| Message ID | 20090916162457.GA84213@freezingfog.local |
|---|---|
| State | New, archived |
| Headers | show |
On Sep 16, 2009 11:24 -0500, Will Drewry wrote: > I have a two questions with an accompanying patch for clarification. > > resize2fs is uninit_bg aware, but when it is expanding an ext4 > filesystem, it will always zero the inode tables. Is it safe to mimick > mke2fs's write_inode_table(.., lazy_flag=1) and leave the new block > groups' inode tables marked INODE_UNINIT, BLOCK_UNINIT and _not_ zero > out the inode table if uninit_bg is supported? > > If it is okay, then it means offline resizing upwards can be just as > fast as mke2fs. I've attached a patch which is probably incomplete. > I'd love feedback as to the feasibility of the change and/or patch > quality. > > As a follow-on, would it be sane to add support like this for > online resizing. From a cursory investigation, it looks like > setup_new_block_groups() could be modified to not zero itables > if uninit_bg is supported, and INODE_ZEROED could be replaced > with ΒG_*_UNINIT. However, I'm not sure if that is a naive > view. I'm happy to send along a patch illustrating this change > if that'd be helpful or welcome. The question is why you would want to risk disk corruption vs. the (likely not performance critical) online resize? > Any and all feedback is appreciated -- even if it just for me > to look at the archives/link/etc. > > diff --git a/resize/resize2fs.c b/resize/resize2fs.c > index 1a5d910..9fcc3b9 100644 > --- a/resize/resize2fs.c > +++ b/resize/resize2fs.c > @@ -497,8 +497,7 @@ retry: > > fs->group_desc[i].bg_flags = 0; > if (csum_flag) > - fs->group_desc[i].bg_flags |= EXT2_BG_INODE_UNINIT | > - EXT2_BG_INODE_ZEROED; > + fs->group_desc[i].bg_flags |= EXT2_BG_INODE_UNINIT; This shouldn't be unconditional, since most users will want the safety of having zeroed inode tables. Cheers, Andreas -- Andreas Dilger Sr. Staff Engineer, Lustre Group Sun Microsystems of Canada, Inc. -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/resize/resize2fs.c b/resize/resize2fs.c index 1a5d910..9fcc3b9 100644 --- a/resize/resize2fs.c +++ b/resize/resize2fs.c @@ -497,8 +497,7 @@ retry: fs->group_desc[i].bg_flags = 0; if (csum_flag) - fs->group_desc[i].bg_flags |= EXT2_BG_INODE_UNINIT | - EXT2_BG_INODE_ZEROED; + fs->group_desc[i].bg_flags |= EXT2_BG_INODE_UNINIT; if (i == fs->group_desc_count-1) { numblocks = (fs->super->s_blocks_count - fs->super->s_first_data_block) % @@ -568,7 +567,7 @@ errout: static errcode_t adjust_superblock(ext2_resize_t rfs, blk_t new_size) { ext2_filsys fs; - int adj = 0; + int adj = 0, csum_flag = 0, num = 0; errcode_t retval; blk_t group_block; unsigned long i; @@ -624,6 +623,9 @@ static errcode_t adjust_superblock(ext2_resize_t rfs, blk_t new_size) &rfs->itable_buf); if (retval) goto errout; + /* Track if we can get by with a lazy init */ + csum_flag = EXT2_HAS_RO_COMPAT_FEATURE(fs->super, + EXT4_FEATURE_RO_COMPAT_GDT_CSUM); memset(rfs->itable_buf, 0, fs->blocksize * fs->inode_blocks_per_group); group_block = fs->super->s_first_data_block + @@ -642,10 +644,24 @@ static errcode_t adjust_superblock(ext2_resize_t rfs, blk_t new_size) /* * Write out the new inode table */ + if (csum_flag) { + /* These are _new_ inode tables. No inodes should be in use. + * (As per ext2fs_set_gdt_csum) */ + fs->group_desc[i].bg_itable_unused = fs->super->s_inodes_per_group; + num = ((((fs->super->s_inodes_per_group - + fs->group_desc[i].bg_itable_unused) * + EXT2_INODE_SIZE(fs->super)) + + EXT2_BLOCK_SIZE(fs->super) - 1) / + EXT2_BLOCK_SIZE(fs->super)); + } else { + num = fs->inode_blocks_per_group; + /* The kernel doesn't need to zero the itable blocks. We will below */ + fs->group_desc[i].bg_flags |= EXT2_BG_INODE_ZEROED; + } retval = io_channel_write_blk(fs->io, - fs->group_desc[i].bg_inode_table, - fs->inode_blocks_per_group, - rfs->itable_buf); + fs->group_desc[i].bg_inode_table, /* blk */ + num, /* count */ + rfs->itable_buf); /* contents */ if (retval) goto errout; io_channel_flush(fs->io);
Hi linux-ext4, I have a two questions with an accompanying patch for clarification. resize2fs is uninit_bg aware, but when it is expanding an ext4 filesystem, it will always zero the inode tables. Is it safe to mimick mke2fs's write_inode_table(.., lazy_flag=1) and leave the new block groups' inode tables marked INODE_UNINIT, BLOCK_UNINIT and _not_ zero out the inode table if uninit_bg is supported? If it is okay, then it means offline resizing upwards can be just as fast as mke2fs. I've attached a patch which is probably incomplete. I'd love feedback as to the feasibility of the change and/or patch quality. As a follow-on, would it be sane to add support like this for online resizing. From a cursory investigation, it looks like setup_new_block_groups() could be modified to not zero itables if uninit_bg is supported, and INODE_ZEROED could be replaced with ΒG_*_UNINIT. However, I'm not sure if that is a naive view. I'm happy to send along a patch illustrating this change if that'd be helpful or welcome. Any and all feedback is appreciated -- even if it just for me to look at the archives/link/etc. Thanks! Signed-off-by: Will Drewry <redpig@dataspill.org> --- resize/resize2fs.c | 28 ++++++++++++++++++++++------ 1 files changed, 22 insertions(+), 6 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html