linkfile: use fops->create in dht_linkfile_create

[chen1195585098]

A newly created linkfile will be deleted for the race, after this, the gfid between linkfile and datafile may get mismatched. And shards deletion process is also affected by such a mismatch, most shards in BRICK_PATH/.shard cannot be removed correctly and space cannot be release.

To fix this race, replace mknod with create during linkfile creation, in this case, we can keep this linkfile opening during the whole linkfile and datafile creations.

A linkfile will not be regarded as stale and deleted then, because its fd is opened and posix_unlink can handle it properly.

Fixes: #4548

[gluster-ant]

Can one of the admins verify this patch?

[gluster-ant]

Can one of the admins verify this patch?

[chen1195585098]

During linkfile and datafile creations, the linkfile is kept opening.

[root@chenjinhao-pc ~]# ps aux|grep glusterfsd
root       44498  0.0  0.0 1221832 23704 ?       SLsl 15:30   0:00 /usr/local/sbin/glusterfsd -s 192.168.122.211 --volfile-id issue.192.168.122.211.brick1 -p /var/run/gluster/vols/issue/192.168.122.211-brick1.pid -S /var/run/gluster/605c42f9050e3ce0.socket --brick-name /brick1 -l /var/log/glusterfs/bricks/brick1.log --xlator-option *-posix.glusterd-uuid=0f58ea27-f808-47b2-9e92-5aed82a3091f --process-name brick --brick-port 59028 --xlator-option issue-server.listen-port=59028
root       44514  0.0  0.0 1221540 23652 ?       SLsl 15:30   0:00 /usr/local/sbin/glusterfsd -s 192.168.122.211 --volfile-id issue.192.168.122.211.brick2 -p /var/run/gluster/vols/issue/192.168.122.211-brick2.pid -S /var/run/gluster/a152b698ffdd88a0.socket --brick-name /brick2 -l /var/log/glusterfs/bricks/brick2.log --xlator-option *-posix.glusterd-uuid=0f58ea27-f808-47b2-9e92-5aed82a3091f --process-name brick --brick-port 51591 --xlator-option issue-server.listen-port=51591

[root@chenjinhao-pc ~]# ll /proc/44498/fd|grep -w 9
l-wx------. 1 root root 64  6月18日 15:32 272 -> /brick1/9
lrwx------. 1 root root 64  6月18日 15:32 9 -> socket:[94888]
[root@chenjinhao-pc ~]# ll /proc/44514/fd|grep -w 9
l-wx------. 1 root root 64  6月18日 15:32 272 -> /brick2/9
lrwx------. 1 root root 64  6月18日 15:32 9 -> socket:[108826]

And a newly created linkfile will not be deleted because its fd is opend and flag open-fd-key-status is set accordingly. So, posix_unlink can handle this properly.

[root@chenjinhao-pc 1]# grep -rn posix_unlink /var/log/glusterfs/bricks/brick{1,2}.log|grep "2025-06-18"
/var/log/glusterfs/bricks/brick1.log:35587:[2025-06-18 07:33:36.604831 +0000] I [MSGID: 113030] [posix-entry-ops.c:1401:posix_unlink] 0-issue-posix: open-fd-key-status: 1 for /brick1/9 
/var/log/glusterfs/bricks/brick1.log:35591:The message "I [MSGID: 113030] [posix-entry-ops.c:1401:posix_unlink] 0-issue-posix: open-fd-key-status: 1 for /brick1/9" repeated 2 times between [2025-06-18 07:33:36.604831 +0000] and [2025-06-18 07:33:36.608125 +0000]
/var/log/glusterfs/bricks/brick2.log:34740:[2025-06-18 07:33:36.604900 +0000] I [MSGID: 113030] [posix-entry-ops.c:1401:posix_unlink] 0-issue-posix: open-fd-key-status: 1 for /brick2/9 
/var/log/glusterfs/bricks/brick2.log:34744:The message "I [MSGID: 113030] [posix-entry-ops.c:1401:posix_unlink] 0-issue-posix: open-fd-key-status: 1 for /brick2/9" repeated 2 times between [2025-06-18 07:33:36.604900 +0000] and [2025-06-18 07:33:36.608189 +0000]

[chen1195585098]

@pranithk @mohit84 Hi, can you have a look at this? Thanks.