{"id":"CVE-2026-90206","published":"2026-09-17T17:17:15.700","lastModified":"2026-09-17T17:17:15.700","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nnvmet: fix max_qid race between configfs and controller allocation\n\nThe function nvmet_subsys_attr_qid_max_store() can race against\nnvmet_alloc_ctrl() when a subsystem's max_qid limit is modified.\n\nSuppose max_qid is currently 64. If nvmet_alloc_ctrl() executes:\nctrl->sqs = kzalloc_objs(struct nvmet_sq *, subsys->max_qid + 1);\nand at this exact point, a userspace process changes max_qid to 128,\nnvmet_subsys_attr_qid_max_store() will set the new max_qid value. It\nattempts to delete active controllers to force a reconnect, but the\nnew controller won't be deleted because it hasn't been added to the\nsubsys->ctrls list yet.\n\nnvmet_alloc_ctrl() then proceeds and adds the new controller to the\nsubsys->ctrls list. Later, when nvmet_install_queue() is called, it\nwill see max_qid set to 128, but the memory allocated for sqs is only\nsized for 64 entries. This results in a KASAN out-of-bounds warning\nand potential memory corruptions.\n\nFix this by protecting the queue allocations and list insertion in\nnvmet_alloc_ctrl() with down_read(&nvmet_config_sem). Because\nnvmet_subsys_attr_qid_max_store() acquires down_write(&nvmet_config_sem)\nto modify the attribute, this safely prevents the configfs writer from\nmodifying max_qid during controller creation.\n\nCopy the max_qid from the subsystem to the controller's structure\nduring the allocation; ctrl->max_qid never changes as long as the\ncontroller remains in LIVE state, so this will prevent similar race\nconditions.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/2c23fc91789dfd9db746edb3ae9b90d65fabd410","tags":[]},{"url":"https://git.kernel.org/stable/c/f1a8846e06388113dfdbb89dee005083fa9afdf9","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}