From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-wr1-f66.google.com ([209.85.221.66]:43864 "EHLO mail-wr1-f66.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1730492AbfHBNsX (ORCPT ); Fri, 2 Aug 2019 09:48:23 -0400 Received: by mail-wr1-f66.google.com with SMTP id p13so2700676wru.10 for ; Fri, 02 Aug 2019 06:48:20 -0700 (PDT) Date: Fri, 2 Aug 2019 15:48:17 +0200 From: Carlos Maiolino Subject: Re: [PATCH 8/9] Use FIEMAP for FIBMAP calls Message-ID: <20190802134816.usmauocewduggrjt@pegasus.maiolino.io> References: <20190731141245.7230-1-cmaiolino@redhat.com> <20190731141245.7230-9-cmaiolino@redhat.com> <20190731232254.GW1561054@magnolia> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20190731232254.GW1561054@magnolia> Sender: linux-xfs-owner@vger.kernel.org List-ID: List-Id: xfs To: "Darrick J. Wong" Cc: linux-fsdevel@vger.kernel.org, hch@lst.de, adilger@dilger.ca, jaegeuk@kernel.org, miklos@szeredi.hu, rpeterso@redhat.com, linux-xfs@vger.kernel.org > > -#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) > > +#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC | \ > > + FIEMAP_FLAG_XATTR| \ > > + FIEMAP_KERNEL_FIBMAP) > > > > static int ext4_xattr_fiemap(struct inode *inode, > > struct fiemap_extent_info *fieinfo) > > @@ -5048,6 +5050,9 @@ int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo) > > if (ext4_has_inline_data(inode)) { > > int has_inline = 1; > > > > + if (fieinfo->fi_flags & FIEMAP_KERNEL_FIBMAP) > > + return -EINVAL; > > Wouldn't the inline data case be caught by fiemap_bmap and turned into > -EINVAL? Yes, it does, but until ext4_fiemap() returns the extent with the INLINE flag, it does need to go through the whole fiemap mapping mechanism when we already know the result... So, instead of letting the ext4_fiemap() map the extent, just take the shortcut and return -EINVAL directly. The check in fiemap_bmap() is a 'safe measure' (if it does have other name I don't know :), but if the filesystem already knows it's gonna fall into an inline inode, taking the shortcut is better, isn't it? > > + return 1; > > + return 0; > > +} > > + > > +static int bmap_fiemap(struct inode *inode, sector_t *block) > > +{ > > + struct fiemap_extent_info fieinfo = { 0, }; > > + struct fiemap_extent fextent; > > + u64 start = *block << inode->i_blkbits; > > + int error = -EINVAL; > > + > > + fextent.fe_logical = 0; > > + fextent.fe_physical = 0; > > + fieinfo.fi_extents_max = 1; > > + fieinfo.fi_extents_mapped = 0; > > + fieinfo.fi_cb_data = &fextent; > > + fieinfo.fi_start = start; > > + fieinfo.fi_len = 1 << inode->i_blkbits; > > + fieinfo.fi_cb = fiemap_fill_kernel_extent; > > + fieinfo.fi_flags = (FIEMAP_KERNEL_FIBMAP | FIEMAP_FLAG_SYNC); > > + > > + error = inode->i_op->fiemap(inode, &fieinfo); > > + > > + if (error) > > + return error; > > + > > + if (fieinfo.fi_flags & (FIEMAP_EXTENT_UNKNOWN | > > + FIEMAP_EXTENT_ENCODED | > > + FIEMAP_EXTENT_DATA_INLINE | > > + FIEMAP_EXTENT_UNWRITTEN | > > + FIEMAP_EXTENT_SHARED)) > > + return -EINVAL; > > + > > + *block = (fextent.fe_physical + > > + (start - fextent.fe_logical)) >> inode->i_blkbits; > > + > > + return error; > > +} > > + > > /** > > * bmap - find a block number in a file > > * @inode: inode owning the block number being requested > > @@ -1591,10 +1663,15 @@ EXPORT_SYMBOL(iput); > > */ > > int bmap(struct inode *inode, sector_t *block) > > { > > - if (!inode->i_mapping->a_ops->bmap) > > + if (inode->i_op->fiemap) > > + return bmap_fiemap(inode, block); > > + > > + if (inode->i_mapping->a_ops->bmap) > > + *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, > > + *block); > > + else > > return -EINVAL; > > > > - *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block); > > return 0; > > } > > EXPORT_SYMBOL(bmap); > > diff --git a/fs/ioctl.c b/fs/ioctl.c > > index d72696c222de..0759ac6e4c7e 100644 > > --- a/fs/ioctl.c > > +++ b/fs/ioctl.c > > @@ -77,11 +77,8 @@ static int ioctl_fibmap(struct file *filp, int __user *p) > > return error; > > } > > > > -#define SET_UNKNOWN_FLAGS (FIEMAP_EXTENT_DELALLOC) > > -#define SET_NO_UNMOUNTED_IO_FLAGS (FIEMAP_EXTENT_DATA_ENCRYPTED) > > -#define SET_NOT_ALIGNED_FLAGS (FIEMAP_EXTENT_DATA_TAIL|FIEMAP_EXTENT_DATA_INLINE) > > -int fiemap_fill_user_extent(struct fiemap_extent_info *fieinfo, u64 logical, > > - u64 phys, u64 len, u32 flags) > > +static int fiemap_fill_user_extent(struct fiemap_extent_info *fieinfo, > > + u64 logical, u64 phys, u64 len, u32 flags) > > { > > struct fiemap_extent extent; > > struct fiemap_extent __user *dest = fieinfo->fi_cb_data; > > @@ -89,17 +86,17 @@ int fiemap_fill_user_extent(struct fiemap_extent_info *fieinfo, u64 logical, > > /* only count the extents */ > > if (fieinfo->fi_extents_max == 0) { > > fieinfo->fi_extents_mapped++; > > - return (flags & FIEMAP_EXTENT_LAST) ? 1 : 0; > > + goto out; > > } > > > > if (fieinfo->fi_extents_mapped >= fieinfo->fi_extents_max) > > return 1; > > > > - if (flags & SET_UNKNOWN_FLAGS) > > + if (flags & FIEMAP_EXTENT_DELALLOC) > > flags |= FIEMAP_EXTENT_UNKNOWN; > > - if (flags & SET_NO_UNMOUNTED_IO_FLAGS) > > + if (flags & FIEMAP_EXTENT_DATA_ENCRYPTED) > > flags |= FIEMAP_EXTENT_ENCODED; > > - if (flags & SET_NOT_ALIGNED_FLAGS) > > It's too bad that we lose the "not aligned" semantic meaning here. May you explain a bit better what you mean? We don't lose it, just the define goes away, the reason I dropped these defines is because the same flags are used in both functions, fiemap_fill_{user,kernel}_extent(), and I didn't think defining them on both places (or in fs.h) has any benefit here, so I opted to remove them. > > > + if (flags & (FIEMAP_EXTENT_DATA_TAIL | FIEMAP_EXTENT_DATA_INLINE)) > > flags |= FIEMAP_EXTENT_NOT_ALIGNED; > > Why doesn't this function just call fiemap_fill_kernel_extent to fill > out the onstack @extent structure? We've now implemented "fill out out > a struct fiemap_extent" twice. fiemap_fill_{user, kernel}_extent() have different purposes, and the big difference is one handles a userspace pointer memory and the other don't. IIRC the original proposal was some sort of sharing a single function, but then Christoph suggested a new design, using different functions as callbacks. > > diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c > > index b485190b7ecd..18a798e9076b 100644 > > --- a/fs/xfs/xfs_iops.c > > +++ b/fs/xfs/xfs_iops.c > > @@ -1113,6 +1113,11 @@ xfs_vn_fiemap( > > struct fiemap_extent_info *fieinfo) > > { > > int error; > > + struct xfs_inode *ip = XFS_I(inode); > > Would you mind fixing the indentation to match usual xfs style? Sure, will fix it > > > + > > + if (fieinfo->fi_flags & FIEMAP_KERNEL_FIBMAP) > > + if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip)) > > + return -EINVAL; > > The xfs part looks ok to me. > > --D > -- Carlos