Eric Ren
2016-Aug-30 03:33 UTC
[Ocfs2-devel] [PATCH v2] ocfs2: Fix start offset to ocfs2_zero_range_for_truncate()
Hello, On 08/30/2016 03:23 AM, Ashish Samant wrote:> Hi Eric, > > The easiest way to reproduce this is : > > 1. Create a random file of say 10 MB > xfs_io -c 'pwrite -b 4k 0 10M' -f 10MBfile > 2. Reflink it > reflink -f 10MBfile reflnktest > 3. Punch a hole at starting at cluster boundary with range greater that 1MB. You can also > use a range that will put the end offset in another extent. > fallocate -p -o 0 -l 1048615 reflnktest > 4. sync > 5. Check the first cluster in the source file. (It will be zeroed out). > dd if=10MBfile iflag=direct bs=<cluster size> count=1 | hexdump -CThanks! I have a try myself, but I'm not sure what is our expected output and if the test result meet it: 1. After applying this patch: ocfs2dev1:/mnt/ocfs2 # rm 10MBfile reflnktest ocfs2dev1:/mnt/ocfs2 # xfs_io -c 'pwrite -b 4k 0 10M' -f 10MBfile wrote 10485760/10485760 bytes at offset 0 10 MiB, 2560 ops; 0.0000 sec (1.089 GiB/sec and 285427.5839 ops/sec) ocfs2dev1:/mnt/ocfs2 # reflink -f 10MBfile reflnktest ocfs2dev1:/mnt/ocfs2 # fallocate -p -o 0 -l 1048615 reflnktest ocfs2dev1:/mnt/ocfs2 # dd if=10MBfile iflag=direct bs=1048576 count=1 | hexdump -C 00000000 cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd |................| * 1+0 records in 1+0 records out 1048576 bytes (1.0 MB, 1.0 MiB) copied, 0.0952464 s, 11.0 MB/s 00100000 2. Before this patch: .... ocfs2dev1:/mnt/ocfs2 # dd if=10MBfile iflag=direct bs=1048576 count=1 | hexdump -C 00000000 cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd |................| * 1+0 records in 1+0 records out 1048576 bytes (1.0 MB, 1.0 MiB) copied, 0.0954648 s, 11.0 MB/s 00100000 3. debugfs.ocfs2 -R stats /dev/sdb ... Block Size Bits: 12 Cluster Size Bits: 20 ... Eric> > Thanks, > Ashish > > On 08/28/2016 10:39 PM, Eric Ren wrote: >> Hi, >> >> Thanks for this fix. I'd like to reproduce this issue locally and test this patch, >> could you elaborate the detailed steps of reproduction? >> >> Thanks, >> Eric >> >> On 08/27/2016 07:04 AM, Ashish Samant wrote: >>> If we punch a hole on a reflink such that following conditions are met: >>> >>> 1. start offset is on a cluster boundary >>> 2. end offset is not on a cluster boundary >>> 3. (end offset is somewhere in another extent) or >>> (hole range > MAX_CONTIG_BYTES(1MB)), >>> >>> we dont COW the first cluster starting at the start offset. But in this >>> case, we were wrongly passing this cluster to >>> ocfs2_zero_range_for_truncate() to zero out. This will modify the cluster >>> in place and zero it in the source too. >>> >>> Fix this by skipping this cluster in such a scenario. >>> >>> Reported-by: Saar Maoz <saar.maoz at oracle.com> >>> Signed-off-by: Ashish Samant <ashish.samant at oracle.com> >>> Reviewed-by: Srinivas Eeda <srinivas.eeda at oracle.com> >>> --- >>> v1->v2: >>> -Changed the commit msg to include a better and generic description of >>> the problem, for all cluster sizes. >>> -Added Reported-by and Reviewed-by tags. >>> fs/ocfs2/file.c | 34 ++++++++++++++++++++++++---------- >>> 1 file changed, 24 insertions(+), 10 deletions(-) >>> >>> diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c >>> index 4e7b0dc..0b055bf 100644 >>> --- a/fs/ocfs2/file.c >>> +++ b/fs/ocfs2/file.c >>> @@ -1506,7 +1506,8 @@ static int ocfs2_zero_partial_clusters(struct inode *inode, >>> u64 start, u64 len) >>> { >>> int ret = 0; >>> - u64 tmpend, end = start + len; >>> + u64 tmpend = 0; >>> + u64 end = start + len; >>> struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); >>> unsigned int csize = osb->s_clustersize; >>> handle_t *handle; >>> @@ -1538,18 +1539,31 @@ static int ocfs2_zero_partial_clusters(struct inode *inode, >>> } >>> /* >>> - * We want to get the byte offset of the end of the 1st cluster. >>> + * If start is on a cluster boundary and end is somewhere in another >>> + * cluster, we have not COWed the cluster starting at start, unless >>> + * end is also within the same cluster. So, in this case, we skip this >>> + * first call to ocfs2_zero_range_for_truncate() truncate and move on >>> + * to the next one. >>> */ >>> - tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); >>> - if (tmpend > end) >>> - tmpend = end; >>> + if ((start & (csize - 1)) != 0) { >>> + /* >>> + * We want to get the byte offset of the end of the 1st >>> + * cluster. >>> + */ >>> + tmpend = (u64)osb->s_clustersize + >>> + (start & ~(osb->s_clustersize - 1)); >>> + if (tmpend > end) >>> + tmpend = end; >>> - trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start, >>> - (unsigned long long)tmpend); >>> + trace_ocfs2_zero_partial_clusters_range1( >>> + (unsigned long long)start, >>> + (unsigned long long)tmpend); >>> - ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); >>> - if (ret) >>> - mlog_errno(ret); >>> + ret = ocfs2_zero_range_for_truncate(inode, handle, start, >>> + tmpend); >>> + if (ret) >>> + mlog_errno(ret); >>> + } >>> if (tmpend < end) { >>> /* >> >> >
Ashish Samant
2016-Aug-30 04:11 UTC
[Ocfs2-devel] [PATCH v2] ocfs2: Fix start offset to ocfs2_zero_range_for_truncate()
Hmm, thats weird. I see this on 4.7 kernel without the patch: # xfs_io -c 'pwrite -b 4k 0 10M' -f 10MBfile wrote 10485760/10485760 bytes at offset 0 10 MiB, 2560 ops; 0.0000 sec (683.995 MiB/sec and 175102.5992 ops/sec) # reflink -f 10MBfile reflnktest # fallocate -p -o 0 -l 1048615 reflnktest # dd if=10MBfile iflag=direct bs=1048576 count=1 | hexdump -C 00000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................| * 1+0 records in 1+0 records out 1048576 bytes (1.0 MB) copied, 0.0321517 s, 32.6 MB/s 00100000 and with patch ---- # dd if=10MBfile iflag=direct bs=1M count=1 | hexdump -C 00000000 cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd |................| * 1+0 records in 1+0 records out 00100000 Thanks, Ashish On 08/29/2016 08:33 PM, Eric Ren wrote:> Hello, > > On 08/30/2016 03:23 AM, Ashish Samant wrote: >> Hi Eric, >> >> The easiest way to reproduce this is : >> >> 1. Create a random file of say 10 MB >> xfs_io -c 'pwrite -b 4k 0 10M' -f 10MBfile >> 2. Reflink it >> reflink -f 10MBfile reflnktest >> 3. Punch a hole at starting at cluster boundary with range greater >> that 1MB. You can also use a range that will put the end offset in >> another extent. >> fallocate -p -o 0 -l 1048615 reflnktest >> 4. sync >> 5. Check the first cluster in the source file. (It will be zeroed out). >> dd if=10MBfile iflag=direct bs=<cluster size> count=1 | hexdump -C > > Thanks! I have a try myself, but I'm not sure what is our expected > output and if the test result meet > it: > > 1. After applying this patch: > ocfs2dev1:/mnt/ocfs2 # rm 10MBfile reflnktest > ocfs2dev1:/mnt/ocfs2 # xfs_io -c 'pwrite -b 4k 0 10M' -f 10MBfile > wrote 10485760/10485760 bytes at offset 0 > 10 MiB, 2560 ops; 0.0000 sec (1.089 GiB/sec and 285427.5839 ops/sec) > ocfs2dev1:/mnt/ocfs2 # reflink -f 10MBfile reflnktest > ocfs2dev1:/mnt/ocfs2 # fallocate -p -o 0 -l 1048615 reflnktest > ocfs2dev1:/mnt/ocfs2 # dd if=10MBfile iflag=direct bs=1048576 count=1 > | hexdump -C > 00000000 cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd > |................| > * > 1+0 records in > 1+0 records out > 1048576 bytes (1.0 MB, 1.0 MiB) copied, 0.0952464 s, 11.0 MB/s > 00100000 > > 2. Before this patch: > .... > ocfs2dev1:/mnt/ocfs2 # dd if=10MBfile iflag=direct bs=1048576 count=1 > | hexdump -C > 00000000 cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd cd > |................| > * > 1+0 records in > 1+0 records out > 1048576 bytes (1.0 MB, 1.0 MiB) copied, 0.0954648 s, 11.0 MB/s > 00100000 > > 3. debugfs.ocfs2 -R stats /dev/sdb > ... > Block Size Bits: 12 Cluster Size Bits: 20 > ... > > Eric >> >> Thanks, >> Ashish >> >> On 08/28/2016 10:39 PM, Eric Ren wrote: >>> Hi, >>> >>> Thanks for this fix. I'd like to reproduce this issue locally and >>> test this patch, >>> could you elaborate the detailed steps of reproduction? >>> >>> Thanks, >>> Eric >>> >>> On 08/27/2016 07:04 AM, Ashish Samant wrote: >>>> If we punch a hole on a reflink such that following conditions are >>>> met: >>>> >>>> 1. start offset is on a cluster boundary >>>> 2. end offset is not on a cluster boundary >>>> 3. (end offset is somewhere in another extent) or >>>> (hole range > MAX_CONTIG_BYTES(1MB)), >>>> >>>> we dont COW the first cluster starting at the start offset. But in >>>> this >>>> case, we were wrongly passing this cluster to >>>> ocfs2_zero_range_for_truncate() to zero out. This will modify the >>>> cluster >>>> in place and zero it in the source too. >>>> >>>> Fix this by skipping this cluster in such a scenario. >>>> >>>> Reported-by: Saar Maoz <saar.maoz at oracle.com> >>>> Signed-off-by: Ashish Samant <ashish.samant at oracle.com> >>>> Reviewed-by: Srinivas Eeda <srinivas.eeda at oracle.com> >>>> --- >>>> v1->v2: >>>> -Changed the commit msg to include a better and generic description of >>>> the problem, for all cluster sizes. >>>> -Added Reported-by and Reviewed-by tags. >>>> fs/ocfs2/file.c | 34 ++++++++++++++++++++++++---------- >>>> 1 file changed, 24 insertions(+), 10 deletions(-) >>>> >>>> diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c >>>> index 4e7b0dc..0b055bf 100644 >>>> --- a/fs/ocfs2/file.c >>>> +++ b/fs/ocfs2/file.c >>>> @@ -1506,7 +1506,8 @@ static int ocfs2_zero_partial_clusters(struct >>>> inode *inode, >>>> u64 start, u64 len) >>>> { >>>> int ret = 0; >>>> - u64 tmpend, end = start + len; >>>> + u64 tmpend = 0; >>>> + u64 end = start + len; >>>> struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); >>>> unsigned int csize = osb->s_clustersize; >>>> handle_t *handle; >>>> @@ -1538,18 +1539,31 @@ static int >>>> ocfs2_zero_partial_clusters(struct inode *inode, >>>> } >>>> /* >>>> - * We want to get the byte offset of the end of the 1st cluster. >>>> + * If start is on a cluster boundary and end is somewhere in >>>> another >>>> + * cluster, we have not COWed the cluster starting at start, >>>> unless >>>> + * end is also within the same cluster. So, in this case, we >>>> skip this >>>> + * first call to ocfs2_zero_range_for_truncate() truncate and >>>> move on >>>> + * to the next one. >>>> */ >>>> - tmpend = (u64)osb->s_clustersize + (start & >>>> ~(osb->s_clustersize - 1)); >>>> - if (tmpend > end) >>>> - tmpend = end; >>>> + if ((start & (csize - 1)) != 0) { >>>> + /* >>>> + * We want to get the byte offset of the end of the 1st >>>> + * cluster. >>>> + */ >>>> + tmpend = (u64)osb->s_clustersize + >>>> + (start & ~(osb->s_clustersize - 1)); >>>> + if (tmpend > end) >>>> + tmpend = end; >>>> - trace_ocfs2_zero_partial_clusters_range1((unsigned long >>>> long)start, >>>> - (unsigned long long)tmpend); >>>> + trace_ocfs2_zero_partial_clusters_range1( >>>> + (unsigned long long)start, >>>> + (unsigned long long)tmpend); >>>> - ret = ocfs2_zero_range_for_truncate(inode, handle, start, >>>> tmpend); >>>> - if (ret) >>>> - mlog_errno(ret); >>>> + ret = ocfs2_zero_range_for_truncate(inode, handle, start, >>>> + tmpend); >>>> + if (ret) >>>> + mlog_errno(ret); >>>> + } >>>> if (tmpend < end) { >>>> /* >>> >>> >> >