about I/O restrictions such as alignment requirements and the maximum > atomic write size. > > The implementation is simple and clean, but I'm rather unhappy about > the interface as it has too many failure modes to bullet proof. For > one old kernels ignore unknown open flags silently, so applications Ok, heh, disregard my review comment (for the xfs part) about the seemingly insufficient O_ATOMIC validation. > have to check the F_IOINFO fcntl before, which is a bit of a killer. > Because of that I've also not implemented any other validity checks > yet, as they might make thing even worse when an open on a not supported > file system or device fails, but not on an old kernel. Maybe we need > a new open version that checks arguments properly first? Does fcntl(F_SETFL...) suffer from this? > Also I'm really worried about the NVMe failure modes - devices simply > advertise an atomic write size, with no way for the device to know > that the host requested a given write to be atomic, and thus no > error reporting. Yikes! > This is made worse by NVMe 1.2 adding per-namespace > atomic I/O parameters that devices can use to introduce additional > odd alignment quirks - while there is some language in the spec > requiring them not to weaken the per-controller guarantees it all > looks rather weak and I'm not too confident in all implementations > getting everything right. > > Last but not least this depends on a few XFS patches, so to actually > apply / run the patches please use this git tree: Well, the XFS parts don't look too bad --D > > git://git.infradead.org/users/hch/vfs.git O_ATOMIC > > Gitweb: > > http://git.infradead.org/users/hch/vfs.git/shortlog/refs/heads/O_ATOMIC