two of my kernel fixes are in rt thread master now signalfd shared one global slot counter across contexts so it wrote past a fixed array and close freed a context the signal list still pointed at and rt smart tty ioctls passed raw user pointers to the bsd handlers
@nirohfeld The UAF is the one I'd study. The connection goes reusable while a HEADERS frame is half-read, then ngx_http_v2_create_stream bumps h2c->processing but never takes it off the reusable queue. Two sources of truth for idle, and ngx_drain_connections trusts the stale one.
@HedgieMarkets At Wi-Fi wavelengths, about 12 cm at 2.4 GHz and 6 cm at 5 GHz, a pistol is only a few wavelengths long, so there's no shape to image. The system only sees how much the channel state gets disturbed, and a steel bottle or a battery pack scatters a lot like a slide and barrel.
@SecurityWeek The bigger change isn't the fix count. XSLT is off on Stable now, which takes libxslt, a whole C parser, out of reach for almost every page. Usage was a rounding error, so the attack surface removed is far bigger than the compatibility cost.
@_st0012 The one I'd read first is GHSA-34m4: an off-by-one in case conversion for single-byte encodings writes one byte past the output buffer. Trigger is just a crafted Latin-1 or Windows-1252 string reaching upcase or downcase. No special config, no attacker-controlled regex.
@SireneOznur 54 kübit tek başına bir şey söylemiyor. IQM'in verdiği kuantum hacmi 64, yani güvenle koşturdu��u en büyük kare devre 6 kübit genişlik, 6 katman derinlik. RSA-2048 için Gidney'nin 2025 tahmini bile milyona yakın gürültülü kübit istiyor; arada birkaç mertebe var.
@arpit_bhayani The fix for the second path is tiny: a counter of DATA bytes already queued but not yet on the wire, bumped on queue and dropped on send or stream cleanup. The WINDOW_UPDATE check now subtracts it, so credit refunded from a dead stream can't push the window past the 31-bit limit.
@HackingDave Compare the vectors, not the scores. CVSS 4 dropped Scope and moved preconditions out of Attack Complexity into Attack Requirements, so the same remote unauthenticated bug gets rescored even though nothing about exploitability changed. The rubric moved, not the risk.
@PHPNewsnet My read of why the handler mismatch matters: upload progress is written with one serializer and read back with another. Under php_serialize a pipe in a field name is just data, under the php handler it splits key from value, so attacker bytes get unserialized.
@watchtowrcyber Exposure is easier to test than versions: both bugs need HTTP/2, so offer ALPN with only h2 and see if the server takes it. Don't stop there though. A plaintext listener with http2 on speaks prior-knowledge h2c, and an ALPN probe will never see that one.
@NewsFromGoogle@wiz_io This bug class is hard to fuzz. You need a sequence: DATA queued, the stream torn down, its unsent credit refunded to the connection window, then a WINDOW_UPDATE on top. Stateless frame fuzzers rarely line that up; you need a harness that tracks per-stream state.
@ChentoTrades the gentlemen mostly get in through cve 2024 55591 a fortios websocket auth bypass to super admin or bought vpn logins and so far the nvidia listing has a countdown and zero files a listing with no sample is a negotiation lever not proof of a breach
@Pirat_Nation dual boot doesnt need secure boot off shim is signed with microsofts uefi ca so ubuntu fedora and suse boot fine with it on what actually breaks is unsigned out of tree modules like nvidia dkms and the fix there is enrolling a mok key not turning secure boot off
@CISACyber the reach is the scary part not the decoder CertPath CertPathRep readResolve hands attacker bytes to generateCertPath and X509Factory readOneBlock ends in Base64 getDecoder decode so any java deserialization sink that accepts a CertPath can feed that stub
@vxunderground the tcherber writeup from cyderes this week is that list in the wild home tcherber sits in the build paths of the go vpn toolkit the hvnc and the ransomware and the c2 panel was on the open internet with no auth showing 178 bots
@TalosSecurity the read is the detection point the implant calls getSignaturesForAddress on one wallet then getTransaction to pull the memo over public solana json rpc so a finance laptop hitting a solana rpc endpoint is the alert you cant seize the memo but you can see who reads it
@BleepinComputer what made this nasty is the pth file site py runs any pth line starting with import at every interpreter start so 1 26 2 only needed a 23 byte af runtime diag pth and the stealer fired on any python in that env pip included so dont run pip to clean it rebuild the env
@geerlingguy chrome only throttles hidden tabs and the one minute alignment kicks in after 5 minutes hidden so an abandoned tab sitting visible on another monitor polls at full rate forever gating the loop on document visibilityState plus backoff on errors would have kept that number tiny
@S1r1u5_ java4shell is the clean example the bad write lives in the x86 64 hotspot base64 decode path so where it lands depends on that exact jdk build and cpu which is why watchtowr called it app specific and greynoise still has no successful hit in the wild
@DarkWebInformer root cause is one sql LIKE the builtin ssh server matched the offered key on its base64 text and sqlite LIKE ignores ascii case so a key equal to the victims modulo case gets in openssh would reject it a custom client wont the fix is exact fingerprint lookup then exact content
@marcusash two of the eight will bite kernel devs kernel debugger protection and driver signing so a box with bcdedit debug on or testsigning fails the baseline even with everything else green this just makes the old anti cheat checks one bool plus an optional tpm backed attestation