CVE-2022-49536

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> scsi: lpfc: Fix SCSI I/O completion and abort handler deadlock<br /> <br /> During stress I/O tests with 500+ vports, hard LOCKUP call traces are<br /> observed.<br /> <br /> CPU A:<br /> native_queued_spin_lock_slowpath+0x192<br /> _raw_spin_lock_irqsave+0x32<br /> lpfc_handle_fcp_err+0x4c6<br /> lpfc_fcp_io_cmd_wqe_cmpl+0x964<br /> lpfc_sli4_fp_handle_cqe+0x266<br /> __lpfc_sli4_process_cq+0x105<br /> __lpfc_sli4_hba_process_cq+0x3c<br /> lpfc_cq_poll_hdler+0x16<br /> irq_poll_softirq+0x76<br /> __softirqentry_text_start+0xe4<br /> irq_exit+0xf7<br /> do_IRQ+0x7f<br /> <br /> CPU B:<br /> native_queued_spin_lock_slowpath+0x5b<br /> _raw_spin_lock+0x1c<br /> lpfc_abort_handler+0x13e<br /> scmd_eh_abort_handler+0x85<br /> process_one_work+0x1a7<br /> worker_thread+0x30<br /> kthread+0x112<br /> ret_from_fork+0x1f<br /> <br /> Diagram of lockup:<br /> <br /> CPUA CPUB<br /> ---- ----<br /> lpfc_cmd-&gt;buf_lock<br /> phba-&gt;hbalock<br /> lpfc_cmd-&gt;buf_lock<br /> phba-&gt;hbalock<br /> <br /> Fix by reordering the taking of the lpfc_cmd-&gt;buf_lock and phba-&gt;hbalock in<br /> lpfc_abort_handler routine so that it tries to take the lpfc_cmd-&gt;buf_lock<br /> first before phba-&gt;hbalock.