From patchwork Thu Mar 7 17:24:23 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Brian Foster X-Patchwork-Id: 10843489 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 873B617E9 for ; Thu, 7 Mar 2019 17:24:28 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 650C02D9B6 for ; Thu, 7 Mar 2019 17:24:28 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 597242DB37; Thu, 7 Mar 2019 17:24:28 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.9 required=2.0 tests=BAYES_00,MAILING_LIST_MULTI, RCVD_IN_DNSWL_HI autolearn=ham version=3.3.1 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 335A62D9B6 for ; Thu, 7 Mar 2019 17:24:27 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726234AbfCGRY0 (ORCPT ); Thu, 7 Mar 2019 12:24:26 -0500 Received: from mx1.redhat.com ([209.132.183.28]:57768 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726233AbfCGRY0 (ORCPT ); Thu, 7 Mar 2019 12:24:26 -0500 Received: from smtp.corp.redhat.com (int-mx03.intmail.prod.int.phx2.redhat.com [10.5.11.13]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id F2DF9C0587DD for ; Thu, 7 Mar 2019 17:24:25 +0000 (UTC) Received: from bfoster.bos.redhat.com (dhcp-41-66.bos.redhat.com [10.18.41.66]) by smtp.corp.redhat.com (Postfix) with ESMTP id 9DAE960C1B for ; Thu, 7 Mar 2019 17:24:25 +0000 (UTC) From: Brian Foster To: linux-xfs@vger.kernel.org Subject: [PATCH RFCv2 2/3] xfs: introduce generic extent allocation infrastructure Date: Thu, 7 Mar 2019 12:24:23 -0500 Message-Id: <20190307172424.30316-3-bfoster@redhat.com> In-Reply-To: <20190307172424.30316-1-bfoster@redhat.com> References: <20190307172424.30316-1-bfoster@redhat.com> X-Scanned-By: MIMEDefang 2.79 on 10.5.11.13 X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.5.16 (mx1.redhat.com [10.5.110.32]); Thu, 07 Mar 2019 17:24:26 +0000 (UTC) Sender: linux-xfs-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-xfs@vger.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP The extent allocation code in XFS has several allocation modes with unique implementations. This is slightly unfortunate as the allocation modes are not all that different from a high level perspective. The most involved mode is the near allocation mode which attempts to allocate an optimally sized extent with ideal locality with respect to a provided agbno. In the common case of suitable extent availability, a near mode allocation consists of a conditional scan of the last cntbt block followed by a concurrent left and right spanning search of the bnobt starting from the ideal point of locality in the bnobt. This works reasonably well as filesystems age via most common allocation patterns. If free space fragments as the filesystem ages, however, the near algorithm has very poor breakdown characteristics. If the extent size lookup happens to land outside of the last cntbt block, the alloc bypasses the cntbt entirely. If the target extent lies beyond a large enough number of unusable extents from the starting point(s) of the bnobt search, the bnobt search can take a significant amount of time to locate the target extent. This leads to pathological allocation latencies in certain workloads. The near allocation algorithm can be fundamentally improved to take advantage of a preexisting mechanism: that by-size cntbt record lookups can incorporate locality. This means that a single cntbt lookup can return the extent with ideal locality (with respect to an agbno param) of a particular size. This means that for larger extent allocations, repeated lookups of the cntbt with an agbno hint can be far more reliable and efficient than a brute force bnobt search. Such a cntbt search may not always find the extent with absolute best locality, but the tradeoff for good enough locality for a more efficient scan is worthwhile because more often than not, extent contiguity is more important for performance than locality. Introduce a new allocation mechanism based on the existing near allocation mode with the aforementioned tweaks. The new mechanism initiates concurrent bnobt spanning and cntbt lookup searches to naturally balance the optimal approach for smaller vs. larger allocation requests. For example, smaller allocation requests are highly likely to be satisfied by the bnobt search. The larger the allocation request, the more likely the cntbt lookup search locates the ideal extent. Once the cntbt search locates at least one suitable allocation candidate, the remaining search for better locality is boosted to throttle bnobt scans and the overall latency of the allocation. Implement the new mechanism with a generic interface and code factoring to facilitate reuse for other allocation modes. For example, the exact and size allocation modes are a subset of this implementation. It should be possible to perform such allocations in the future with fairly isolated logic changes to incorporate the different allocation requirements. Note that this patch introduces mechanism only and makes no functional changes. [XXX: Squash with follow on patch to address static function warnings, etc.]. Signed-off-by: Brian Foster --- fs/xfs/libxfs/xfs_alloc.c | 449 ++++++++++++++++++++++++++++++++++++++ fs/xfs/xfs_trace.h | 22 ++ 2 files changed, 471 insertions(+) diff --git a/fs/xfs/libxfs/xfs_alloc.c b/fs/xfs/libxfs/xfs_alloc.c index 10a6e22b764d..2f09d71ea909 100644 --- a/fs/xfs/libxfs/xfs_alloc.c +++ b/fs/xfs/libxfs/xfs_alloc.c @@ -981,6 +981,455 @@ xfs_alloc_find_best_extent( return error; } +/* + * NEAR allocation algorithm and data structures. + */ +struct xfs_best_cur { + struct xfs_btree_cur *cur; /* cursor */ + bool done; /* search done flag */ +}; + +struct xfs_alloc_best { + struct xfs_best_cur cnt; /* btree cursors */ + struct xfs_best_cur bnolt; + struct xfs_best_cur bnogt; + xfs_extlen_t cur_len; /* current search length */ + xfs_agblock_t rec_bno; /* extent startblock */ + xfs_extlen_t rec_len; /* extent length */ + xfs_agblock_t bno; /* alloc bno */ + xfs_extlen_t len; /* alloc len */ + xfs_extlen_t diff; /* diff from search bno */ + unsigned busy_gen; /* busy state */ + bool busy; +}; + +/* + * Set up the cursors, etc. in the best extent allocation tracking structure. + * This function can be called multiple times to reset an initialized structure + * without having to reallocate cursors. + */ +static int +xfs_alloc_best_setup( + struct xfs_alloc_arg *args, + struct xfs_alloc_best *best) +{ + int error; + int i; + + best->cnt.done = best->bnolt.done = best->bnogt.done = true; + best->cur_len = args->maxlen; + best->rec_bno = 0; + best->rec_len = 0; + best->bno = 0; + best->len = 0; + best->diff = -1; + best->busy = false; + best->busy_gen = 0; + + /* + * Initialize the cntbt cursor and determine whether to start the search + * at maxlen or minlen. + */ + if (!best->cnt.cur) + best->cnt.cur = xfs_allocbt_init_cursor(args->mp, args->tp, + args->agbp, args->agno, XFS_BTNUM_CNT); + error = xfs_alloc_lookup_ge(best->cnt.cur, args->agbno, best->cur_len, + &i); + if (!i) { + best->cur_len = args->minlen; + error = xfs_alloc_lookup_ge(best->cnt.cur, args->agbno, + best->cur_len, &i); + if (error) + return error; + } + if (i) + best->cnt.done = false; + + /* + * Initialize bnobt left/right search cursors. Mark the cursor done if + * either lands at the associated end of the tree. + */ + if (!best->bnolt.cur) + best->bnolt.cur = xfs_allocbt_init_cursor(args->mp, args->tp, + args->agbp, args->agno, XFS_BTNUM_BNO); + error = xfs_alloc_lookup_le(best->bnolt.cur, args->agbno, args->maxlen, + &i); + if (error) + return error; + if (i) + best->bnolt.done = false; + + if (!best->bnogt.cur) + best->bnogt.cur = xfs_allocbt_init_cursor(args->mp, args->tp, + args->agbp, args->agno, XFS_BTNUM_BNO); + error = xfs_alloc_lookup_ge(best->bnogt.cur, args->agbno, args->maxlen, + &i); + if (error) + return error; + if (i) + best->bnogt.done = false; + + return 0; +} + +static void +xfs_alloc_best_close( + struct xfs_alloc_best *best, + bool error) +{ + int cur_error = XFS_BTREE_NOERROR; + + if (error) + cur_error = XFS_BTREE_ERROR; + + if (best->cnt.cur) + xfs_btree_del_cursor(best->cnt.cur, cur_error); + if (best->bnolt.cur) + xfs_btree_del_cursor(best->bnolt.cur, cur_error); + if (best->bnogt.cur) + xfs_btree_del_cursor(best->bnogt.cur, cur_error); + best->cnt.cur = best->bnolt.cur = best->bnogt.cur = NULL; +} + +/* + * Consider an extent for allocation. If the provided extent fits allocation + * criteria and has better locality than the current candidate, store it in the + * best extent tracking structure. Mark the search cursor done if it has entered + * an out of range state based on allocation criteria. Returns true if the + * provided extent has been assigned as the new allocation candidate, false + * otherwise. + */ +static bool +xfs_alloc_best_check( + struct xfs_alloc_arg *args, + xfs_agblock_t bno, + xfs_extlen_t len, + struct xfs_alloc_best *best, + struct xfs_best_cur *bcur) +{ + xfs_agblock_t bnoa, bnew; + xfs_extlen_t lena, diff = -1; + bool busy; + unsigned busy_gen = 0; + bool badrange = false; + bool isbnobt = bcur->cur->bc_btnum == XFS_BTNUM_BNO; + + /* + * Check against minlen. If this is a cntbt cursor, it has gone out of + * range. This should only happen when walking backwards. + */ + if (len < args->minlen) { + badrange = !isbnobt; + goto fail; + } + + /* + * Compute aligned extent and check length and range. If this is a bnobt + * record that violates the range criteria, mark the bnobt cursor done. + */ + busy = xfs_alloc_compute_aligned(args, bno, len, &bnoa, &lena, + &busy_gen); + best->busy |= busy; + if (busy) + best->busy_gen = busy_gen; + if (bnoa < args->min_agbno || bnoa > args->max_agbno) { + badrange = isbnobt; + goto fail; + } + if (lena < args->minlen) + goto fail; + + args->len = XFS_EXTLEN_MIN(lena, args->maxlen); + xfs_alloc_fix_len(args); + ASSERT(args->len >= args->minlen); + if (args->len < best->len) + goto fail; + + /* + * If we have a usable extent, compare it to the current allocation + * candidate. Mark the cursor done if this is a bnobt cursor with worse + * locality. + */ + diff = xfs_alloc_compute_diff(args->agbno, args->len, + args->alignment, args->datatype, + bnoa, lena, &bnew); + if (bnew == NULLAGBLOCK) + goto fail; + if (diff > best->diff) { + badrange = isbnobt; + goto fail; + } + if (args->len < best->len) + goto fail; + + /* new candidate extent */ + best->rec_bno = bno; + best->rec_len = len; + best->bno = bnew; + best->len = args->len; + best->diff = diff; + trace_xfs_alloc_best_new(args->mp, best->bno, best->len, best->diff); + return true; + +fail: + if (badrange) + bcur->done = true; + return false; +} + +/* + * Perform an allocation of a candidate extent. Remove the extent from both + * trees and update the caller's allocation structure. + */ +STATIC int +xfs_alloc_best_finish( + struct xfs_alloc_arg *args, + struct xfs_alloc_best *best, + int *stat) +{ + int error; + + ASSERT(best->len); + ASSERT(best->cnt.cur && best->bnolt.cur); + + error = xfs_alloc_fixup_trees(best->cnt.cur, best->bnolt.cur, + best->rec_bno, best->rec_len, best->bno, + best->len, 0); + if (error) + return error; + + args->agbno = best->bno; + args->len = best->len; + args->wasfromfl = 0; + *stat = 1; + + trace_xfs_alloc_best(args); + return 0; +} + +/* + * Perform an iterative by-size lookup based on the allocation request. This + * generally expects a cntbt cursor and uses the bno optimized lookup to find a + * suitably sized extent with ideal locality. + */ +STATIC int +xfs_alloc_lookup_iter( + struct xfs_alloc_arg *args, + struct xfs_alloc_best *best, + struct xfs_best_cur *bcur) +{ + struct xfs_btree_cur *cur = bcur->cur; + xfs_agblock_t bno; + xfs_extlen_t len, cur_len; + int error; + int i; + + /* + * Store the current search key and perform a locality optimized lookup. + */ + cur_len = best->cur_len; + error = xfs_alloc_lookup_ge(cur, args->agbno, cur_len, &i); + if (error) + return error; + if (i == 0) { + bcur->done = true; + return 0; + } + error = xfs_alloc_get_rec(cur, &bno, &len, &i); + if (error) + return error; + + /* + * Check the record and update the search key to the extent length we + * actually found in the tree. + */ + XFS_WANT_CORRUPTED_RETURN(args->mp, i == 1); + xfs_alloc_best_check(args, bno, len, best, bcur); + ASSERT(len >= best->cur_len); + best->cur_len = len; + + /* + * We looked up the first record >= [agbno, len] above. If this is a + * cntbt search, the agbno is a secondary key and so the record may not + * start beyond agbno if there are no such records for the particular + * length. If it is past agbno, check the previous record too so long as + * the length matches as it may be closer. + */ + if (bno > args->agbno) { + error = xfs_btree_decrement(cur, 0, &i); + if (error) + return error; + if (i) { + error = xfs_alloc_get_rec(cur, &bno, &len, &i); + if (error) + return error; + XFS_WANT_CORRUPTED_RETURN(args->mp, i == 1); + if (len == best->cur_len) + xfs_alloc_best_check(args, bno, len, best, + bcur); + } + } + + /* + * Increment the search key if we haven't yet found a candidate extent + * or this lookup already significantly bumped the key. We have to make + * sure to not skip any records until we have at least one allocatable + * extent. Note that if the allocation ultimately fails due to busy + * extents, we'll flush the busy list and restart the whole thing. + * + * Otherwise, double the search key for the next lookup to optimize the + * search. This allows us to find good enough locality at the expense of + * absolute best locality. Max extent size and reasonable allocation + * efficiency are more important here than perfect locality. + */ + cur_len <<= 1; + if (!best->len || best->cur_len >= cur_len) + best->cur_len++; + else + best->cur_len = cur_len; + + return error; +} + +/* + * Perform an iterative record at a time walk of a btree to find an allocation + * candidate extent. This is generally used for left/right spanning searches of + * the bnobt, but this also works on a cntbt cursor for cases where a minimal + * number of suitably sized extents are available and a more aggressive search + * is required. + */ +STATIC int +xfs_alloc_walk_iter( + struct xfs_alloc_arg *args, + struct xfs_alloc_best *best, + struct xfs_best_cur *bcur, + bool increment, + int iters, + int *stat) +{ + struct xfs_btree_cur *cur; + int error; + int i; + xfs_agblock_t bno; + xfs_extlen_t len; + bool found = false; + + if (bcur->done) + return 0; + + *stat = 0; + cur = bcur->cur; + for (; iters > 0; iters--) { + error = xfs_alloc_get_rec(cur, &bno, &len, &i); + if (error) + return error; + XFS_WANT_CORRUPTED_RETURN(args->mp, i == 1); + found = xfs_alloc_best_check(args, bno, len, best, bcur); + if (found) { + *stat = 1; + break; + } + if (bcur->done) + break; + + if (increment) + error = xfs_btree_increment(cur, 0, &i); + else + error = xfs_btree_decrement(cur, 0, &i); + if (error) + return error; + if (i == 0) { + bcur->done = true; + break; + } + } + + return error; +} + +STATIC int +xfs_alloc_ag_vextent_best( + struct xfs_alloc_arg *args, + struct xfs_alloc_best *best, + int *stat) +{ + int error; + int i; + unsigned int findbestcount = args->mp->m_alloc_mxr[0]; + + ASSERT(best->cnt.cur); + *stat = 0; + + /* search as long as we have at least one active cursor */ + while (!best->cnt.done || !best->bnolt.done || !best->bnogt.done) { + /* + * Search the bnobt left and right. If either of these find a + * suitable extent, we know we've found ideal locality. Do a + * capped search in the opposite direction and we're done. + */ + error = xfs_alloc_walk_iter(args, best, &best->bnolt, false, 1, + &i); + if (error) + return error; + if (i) { + error = xfs_alloc_walk_iter(args, best, &best->bnogt, + true, findbestcount, &i); + if (error) + return error; + break; + } + + error = xfs_alloc_walk_iter(args, best, &best->bnogt, true, 1, + &i); + if (error) + return error; + if (i) { + error = xfs_alloc_walk_iter(args, best, &best->bnolt, + false, findbestcount, &i); + if (error) + return error; + break; + } + + /* + * Search the cntbt for a maximum sized extent with ideal + * locality. The lookup search terminates on reaching the end of + * the cntbt. Break out of the loop if this occurs to throttle + * the bnobt scans. + */ + error = xfs_alloc_lookup_iter(args, best, &best->cnt); + if (error) + return error; + if (best->cnt.done) + break; + } + + /* + * If the lookup search terminated and we still have no suitable + * allocation, scan the largest extents available in the cntbt as a last + * resort. The cntbt done flag means the cursor points off the end of + * the tree. Point it back to the last record in the tree and walk + * backwards from there. + */ + if (!best->len && best->cnt.done) { + best->cnt.done = false; + error = xfs_btree_decrement(best->cnt.cur, 0, &i); + if (error) + return error; + if (i) { + error = xfs_alloc_walk_iter(args, best, &best->cnt, + false, findbestcount, &i); + if (error) + return error; + } + } + + if (best->len) + error = xfs_alloc_best_finish(args, best, stat); + + return error; +} + /* * Allocate a variable extent near bno in the allocation group agno. * Extent's length (returned in len) will be between minlen and maxlen, diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h index 47fb07d86efd..7346646405cd 100644 --- a/fs/xfs/xfs_trace.h +++ b/fs/xfs/xfs_trace.h @@ -1642,6 +1642,7 @@ DEFINE_ALLOC_EVENT(xfs_alloc_near_lesser); DEFINE_ALLOC_EVENT(xfs_alloc_near_error); DEFINE_ALLOC_EVENT(xfs_alloc_near_noentry); DEFINE_ALLOC_EVENT(xfs_alloc_near_busy); +DEFINE_ALLOC_EVENT(xfs_alloc_best); DEFINE_ALLOC_EVENT(xfs_alloc_size_neither); DEFINE_ALLOC_EVENT(xfs_alloc_size_noentry); DEFINE_ALLOC_EVENT(xfs_alloc_size_nominleft); @@ -1658,6 +1659,27 @@ DEFINE_ALLOC_EVENT(xfs_alloc_vextent_noagbp); DEFINE_ALLOC_EVENT(xfs_alloc_vextent_loopfailed); DEFINE_ALLOC_EVENT(xfs_alloc_vextent_allfailed); +TRACE_EVENT(xfs_alloc_best_new, + TP_PROTO(struct xfs_mount *mp, xfs_agblock_t bno, xfs_extlen_t len, + xfs_extlen_t diff), + TP_ARGS(mp, bno, len, diff), + TP_STRUCT__entry( + __field(dev_t, dev) + __field(xfs_agblock_t, bno) + __field(xfs_extlen_t, len) + __field(xfs_extlen_t, diff) + ), + TP_fast_assign( + __entry->dev = mp->m_super->s_dev; + __entry->bno = bno; + __entry->len = len; + __entry->diff = diff; + ), + TP_printk("dev %d:%d bno 0x%x len 0x%x diff 0x%x", + MAJOR(__entry->dev), MINOR(__entry->dev), + __entry->bno, __entry->len, __entry->diff) +) + DECLARE_EVENT_CLASS(xfs_da_class, TP_PROTO(struct xfs_da_args *args), TP_ARGS(args),