* [9fans] ramfs bug
@ 2022-08-15 11:02 Richard Miller
2022-08-15 14:20 ` adr
0 siblings, 1 reply; 2+ messages in thread
From: Richard Miller @ 2022-08-15 11:02 UTC (permalink / raw)
To: 9fans
Deleting an open file on ramfs(4) can have surpising results.
Here's how it's supposed to behave. When a file is removed, the
Ram struct which represents the in-memory file has its busy flag
reset, so a subsequent i/o will see that it's been deleted.
cpu% ramfs
cpu% echo this is x >/tmp/x
cpu% { sleep 10; cat } </tmp/x &
cpu% rm /tmp/x
cpu% cat: error reading <stdin>: file does not exist
However, before the i/o takes place, it's possible for another
file to be created and re-use the same Ram struct, setting
the busy flag again. The process which had the deleted file open
is now doing i/o to a different file, without any checks on its
permissions, exclusive use flag, etc.
cpu% ramfs
cpu% echo this is x >/tmp/x
cpu% { sleep 10; cat } </tmp/x &
cpu% rm /tmp/x; echo actually this is y >/tmp/y
cpu% actually this is y
A simple fix is to add a reference count to the Ram struct, to
keep track of how many Fids point to it, and only allow a Ram to
be re-used if both busy==0 and ref==0. The new 9legacy patch
ramfs-refcount makes this change.
Note to 9front users: don't worry, your ramfs is a completely
different program using the 9p(2) library, which doesn't have
this flaw.
------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/Tc364b766ff6473bc-M52976e87f40d78ae095f576e
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2022-08-15 14:21 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-08-15 11:02 [9fans] ramfs bug Richard Miller
2022-08-15 14:20 ` adr
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).