Bug 1802013

+++ This bug was initially created as a clone of Bug #1738419 +++

+++ This bug was initially created as a clone of Bug #1737141 +++

Description of problem:

When using direct I/O, reading from a file returns more data, padding the file
data with zeroes.

Here is an example.

## On a host mounting gluster using fuse

$ pwd
/rhev/data-center/mnt/glusterSD/voodoo4.tlv.redhat.com:_gv0/de566475-5b67-4987-abf3-3dc98083b44c/dom_md


$ mount | grep glusterfs
voodoo4.tlv.redhat.com:/gv0 on /rhev/data-center/mnt/glusterSD/voodoo4.tlv.redhat.com:_gv0 type fuse.glusterfs (rw,relatime,user_id=0,group_id=0,default_permissions,allow_other,max_read=131072)


$ stat metadata 
  File: metadata
  Size: 501       	Blocks: 1          IO Block: 131072 regular file
Device: 31h/49d	Inode: 13313776956941938127  Links: 1
Access: (0644/-rw-r--r--)  Uid: (   36/    vdsm)   Gid: (   36/     kvm)
Context: system_u:object_r:fusefs_t:s0
Access: 2019-08-01 22:21:49.186381528 +0300
Modify: 2019-08-01 22:21:49.427404135 +0300
Change: 2019-08-01 22:21:49.969739575 +0300
 Birth: -


$ cat metadata 
ALIGNMENT=1048576
BLOCK_SIZE=4096
CLASS=Data
DESCRIPTION=gv0
IOOPTIMEOUTSEC=10
LEASERETRIES=3
LEASETIMESEC=60
LOCKPOLICY=
LOCKRENEWALINTERVALSEC=5
MASTER_VERSION=1
POOL_DESCRIPTION=4k-gluster
POOL_DOMAINS=de566475-5b67-4987-abf3-3dc98083b44c:Active
POOL_SPM_ID=-1
POOL_SPM_LVER=-1
POOL_UUID=44cfb532-3144-48bd-a08c-83065a5a1032
REMOTE_PATH=voodoo4.tlv.redhat.com:/gv0
ROLE=Master
SDUUID=de566475-5b67-4987-abf3-3dc98083b44c
TYPE=GLUSTERFS
VERSION=5
_SHA_CKSUM=3d1cb836f4c93679fc5a4e7218425afe473e3cfa


$ dd if=metadata bs=4096 count=1 of=/dev/null
0+1 records in
0+1 records out
501 bytes copied, 0.000340298 s, 1.5 MB/s


$ dd if=metadata bs=4096 count=1 of=/dev/null iflag=direct
1+0 records in
1+0 records out
4096 bytes (4.1 kB, 4.0 KiB) copied, 0.00398529 s, 1.0 MB/s

Checking the copied data, the actual content of the file is padded
with zeros to 4096 bytes.


## On the one of the gluster nodes

$ pwd
/export/vdo0/brick/de566475-5b67-4987-abf3-3dc98083b44c/dom_md


$ stat metadata 
  File: metadata
  Size: 501       	Blocks: 16         IO Block: 4096   regular file
Device: fd02h/64770d	Inode: 149         Links: 2
Access: (0644/-rw-r--r--)  Uid: (   36/ UNKNOWN)   Gid: (   36/     kvm)
Context: system_u:object_r:usr_t:s0
Access: 2019-08-01 22:21:50.380425478 +0300
Modify: 2019-08-01 22:21:49.427397589 +0300
Change: 2019-08-01 22:21:50.374425302 +0300
 Birth: -


$ dd if=metadata bs=4096 count=1 of=/dev/null
0+1 records in
0+1 records out
501 bytes copied, 0.000991636 s, 505 kB/s


$ dd if=metadata bs=4096 count=1 of=/dev/null iflag=direct
0+1 records in
0+1 records out
501 bytes copied, 0.0011381 s, 440 kB/s

This proves that the issue is in gluster.


# gluster volume info gv0
 
Volume Name: gv0
Type: Replicate
Volume ID: cbc5a2ad-7246-42fc-a78f-70175fb7bf22
Status: Started
Snapshot Count: 0
Number of Bricks: 1 x (2 + 1) = 3
Transport-type: tcp
Bricks:
Brick1: voodoo4.tlv.redhat.com:/export/vdo0/brick
Brick2: voodoo5.tlv.redhat.com:/export/vdo0/brick
Brick3: voodoo8.tlv.redhat.com:/export/vdo0/brick (arbiter)
Options Reconfigured:
storage.owner-gid: 36
storage.owner-uid: 36
server.event-threads: 4
client.event-threads: 4
cluster.choose-local: off
user.cifs: off
features.shard: on
cluster.shd-wait-qlength: 10000
cluster.shd-max-threads: 8
cluster.locking-scheme: granular
cluster.data-self-heal-algorithm: full
cluster.server-quorum-type: server
cluster.quorum-type: auto
cluster.eager-lock: enable
network.remote-dio: disable
performance.low-prio-threads: 32
performance.io-cache: off
performance.read-ahead: off
performance.quick-read: off
transport.address-family: inet
nfs.disable: on
performance.client-io-threads: on


$ xfs_info /export/vdo0
meta-data=/dev/mapper/vdo0       isize=512    agcount=4, agsize=6553600 blks
         =                       sectsz=4096  attr=2, projid32bit=1
         =                       crc=1        finobt=1, sparse=1, rmapbt=0
         =                       reflink=0
data     =                       bsize=4096   blocks=26214400, imaxpct=25
         =                       sunit=0      swidth=0 blks
naming   =version 2              bsize=4096   ascii-ci=0, ftype=1
log      =internal log           bsize=4096   blocks=12800, version=2
         =                       sectsz=4096  sunit=1 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0


Version-Release number of selected component (if applicable):

Server:

$ rpm -qa | grep glusterfs
glusterfs-libs-6.4-1.fc29.x86_64
glusterfs-api-6.4-1.fc29.x86_64
glusterfs-client-xlators-6.4-1.fc29.x86_64
glusterfs-fuse-6.4-1.fc29.x86_64
glusterfs-6.4-1.fc29.x86_64
glusterfs-cli-6.4-1.fc29.x86_64
glusterfs-server-6.4-1.fc29.x86_64

Client:

$ rpm -qa | grep glusterfs
glusterfs-client-xlators-6.4-1.fc29.x86_64
glusterfs-6.4-1.fc29.x86_64
glusterfs-rdma-6.4-1.fc29.x86_64
glusterfs-cli-6.4-1.fc29.x86_64
glusterfs-libs-6.4-1.fc29.x86_64
glusterfs-fuse-6.4-1.fc29.x86_64
glusterfs-api-6.4-1.fc29.x86_64


How reproducible:
Always.

Steps to Reproduce:
1. Provision gluster volume over vdo (did not check without vdo)
2. Create a file of 501 bytes
3. Read the file using direct I/O

Actual results:
read() returns 4096 bytes, padding the file data with zeroes

Expected results:
read() returns actual file data (501 bytes)

--- Additional comment from Nir Soffer on 2019-08-02 19:21:20 UTC ---

David, do you think this can affect sanlock?

--- Additional comment from Nir Soffer on 2019-08-02 19:25:02 UTC ---

Kevin, do you think this can affect qemu/qemu-img?

--- Additional comment from Amar Tumballi on 2019-08-05 05:33:57 UTC ---

@Nir, thanks for the report. We will look into this.

--- Additional comment from Kevin Wolf on 2019-08-05 09:16:16 UTC ---

(In reply to Nir Soffer from comment #2)
> Kevin, do you think this can affect qemu/qemu-img?

This is not a problem for QEMU as long as the file size is correct. If gluster didn't do the zero padding, QEMU would do it internally.

In fact, fixing this in gluster may break the case of unaligned image sizes with QEMU because the image size is rounded up to sector (512 byte) granularity and the gluster driver turns short reads into errors. This would actually affect non-O_DIRECT, too, which already seems to behave this way, so can you just give this a quick test?

--- Additional comment from David Teigland on 2019-08-05 15:08:32 UTC ---

(In reply to Nir Soffer from comment #1)
> David, do you think this can affect sanlock?

I don't think so.  sanlock doesn't use any space that it didn't first write to initialize.

--- Additional comment from Worker Ant on 2019-08-08 05:56:04 UTC ---

REVIEW: https://review.gluster.org/23175 (features/shard: Send correct size when reads are sent beyond file size) posted (#1) for review on master by Krutika Dhananjay

--- Additional comment from Worker Ant on 2019-08-12 13:30:56 UTC ---

REVIEW: https://review.gluster.org/23175 (features/shard: Send correct size when reads are sent beyond file size) merged (#3) on master by Krutika Dhananjay