From mboxrd@z Thu Jan 1 00:00:00 1970 From: tm@tao.ma Subject: Re: [PATCH 4/4 v2] ext4: Speed up FITRIM by recording flags in ext4_group_info. Date: Thu, 10 Feb 2011 07:58:13 -0700 Message-ID: References: <1297231048-3458-4-git-send-email-tm@tao.ma> <1297323206-9874-1-git-send-email-tm@tao.ma> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7BIT Cc: "Tao Ma" , linux-ext4@vger.kernel.org, "Andreas Dilger" , "Lukas Czerner" To: "Lukas Czerner" Return-path: Received: from cpoproxy2-pub.bluehost.com ([67.222.39.38]:34762 "HELO cpoproxy2-pub.bluehost.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S1751179Ab1BJO6Q (ORCPT ); Thu, 10 Feb 2011 09:58:16 -0500 In-Reply-To: Sender: linux-ext4-owner@vger.kernel.org List-ID: > On Thu, 10 Feb 2011, Tao Ma wrote: > >> From: Tao Ma >> >> In ext4, when FITRIM is called every time, we iterate all the >> groups and do trim one by one. It is a bit time wasting if the >> group has been trimmed and there is no change since the last >> trim. >> >> So this patch adds a new flag in ext4_group_info->bb_state to >> indicate that the group has been trimmed, and it will be cleared >> if some blocks is freed(in release_blocks_on_commit). Another >> trim_minlen is added in ext4_sb_info to record the last minlen >> we use to trim the volume, so that if the caller provide a small >> one, we will go on the trim regardless of the bb_state. >> >> A simple test with my intel x25m ssd: >> df -h shows: >> /dev/sdb2 108G 35G 68G 34% /mnt/ext4 >> Block size: 4096 >> >> run the FITRIM with the following parameter: >> range.start = 0; >> range.len = UINT64_MAX; >> range.minlen = 1048576; >> >> without the patch: >> [root@boyu-tm test]# time ./ftrim /mnt/ext4/a >> real 0m4.039s >> user 0m0.000s >> sys 0m1.020s >> [root@boyu-tm test]# time ./ftrim /mnt/ext4/a >> real 0m3.577s >> user 0m0.001s >> sys 0m1.004s >> [root@boyu-tm test]# time ./ftrim /mnt/ext4/a >> real 0m3.380s >> user 0m0.000s >> sys 0m0.991s >> >> with the patch: >> [root@boyu-tm test]# time ./ftrim /mnt/ext4/a >> real 0m3.466s >> user 0m0.000s >> sys 0m0.966s >> [root@boyu-tm test]# time ./ftrim /mnt/ext4/a >> real 0m0.001s >> user 0m0.000s >> sys 0m0.001s >> [root@boyu-tm test]# time ./ftrim /mnt/ext4/a >> real 0m0.001s >> user 0m0.000s >> sys 0m0.000s >> >> A big improvement for the 2nd and 3rd run. >> >> After I delete some big image files and re-run the trim, >> it is still much faster than iterating the whole disk. >> /dev/sdb2 108G 25G 78G 24% /mnt/ext4 >> >> [root@boyu-tm test]# time ./ftrim /mnt/ext4/a >> real 0m0.513s >> user 0m0.000s >> sys 0m0.069s > > Great it looks really good. > >> >> Cc: Andreas Dilger >> Cc: Lukas Czerner >> Signed-off-by: Tao Ma >> --- >> fs/ext4/ext4.h | 8 +++++++- >> fs/ext4/mballoc.c | 22 ++++++++++++++++++++++ >> 2 files changed, 29 insertions(+), 1 deletions(-) >> >> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h >> index 0c8d97b..1d59a63 100644 >> --- a/fs/ext4/ext4.h >> +++ b/fs/ext4/ext4.h >> @@ -1200,6 +1200,9 @@ struct ext4_sb_info { >> struct ext4_li_request *s_li_request; >> /* Wait multiplier for lazy initialization thread */ >> unsigned int s_li_wait_mult; >> + >> + /* record the last minlen when FITRIM is called. */ >> + u64 s_last_trim_minblks; >> }; >> >> static inline struct ext4_sb_info *EXT4_SB(struct super_block *sb) >> @@ -1970,10 +1973,13 @@ struct ext4_group_info { >> * 5 free 8-block regions. */ >> }; >> >> -#define EXT4_GROUP_INFO_NEED_INIT_BIT 0 >> +#define EXT4_GROUP_INFO_NEED_INIT_BIT 0 >> +#define EXT4_GROUP_INFO_WAS_TRIMMED_BIT 1 >> >> #define EXT4_MB_GRP_NEED_INIT(grp) \ >> (test_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &((grp)->bb_state))) >> +#define EXT4_MB_GRP_HAS_BEEN_TRIMMED(grp) \ >> + (test_bit(EXT4_GROUP_INFO_WAS_TRIMMED_BIT, &((grp)->bb_state))) >> >> #define EXT4_MAX_CONTENTION 8 >> #define EXT4_CONTENTION_THRESHOLD 2 >> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c >> index 4eadac8..c7aa094 100644 >> --- a/fs/ext4/mballoc.c >> +++ b/fs/ext4/mballoc.c >> @@ -2687,6 +2687,16 @@ static void release_blocks_on_commit(journal_t >> *journal, transaction_t *txn) >> rb_erase(&entry->node, &(db->bb_free_root)); >> mb_free_blocks(NULL, &e4b, entry->start_blk, entry->count); >> >> + /* >> + * Clear the trimmed flag for the group so that the next >> + * ext4_trim_fs can trim it. >> + * If the volume is mounted with -o discard, online discard >> + * is supported and the free blocks will be trimmed online. >> + */ >> + if (!test_opt(sb, DISCARD)) >> + clear_bit(EXT4_GROUP_INFO_WAS_TRIMMED_BIT, >> + &(db->bb_state)); >> + >> if (!db->bb_free_root.rb_node) { >> /* No more items in the per group rb tree >> * balance refcounts from ext4_mb_free_metadata() >> @@ -4772,6 +4782,10 @@ ext4_grpblk_t ext4_trim_all_free(struct >> super_block *sb, struct ext4_buddy *e4b, >> >> ext4_lock_group(sb, group); >> >> + if (EXT4_MB_GRP_HAS_BEEN_TRIMMED(e4b->bd_info) && >> + minblocks >= EXT4_SB(sb)->s_last_trim_minblks) >> + goto out; >> + >> trace_ext4_trim_all_free(sb, group, start, max); >> >> while (start < max) { >> @@ -4804,6 +4818,11 @@ ext4_grpblk_t ext4_trim_all_free(struct >> super_block *sb, struct ext4_buddy *e4b, >> if ((e4b->bd_info->bb_free - free_count) < minblocks) >> break; >> } >> + >> + if (!ret) >> + set_bit(EXT4_GROUP_INFO_WAS_TRIMMED_BIT, >> + &(e4b->bd_info->bb_state)); >> +out: >> ext4_unlock_group(sb, group); >> >> ext4_debug("trimmed %d blocks in the group %d\n", >> @@ -4892,6 +4911,9 @@ int ext4_trim_fs(struct super_block *sb, struct >> fstrim_range *range) >> } >> range->len = trimmed * sb->s_blocksize; >> >> + if (!ret) >> + EXT4_SB(sb)->s_last_trim_minblks = minlen; >> + > > Since this is not protected by any lock, would not it race in case of > multiple FITRIM calls ? yeah, I am also thinking of this, but I don't think we need a new lock just for this. And I guess atomic_t isn't good here because minlen is a u64. Do you think we can use some other spin_lock in ext4 system? I am not quite familiar with ext4 by now, so do you have any suggestion? Regards, Tao