Date: Mon, 14 Oct 2013 21:50:57 +0000 (UTC) From: Alan Somers <asomers@FreeBSD.org> To: src-committers@freebsd.org, svn-src-projects@freebsd.org Subject: svn commit: r256463 - projects/zfsd/head/cddl/sbin/zfsd Message-ID: <201310142150.r9ELovC1082211@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: asomers Date: Mon Oct 14 21:50:56 2013 New Revision: 256463 URL: http://svnweb.freebsd.org/changeset/base/256463 Log: Fix a bug in zfsd: when a drive is experiencing a rapid storm of IO or checksum errors, zfsd will not degrade/fault it until hundreds or thousands of errors have occured. cddl/sbin/zfsd/case_file.cc RefreshVdevState() iterates through all the system's zpools, which involves the ioctls ZFS_IOC_POOL_CONFIGS and ZFS_IOC_POOL_STATS. Both of those acquire spa_namespace_lock, which may block for a long time under certain circumstances, including when the system has a storm of IO or checksum errors. This change eliminates the call to RefreshVdevState() whenever a ZFSEvent is received. Instead, RefreshVdevState() will only be called when a CaseFile is closed, if necessary. This way, zfsd won't spend too much time blocking on ioctl()s and miss reading events from devd. Submitted by: alans Approved by: ken (mentor) Sponsored by: Spectra Logic Corporation Modified: projects/zfsd/head/cddl/sbin/zfsd/case_file.cc Modified: projects/zfsd/head/cddl/sbin/zfsd/case_file.cc ============================================================================== --- projects/zfsd/head/cddl/sbin/zfsd/case_file.cc Mon Oct 14 21:41:36 2013 (r256462) +++ projects/zfsd/head/cddl/sbin/zfsd/case_file.cc Mon Oct 14 21:50:56 2013 (r256463) @@ -298,28 +298,6 @@ CaseFile::ReEvaluate(const ZfsEvent &eve { bool consumed(false); - if (!RefreshVdevState()) { - /* - * The pool or vdev for this case file is no longer - * part of the configuration. This can happen - * if we process a device arrival notification - * before seeing the ZFS configuration change - * event. - */ - syslog(LOG_INFO, - "CaseFile::ReEvaluate(%s,%s) Pool/Vdev unconfigured. " - "Closing\n", - PoolGUIDString().c_str(), - VdevGUIDString().c_str()); - Close(); - - /* - * Since this event was not used to close this - * case, do not report it as consumed. - */ - return (/*consumed*/false); - } - if (event.Value("type") == "misc.fs.zfs.vdev_remove") { /* * The Vdev we represent has been removed from the @@ -333,6 +311,28 @@ CaseFile::ReEvaluate(const ZfsEvent &eve if (event.Value("class") == "resource.fs.zfs.removed") { bool spare_activated; + if (!RefreshVdevState()) { + /* + * The pool or vdev for this case file is no longer + * part of the configuration. This can happen + * if we process a device arrival notification + * before seeing the ZFS configuration change + * event. + */ + syslog(LOG_INFO, + "CaseFile::ReEvaluate(%s,%s) Pool/Vdev " + "unconfigured. Closing\n", + PoolGUIDString().c_str(), + VdevGUIDString().c_str()); + Close(); + + /* + * Since this event was not used to close this + * case, do not report it as consumed. + */ + return (/*consumed*/false); + } + /* * Discard any tentative I/O error events for * this case. They were most likely caused by the
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201310142150.r9ELovC1082211>