blob: 3743886a91dc12765454963c651019904605eb7b (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
|
When using git-annex get -J2, one thread can clear another thread's
transfer lock.
This happens because fcntl locking is used, and it is not thread-safe.
<https://github.com/haskell/unix/issues/44>
> * If a process closes any file descriptor referring to a
> file, then all of the process's locks on that file are
> released, regardless of the file descriptor(s) on which the
> locks were obtained.
One thread starts a transfer and locks it; the second thread starts a
transfer of a different file, and in order to check annex.diskreserve,
checks to see which other transfers are currently running. In doing this
check, it closes a fd attached to the first thread's lock, which causes the
lock to be dropped.
This only affects -J mode.
To fix it, probably need to use STM to keep a list of transfers all threads
in the current process are doing. Then the lock checking code can avoid
re-opening locks for transfers in the STM list.
A more generic approach is to use Annex.LockFile for everything,
and make it check its lock pool via STM for other threads holding a lock.
(Currently, the pool is only used for shared locks.)
--[[Joey]]
> [[fixed|done]] via LockPools. --[[Joey]]
|
