Date: Sun, 19 Mar 2006 20:25:26 GMT From: Robert Watson <rwatson@FreeBSD.org> To: Perforce Change Reviews <perforce@freebsd.org> Subject: PERFORCE change 93596 for review Message-ID: <200603192025.k2JKPQYd023419@repoman.freebsd.org>
next in thread | raw e-mail | index | archive | help
http://perforce.freebsd.org/chv.cgi?CH=93596 Change 93596 by rwatson@rwatson_peppercorn on 2006/03/19 20:24:58 No longer unconditionally drain the audit record queue if there is not an active audit trail: instead, continue to iterate through each record in case an audit pipe is interested. Affected files ... .. //depot/projects/trustedbsd/audit3/sys/security/audit/audit_worker.c#8 edit Differences ... ==== //depot/projects/trustedbsd/audit3/sys/security/audit/audit_worker.c#8 (text+ko) ==== @@ -304,24 +304,6 @@ } /* - * Drain the audit commit queue and free the records. Used if there are - * records present, but no audit log target. - */ -static void -audit_worker_drain(void) -{ - struct kaudit_record *ar; - - mtx_assert(&audit_mtx, MA_OWNED); - - while ((ar = TAILQ_FIRST(&audit_q))) { - TAILQ_REMOVE(&audit_q, ar, k_q); - audit_free(ar); - audit_q_len--; - } -} - -/* * Given a kernel audit record, process as required. Kernel audit records * are converted to one, or possibly two, BSM records, depending on whether * there is a user audit record present also. Kernel records need be @@ -424,23 +406,11 @@ audit_worker_rotate(&audit_cred, &audit_vp, audit_td); /* - * If we have records, but there's no active vnode to write - * to, drain the record queue. Generally, we prevent the - * unnecessary allocation of records elsewhere, but we need - * to allow for races between conditional allocation and - * queueing. Go back to waiting when we're done. - */ - if (audit_vp == NULL) { - audit_worker_drain(); - continue; - } - - /* - * We have both records to write and an active vnode to write - * to. Dequeue a record, and start the write. Eventually, - * it might make sense to dequeue several records and perform - * our own clustering, if the lower layers aren't doing it - * automatically enough. + * If there are records in the global audit record queue, + * transfer them to a thread-local queue and process them + * one by one. If we cross the low watermark threshold, + * signal any waiting processes that they may wake up and + * continue generating records. */ lowater_signal = 0; while ((ar = TAILQ_FIRST(&audit_q))) {
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200603192025.k2JKPQYd023419>