In the Linux kernel, the following vulnerability has been resolved:<\/p>\n
ceph: fix hanging __ceph_get_caps() with stale mds_wanted<\/p>\n
A reader can hang forever in __ceph_get_caps() when the client no
\nlonger holds `FILE_RD`, but local cap state still says that the
\ncapability is already wanted (via `mds_wanted`).<\/p>\n
One way to trigger this is through MDS cap revocation. If another
\nclient performs a conflicting operation, the MDS can revoke `FILE_RD`
\nfrom the reader; the next read then has to reacquire `FILE_RD`. If
\nthe cap update that should request `FILE_RD` never reaches the MDS
\nafter `cap->mds_wanted` was raised, the reader is left holding only
\nnon-file caps while local `mds_wanted` still includes the file read
\ncaps.<\/p>\n
In that state, try_get_cap_refs() sees `need mds_wanted was` raised, no further request is sent and the
\nwaiter can sleep indefinitely until unrelated cap traffic happens to
\nwake it up.<\/p>\n
The ordering issue is that `cap->mds_wanted` is updated in
\n__prep_cap() before the `CEPH_MSG_CLIENT_CAPS message` is actually
\nqueued for send. That makes one field serve two different meanings at
\nonce: what this client wants, and what the client believes the MDS
\nalready knows it wants.<\/p>\n
A proper fix would be to split those states and track whether a cap
\nupdate is actually in flight or has been observed by the MDS.
\nHowever, simply moving the `cap->mds_wanted assignment` later would
\nnot be sufficient: queueing the message in the messenger does not
\nguarantee that the MDS processed that specific wanted set, and
\nreconnect or message loss can still invalidate that assumption.
\nFixing that properly would require a larger rework of the cap state
\nmachine.<\/p>\n
To allow simpler backports to stable kernels, this patch implements a
\nsimpler workaround:<\/p>\n
– stop waiting forever in __ceph_get_caps(); after a bounded wait,
\n fall back to the renew path<\/p>\n
– make ceph_renew_caps() issue a synchronous `OPEN` request whenever
\n the inode still does not actually hold the wanted caps, instead of
\n only calling ceph_check_caps()<\/p>\n
The extra issued-vs-wanted check in ceph_renew_caps() is necessary
\nbecause the previous test only checked whether the inode still had any
\nreal caps at all. That is not enough after revocation: the client can
\nstill hold something like `pLs` and yet be missing `FILE_RD`
\ncompletely. In that case, falling back to ceph_check_caps() is not
\nsufficient, because it still trusts `cap->mds_wanted` and may resend
\nnothing. By requiring `(issued & wanted) == wanted` before taking the
\nasynchronous path, the code only uses ceph_check_caps() when the
\n`wanted caps` are already actually issued. Otherwise, it sends the
\nsynchronous `OPEN` renew.<\/p>\n
This preserves the existing asynchronous fast path when the wanted
\ncaps are already issued, avoids changing cap-state semantics, and
\nfixes the hang by guaranteeing that a stalled waiter eventually
\nretries through a path that does not rely on the stale `mds_wanted`
\nstate.<\/p>\n
[ idryomov: move CEPH_GET_CAPS_WAIT_TIMEOUT from libceph.h to
\n mds_client.h, formatting ]<\/p>\n","protected":false},"excerpt":{"rendered":"
In the Linux kernel, the following vulnerability has been resolved:<\/p>\n
ceph: fix hanging __ceph_get_caps() with stale mds_wanted<\/p>\n
A reader can hang forever in __ceph_get_caps() when the client no
\nlonger<\/p>\n","protected":false},"template":"","meta":[],"categories":[],"tags":[],"class_list":["post-11940","vuln","type-vuln","status-publish","hentry"],"_links":{"self":[{"href":"https:\/\/www.54iter.cn\/index.php?rest_route=\/wp\/v2\/vuln\/11940","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.54iter.cn\/index.php?rest_route=\/wp\/v2\/vuln"}],"about":[{"href":"https:\/\/www.54iter.cn\/index.php?rest_route=\/wp\/v2\/types\/vuln"}],"wp:attachment":[{"href":"https:\/\/www.54iter.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=11940"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.54iter.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=11940"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.54iter.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=11940"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}