{"id":"CVE-2026-98068","title":"In the Linux kernel, the following vulnerability has been resolved:\n\nnet/rds: don't let rds_conn_shutdown() consume a concurrent drop\n\nrds_conn_shutdown() finishes by moving the path from\nRDS_CONN_DISCONNECTING to RDS_CONN_DOWN, and also…","summary":"In the Linux kernel, the following vulnerability has been resolved:\n\nnet/rds: don't let rds_conn_shutdown() consume a concurrent drop\n\nrds_conn_shutdown() finishes by moving the path from\nRDS_CONN_DISCONNECTING to RDS_CONN_DOWN, and also…","severity":"none","vendor":"Linux","product":"Linux","affected":["Linux >= e97656d03ca0cea888a0b9d382abce8233771f31 < bd1cb197a07111ca8d4f4c21411c56a3c85a9797","Linux >= e97656d03ca0cea888a0b9d382abce8233771f31 < 4cb9b6d3d31a2dbaa5469981c2c4c03e9acc7aca","Linux >= e97656d03ca0cea888a0b9d382abce8233771f31 < 2941561395066be856a53626d4ca973dd9c982b2","Linux >= e97656d03ca0cea888a0b9d382abce8233771f31 < 260c6308fe2e19ad519389d44d582e292aecc3af","Linux 4.12"],"published":"2026-09-25","updated":"2026-09-25","sourceUpdated":"2026-09-25T11:17:35.903","source":"NVD","sourceUrl":"https://nvd.nist.gov/vuln/detail/CVE-2026-98068","references":[{"url":"https://git.kernel.org/stable/c/260c6308fe2e19ad519389d44d582e292aecc3af","label":"416baaa9-dc9f-4396-8d5f-8c081fb06d67"},{"url":"https://git.kernel.org/stable/c/2941561395066be856a53626d4ca973dd9c982b2","label":"416baaa9-dc9f-4396-8d5f-8c081fb06d67"},{"url":"https://git.kernel.org/stable/c/4cb9b6d3d31a2dbaa5469981c2c4c03e9acc7aca","label":"416baaa9-dc9f-4396-8d5f-8c081fb06d67"},{"url":"https://git.kernel.org/stable/c/bd1cb197a07111ca8d4f4c21411c56a3c85a9797","label":"416baaa9-dc9f-4396-8d5f-8c081fb06d67"}],"tags":["nvd","cve.org"],"ingestedAt":"2026-09-25T11:06:38.839Z","slug":"CVE-2026-98068","body":"## Overview\n\nIn the Linux kernel, the following vulnerability has been resolved:\n\nnet/rds: don't let rds_conn_shutdown() consume a concurrent drop\n\nrds_conn_shutdown() finishes by moving the path from\nRDS_CONN_DISCONNECTING to RDS_CONN_DOWN, and also accepts\nRDS_CONN_ERROR as the starting state of that final transition, so that\na FIN processed in softirq context during the teardown does not derail\nthe shutdown into a noisy error path.\n\nBut consuming that RDS_CONN_ERROR also consumes the shutdown pass that\ncame with it: rds_conn_path_drop() sets RDS_CONN_ERROR and then queues\ncp_down_w, and a pass that starts on a path already in RDS_CONN_DOWN\nis a no-op.  For the FIN case that is harmless - the socket the FIN\narrived on is the very socket the teardown just released.  It is not\nharmless for a dropper that attached something to the path first.\n\nrds_tcp_accept_one() is such a dropper.  Its path claim in\nrds_tcp_accept_one_path() transitions RDS_CONN_DOWN ->\nRDS_CONN_CONNECTING, and a concurrent drop - a FIN on a previous\nsocket in softirq context, an administrative reset - can put the path\ninto RDS_CONN_ERROR between that claim and the state check that\nfollows, which accepts RDS_CONN_ERROR.  The accept then installs the\nfreshly accepted socket with rds_tcp_set_callbacks() while the queued\nteardown - which sampled tc->t_sock before this socket existed - is\nstill running.  rds_connect_path_complete() fails its transition to\nRDS_CONN_UP and drops the path again, queueing the pass that should\nreap the socket it just installed.  If the in-flight shutdown's final\ntransition consumes that drop's RDS_CONN_ERROR, the queued pass finds\nthe path in RDS_CONN_DOWN and does nothing.  The installed socket is\nnever torn down: it sits established with its callbacks armed and its\nrds_tcp_connection on rds_tcp_tc_list, the peer sees a connection that\nnothing ever reads, and the path is wedged in RDS_CONN_DOWN until some\nlater event drops it again.  Reproduced with widened race windows as\nan ever-growing receive queue on a socket owned by a path stuck in\nRDS_CONN_DOWN, with the peer's send path wedged behind it.\n\nMake the final transition only DISCONNECTING -> DOWN.  If it fails\nbecause the path is in RDS_CONN_ERROR, a drop raced the teardown:\ncancel the reconnect timer and clear RDS_RECONNECT_PENDING - the one\npiece of the skipped tail that must not be left behind - and return,\nletting the pass the drop queued finish the job: it tears down\nwhatever attached to the path in the meantime, completes the\ntransition to RDS_CONN_DOWN, and re-arms the reconnect from its own\ntail.\n\nThe timer quiesce in that branch matters because the racing drop does\nnot always queue that pass: rds_conn_path_drop() returns without\nqueueing when a destroy is pending - exactly the situation during a\nnetns teardown or module unload, when a FIN on the dying socket is\nprocessed while rds_conn_path_destroy() flushes cp_down_w.  If the\nflushed pass is the one that takes this return, no later pass exists,\nand rds_conn_path_destroy() would find cp_conn_w still armed\n(WARN_ON) and then free a path whose reconnect timer can still fire.\nWith the cancel in the branch, every exit of a shutdown pass leaves\nthe timer quiesced no matter which pass completes the transition.\n\nThe FIN case keeps making progress, one pass later and still without\nnoisy logging.  Any other state keeps today's rds_conn_path_error()\nhandling; no current cp_state writer can leave a DISCONNECTING path\nin anything but RDS_CONN_ERROR (every other writer is a cmpxchg from\na non-DISCONNECTING state), so that branch is defensive.\n\nOn kernels without the preceding patches the same hazard exists with\nthe sample-based quiesce; the fix applies there equally.\n\n## Remediation\n\nRefer to the linked advisories for vendor-supplied fixes and affected version ranges.","depth":"sunlit","depthScore":3,"depthScoreParts":{"impact":2.8,"likelihood":0,"exploitation":0,"ransomware":0},"changes":[]}