eCryptFS race window?
Hi Michael,
it seems there is a few potential race window in eCryptFS which I was trying to fix but it requires more deeper eCrypFS knowledge that have (at least only by understanding eCryptFS in big picture it is possible to fix this problem by elegant path). So what is the problem - the procedures
ecryptfs_miscdev_poll
ecryptfs_miscdev_read
does take ecryptfs_daemon_hash_mux mutex and then daemon->mux _but_ releases them not in exactly backward order as it should. My patches (not in mainline but you saw them) was screwed up 'cause mutex_is_locked could release mutex acquired by another process and that is wrong. But I've a hope that I'm simply *wrong* about this possible races ;) Take a look please.
- Cyrill -
Quoted 9 lines · Cyrill Gorcunov
it seems there is a few potential race window in eCryptFS which I was trying to fix but it requires more deeper eCrypFS knowledge that have (at least only by understanding eCryptFS in big picture it is possible to fix this problem by elegant path). So what is the problem - the procedures
ecryptfs_miscdev_poll
ecryptfs_miscdev_read
Quoted 6 lines
does take
ecryptfs_daemon_hash_muxmutex and thendaemon->mux_but_ releases them not in exactly backward order as it should. My patches (not in mainline but you saw them) was screwed up 'causemutex_is_lockedcould release mutex acquired by another process and that is wrong. But I've a hope that I'm simply *wrong* about this possible races ;) Take a look please.
I cannot find an execution path whereby one of the two must re-acquire ecryptfs_daemon_hash_mux in order to release daemon->mux, so I do not think we will ever have deadlock between those two functions.
Mike
[Michael Halcrow - Tue, May 20, 2008 at 01:27:57PM -0500]
| On Tue, May 20, 2008 at 09:03:21PM +0400, Cyrill Gorcunov wrote: | > it seems there is a few potential race window in eCryptFS which I | > was trying to fix but it requires more deeper eCrypFS knowledge that | > have (at least only by understanding eCryptFS in big picture it is | > possible to fix this problem by elegant path). So what is the | > problem - the procedures | > | > ecryptfs_miscdev_poll | > ecryptfs_miscdev_read | > | > does take ecryptfs_daemon_hash_mux mutex and then daemon->mux _but_ | > releases them not in exactly backward order as it should. My patches | > (not in mainline but you saw them) was screwed up 'cause | > mutex_is_locked could release mutex acquired by another process and | > that is wrong. But I've a hope that I'm simply *wrong* about this | > possible races ;) Take a look please. | | I cannot find an execution path whereby one of the two must re-acquire | ecryptfs_daemon_hash_mux in order to release daemon->mux, so I do not | think we will ever have deadlock between those two functions. | | Mike |
ok, that means I'm just wrong, thanks
- Cyrill -
[Michael Halcrow - Tue, May 20, 2008 at 01:27:57PM -0500]
| On Tue, May 20, 2008 at 09:03:21PM +0400, Cyrill Gorcunov wrote: | > it seems there is a few potential race window in eCryptFS which I | > was trying to fix but it requires more deeper eCrypFS knowledge that | > have (at least only by understanding eCryptFS in big picture it is | > possible to fix this problem by elegant path). So what is the | > problem - the procedures | > | > ecryptfs_miscdev_poll | > ecryptfs_miscdev_read | > | > does take ecryptfs_daemon_hash_mux mutex and then daemon->mux _but_ | > releases them not in exactly backward order as it should. My patches | > (not in mainline but you saw them) was screwed up 'cause | > mutex_is_locked could release mutex acquired by another process and | > that is wrong. But I've a hope that I'm simply *wrong* about this | > possible races ;) Take a look please. | | I cannot find an execution path whereby one of the two must re-acquire | ecryptfs_daemon_hash_mux in order to release daemon->mux, so I do not | think we will ever have deadlock between those two functions. | | Mike |
Anyway, don't apply my patch you've tested (which I've replied to Ingo's circular dep report), please, it's screwed ;) The patch Andrew has taken to its tree is ok.
- Cyrill -