From nobody Tue Oct 10 19:35:16 2023 X-Original-To: dev-commits-src-all@mlmmj.nyi.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2610:1c1:1:606c::19:1]) by mlmmj.nyi.freebsd.org (Postfix) with ESMTP id 4S4mMS70H3z4wDbV; Tue, 10 Oct 2023 19:35:16 +0000 (UTC) (envelope-from git@FreeBSD.org) Received: from mxrelay.nyi.freebsd.org (mxrelay.nyi.freebsd.org [IPv6:2610:1c1:1:606c::19:3]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (4096 bits) server-digest SHA256 client-signature RSA-PSS (4096 bits) client-digest SHA256) (Client CN "mxrelay.nyi.freebsd.org", Issuer "R3" (verified OK)) by mx1.freebsd.org (Postfix) with ESMTPS id 4S4mMS6QxJz3RTj; Tue, 10 Oct 2023 19:35:16 +0000 (UTC) (envelope-from git@FreeBSD.org) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=freebsd.org; s=dkim; t=1696966516; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding; bh=FBwMXuAorpXFpwtX9tnPlO0JcfeaEwuFFiIE/h06RXM=; b=MucVceyV0Bt+9QfIzmAYoP4jVOk55nImxBrVuW/i+OeRkqDuO7Cb0vtPDHlws7lIkeEVBS l0WphKPACj6WecDZTOdbrN1nHhfONLZBzDXpbo+br3l2kSkYpgmNUqRkcAmnv+807+BHQF rgK0qTUXiqW60QC9lifyt2CmVc73Kt+FRnj8W06RE0eyAYfLrHK4HwnNvJ8SR8o28895f1 9xMfL73xBAAQH0dUgPhnvQZ4FK2iygj3FRxuQTGyCsXLl93gUETKSNNApFsnJsQJOFiYnq fKFPIqGc/QGIFgAx7EftCJmgrH1YGQkG/PX0RfLrfTYddrIvFGWwBq85OfC02Q== ARC-Seal: i=1; s=dkim; d=freebsd.org; t=1696966516; a=rsa-sha256; cv=none; b=AkuIleEg3oBiwoZKesjKgXbH/5nqrnBGMq/KksxhsaU7JCIUsEwExFHKfoniXS0A3YYYDI dlqAAT4yGmC7mJWzdkQnZSvyWItro99QTDtkclPVh29bKdtW+dctI/vhEv2n0YdqIt19f5 movkDXh/4bNxov2QvAh/UmvLMlLKHzYtvoJOr/RHnEk9csHyUrIeGOlwjsi099ZIn/E8q1 F9uBj+IX17LReiKFyIr+CicCb1fUtAbLctCTjyU3YZeucdDCfU8lkMxCd/oHTERnu2Ygmt nExqBdhWc6XELlXtpi0bMS5Msde5SA6KOwxF0lddh0qph25OSBgolDofndlb+Q== ARC-Authentication-Results: i=1; mx1.freebsd.org; none ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=freebsd.org; s=dkim; t=1696966516; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding; bh=FBwMXuAorpXFpwtX9tnPlO0JcfeaEwuFFiIE/h06RXM=; b=oqjb7nUPIJJzeDOHBL5ZkUXWCX5Iei95JS0Ty8fKPDdumSnJ5krBLKsKgmS9XpL18YtWoD AkhkGNTnAMDORwh3Tbly/baLbwqq1z1FUqoyX2d/gOXAA/2Ad8dsKEKFAC7AVws+fXRokc 16A32XtNMvqaJwrqhfxQXwP18amLDg7rCdn8OIcDMYG2QNmhaKomGYAnf+VEhybA/gz05T HSQrLaGcGhtxkbWvbw/67GDYYVnQboQ0ed9QU2B3zDw7e+XUZ8IXr3/+LO1Mw1E2x/r9XI xG9HTbbvMCFLeeIYD/OATvazSmVEvWIFZ3u4d+xj070i7XwAS1cQTd8hXZzklQ== Received: from gitrepo.freebsd.org (gitrepo.freebsd.org [IPv6:2610:1c1:1:6068::e6a:5]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (4096 bits) server-digest SHA256) (Client did not present a certificate) by mxrelay.nyi.freebsd.org (Postfix) with ESMTPS id 4S4mMS5TG0zrZB; Tue, 10 Oct 2023 19:35:16 +0000 (UTC) (envelope-from git@FreeBSD.org) Received: from gitrepo.freebsd.org ([127.0.1.44]) by gitrepo.freebsd.org (8.17.1/8.17.1) with ESMTP id 39AJZGsd003685; Tue, 10 Oct 2023 19:35:16 GMT (envelope-from git@gitrepo.freebsd.org) Received: (from git@localhost) by gitrepo.freebsd.org (8.17.1/8.17.1/Submit) id 39AJZGmZ003682; Tue, 10 Oct 2023 19:35:16 GMT (envelope-from git) Date: Tue, 10 Oct 2023 19:35:16 GMT Message-Id: <202310101935.39AJZGmZ003682@gitrepo.freebsd.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-main@FreeBSD.org From: Mateusz Guzik Subject: git: 90a008e94bb2 - main - vfs: prefix regular vnlru with a special case for free vnodes List-Id: Commit messages for all branches of the src repository List-Archive: https://lists.freebsd.org/archives/dev-commits-src-all List-Help: List-Post: List-Subscribe: List-Unsubscribe: Sender: owner-dev-commits-src-all@freebsd.org X-BeenThere: dev-commits-src-all@freebsd.org MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 8bit X-Git-Committer: mjg X-Git-Repository: src X-Git-Refname: refs/heads/main X-Git-Reftype: branch X-Git-Commit: 90a008e94bb205e5b8f3c41d57e155b59a6be95d Auto-Submitted: auto-generated The branch main has been updated by mjg: URL: https://cgit.FreeBSD.org/src/commit/?id=90a008e94bb205e5b8f3c41d57e155b59a6be95d commit 90a008e94bb205e5b8f3c41d57e155b59a6be95d Author: Mateusz Guzik AuthorDate: 2023-09-14 19:08:40 +0000 Commit: Mateusz Guzik CommitDate: 2023-10-10 19:35:12 +0000 vfs: prefix regular vnlru with a special case for free vnodes Works around severe performance problems in certain corner cases, see the commentary added. Modifying vnlru logic has proven rather error prone in the past and a release is near, thus take the easy way out and fix it without having to dig into the current machinery. --- sys/kern/vfs_subr.c | 117 ++++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 113 insertions(+), 4 deletions(-) diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c index 5e39a149ef36..5834feff080c 100644 --- a/sys/kern/vfs_subr.c +++ b/sys/kern/vfs_subr.c @@ -1427,6 +1427,14 @@ vnlru_free_locked(int count) return (ret); } +static int +vnlru_free(int count) +{ + + mtx_lock(&vnode_list_mtx); + return (vnlru_free_locked(count)); +} + void vnlru_free_vfsops(int count, struct vfsops *mnt_op, struct vnode *mvp) { @@ -1594,6 +1602,106 @@ vnlru_kick_cond(void) mtx_unlock(&vnode_list_mtx); } +static void +vnlru_proc_sleep(void) +{ + + if (vnlruproc_sig) { + vnlruproc_sig = 0; + wakeup(&vnlruproc_sig); + } + msleep(vnlruproc, &vnode_list_mtx, PVFS|PDROP, "vlruwt", hz); +} + +/* + * A lighter version of the machinery below. + * + * Tries to reach goals only by recycling free vnodes and does not invoke + * uma_reclaim(UMA_RECLAIM_DRAIN). + * + * This works around pathological behavior in vnlru in presence of tons of free + * vnodes, but without having to rewrite the machinery at this time. Said + * behavior boils down to continuously trying to reclaim all kinds of vnodes + * (cycling through all levels of "force") when the count is transiently above + * limit. This happens a lot when all vnodes are used up and vn_alloc + * speculatively increments the counter. + * + * Sample testcase: vnode limit 8388608, 20 separate directory trees each with + * 1 million files in total and 20 find(1) processes stating them in parallel + * (one per each tree). + * + * On a kernel with only stock machinery this needs anywhere between 60 and 120 + * seconds to execute (time varies *wildly* between runs). With the workaround + * it consistently stays around 20 seconds. + * + * That is to say the entire thing needs a fundamental redesign (most notably + * to accommodate faster recycling), the above only tries to get it ouf the way. + * + * Return values are: + * -1 -- fallback to regular vnlru loop + * 0 -- do nothing, go to sleep + * >0 -- recycle this many vnodes + */ +static long +vnlru_proc_light_pick(void) +{ + u_long rnumvnodes, rfreevnodes; + + if (vstir || vnlruproc_sig == 1) + return (-1); + + rnumvnodes = atomic_load_long(&numvnodes); + rfreevnodes = vnlru_read_freevnodes(); + + /* + * vnode limit might have changed and now we may be at a significant + * excess. Bail if we can't sort it out with free vnodes. + */ + if (rnumvnodes > desiredvnodes) { + if (rnumvnodes - rfreevnodes >= desiredvnodes || + rfreevnodes <= wantfreevnodes) { + return (-1); + } + + return (rnumvnodes - desiredvnodes); + } + + /* + * Don't try to reach wantfreevnodes target if there are too few vnodes + * to begin with. + */ + if (rnumvnodes < wantfreevnodes) { + return (0); + } + + if (rfreevnodes < wantfreevnodes) { + return (-1); + } + + return (0); +} + +static bool +vnlru_proc_light(void) +{ + long freecount; + + mtx_assert(&vnode_list_mtx, MA_NOTOWNED); + + freecount = vnlru_proc_light_pick(); + if (freecount == -1) + return (false); + + if (freecount != 0) { + vnlru_free(freecount); + } + + mtx_lock(&vnode_list_mtx); + vnlru_proc_sleep(); + mtx_assert(&vnode_list_mtx, MA_NOTOWNED); + return (true); +} + static void vnlru_proc(void) { @@ -1609,6 +1717,10 @@ vnlru_proc(void) want_reread = false; for (;;) { kproc_suspend_check(vnlruproc); + + if (force == 0 && vnlru_proc_light()) + continue; + mtx_lock(&vnode_list_mtx); rnumvnodes = atomic_load_long(&numvnodes); @@ -1639,10 +1751,7 @@ vnlru_proc(void) vstir = false; } if (force == 0 && !vnlru_under(rnumvnodes, vlowat)) { - vnlruproc_sig = 0; - wakeup(&vnlruproc_sig); - msleep(vnlruproc, &vnode_list_mtx, - PVFS|PDROP, "vlruwt", hz); + vnlru_proc_sleep(); continue; } rfreevnodes = vnlru_read_freevnodes();