From patchwork Tue Oct 29 23:30:09 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Darrick J. Wong" X-Patchwork-Id: 11218775 Return-Path: Received: from mail.kernel.org (pdx-korg-mail-1.web.codeaurora.org [172.30.200.123]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 3892E1390 for ; Tue, 29 Oct 2019 23:30:16 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 0BE9821924 for ; Tue, 29 Oct 2019 23:30:16 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=pass (2048-bit key) header.d=oracle.com header.i=@oracle.com header.b="QoTCYzZM" Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726157AbfJ2XaP (ORCPT ); Tue, 29 Oct 2019 19:30:15 -0400 Received: from userp2120.oracle.com ([156.151.31.85]:48226 "EHLO userp2120.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726028AbfJ2XaP (ORCPT ); Tue, 29 Oct 2019 19:30:15 -0400 Received: from pps.filterd (userp2120.oracle.com [127.0.0.1]) by userp2120.oracle.com (8.16.0.27/8.16.0.27) with SMTP id x9TNU30t011301 for ; Tue, 29 Oct 2019 23:30:13 GMT DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=oracle.com; h=subject : from : to : cc : date : message-id : in-reply-to : references : mime-version : content-type : content-transfer-encoding; s=corp-2019-08-05; bh=c9Qlx/2N3w0zUDdDyJJA5w9RN5KMAFmaQy7dlZKUxAU=; b=QoTCYzZMswgYO790xhpI4Ffp6HyM6N8usZ/5Yd9UGokazU0t3bxwrikoMypcG8BQphI2 q9m/Dbjg5XzhKWLFDeprJuCEGaKqrP2Ww/qOFh8NB7hdz1OTWYWQXjeafEkkBYGLOKZw Avg/lYzND2nMbljGAk6sjGW6IfKBE6DyDNxjExXOYmRq7KBetAIPtmvAr7ItHSkwqf9k ZNfZ4h1LyDERbuybZ0A4wOVn6BPdVQJCDZut3bskd3dgXeSwskP1oxb56imBx5G1opyx xJj9FbciymtvdeGqFE4QtZ/yqEvFdFD0bJjvHraLdEy6gHny137sugntRHwZMKlaaiAY fQ== Received: from aserp3030.oracle.com (aserp3030.oracle.com [141.146.126.71]) by userp2120.oracle.com with ESMTP id 2vxwhfgav4-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Tue, 29 Oct 2019 23:30:13 +0000 Received: from pps.filterd (aserp3030.oracle.com [127.0.0.1]) by aserp3030.oracle.com (8.16.0.27/8.16.0.27) with SMTP id x9TNSEjR039052 for ; Tue, 29 Oct 2019 23:30:12 GMT Received: from userv0121.oracle.com (userv0121.oracle.com [156.151.31.72]) by aserp3030.oracle.com with ESMTP id 2vxwj8uy35-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Tue, 29 Oct 2019 23:30:12 +0000 Received: from abhmp0021.oracle.com (abhmp0021.oracle.com [141.146.116.27]) by userv0121.oracle.com (8.14.4/8.13.8) with ESMTP id x9TNUAMp006392 for ; Tue, 29 Oct 2019 23:30:11 GMT Received: from localhost (/67.169.218.210) by default (Oracle Beehive Gateway v4.0) with ESMTP ; Tue, 29 Oct 2019 16:30:10 -0700 Subject: [PATCH 1/4] xfs: introduce fake roots for ag-rooted btrees From: "Darrick J. Wong" To: darrick.wong@oracle.com Cc: linux-xfs@vger.kernel.org Date: Tue, 29 Oct 2019 16:30:09 -0700 Message-ID: <157239180957.1266747.9609472431653271537.stgit@magnolia> In-Reply-To: <157239180345.1266747.1586648446090389492.stgit@magnolia> References: <157239180345.1266747.1586648446090389492.stgit@magnolia> User-Agent: StGit/0.17.1-dirty MIME-Version: 1.0 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9425 signatures=668685 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 suspectscore=3 malwarescore=0 phishscore=0 bulkscore=0 spamscore=0 mlxscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1908290000 definitions=main-1910290206 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9425 signatures=668685 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 priorityscore=1501 malwarescore=0 suspectscore=3 phishscore=0 bulkscore=0 spamscore=0 clxscore=1015 lowpriorityscore=0 mlxscore=0 impostorscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1908290000 definitions=main-1910290206 Sender: linux-xfs-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-xfs@vger.kernel.org From: Darrick J. Wong Create an in-core fake root for AG-rooted btree types so that callers can generate a whole new btree using the upcoming btree bulk load function without making the new tree accessible from the rest of the filesystem. It is up to the individual btree type to provide a function to create a staged cursor (presumably with the appropriate callouts to update the fakeroot) and then commit the staged root back into the filesystem. Signed-off-by: Darrick J. Wong --- fs/xfs/libxfs/xfs_btree.c | 117 +++++++++++++++++++++++++++++++++++++++++++++ fs/xfs/libxfs/xfs_btree.h | 42 ++++++++++++++-- fs/xfs/xfs_trace.h | 28 +++++++++++ 3 files changed, 182 insertions(+), 5 deletions(-) diff --git a/fs/xfs/libxfs/xfs_btree.c b/fs/xfs/libxfs/xfs_btree.c index 4fd89c80c821..ab0d459c10ea 100644 --- a/fs/xfs/libxfs/xfs_btree.c +++ b/fs/xfs/libxfs/xfs_btree.c @@ -384,6 +384,8 @@ xfs_btree_del_cursor( /* * Free the cursor. */ + if (unlikely(cur->bc_flags & XFS_BTREE_STAGING)) + kmem_free((void *)cur->bc_ops); kmem_zone_free(xfs_btree_cur_zone, cur); } @@ -4936,3 +4938,118 @@ xfs_btree_has_more_records( else return block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK); } + +/* We don't allow staging cursors to be duplicated. */ +STATIC struct xfs_btree_cur * +xfs_btree_fakeroot_dup_cursor( + struct xfs_btree_cur *cur) +{ + ASSERT(0); + return NULL; +} + +/* Refuse to allow regular block allocation for a staging cursor. */ +STATIC int +xfs_btree_fakeroot_alloc_block( + struct xfs_btree_cur *cur, + union xfs_btree_ptr *start_bno, + union xfs_btree_ptr *new_bno, + int *stat) +{ + ASSERT(0); + return -EFSCORRUPTED; +} + +/* Refuse to allow block freeing for a staging cursor. */ +STATIC int +xfs_btree_fakeroot_free_block( + struct xfs_btree_cur *cur, + struct xfs_buf *bp) +{ + ASSERT(0); + return -EFSCORRUPTED; +} + +/* Initialize a pointer to the root block from the fakeroot. */ +STATIC void +xfs_btree_fakeroot_init_ptr_from_cur( + struct xfs_btree_cur *cur, + union xfs_btree_ptr *ptr) +{ + struct xbtree_afakeroot *afake; + + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + + afake = cur->bc_private.a.afake; + ptr->s = cpu_to_be32(afake->af_root); +} + +/* Set the root block when our tree has a fakeroot. */ +STATIC void +xfs_btree_afakeroot_set_root( + struct xfs_btree_cur *cur, + union xfs_btree_ptr *ptr, + int inc) +{ + struct xbtree_afakeroot *afake = cur->bc_private.a.afake; + + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + afake->af_root = be32_to_cpu(ptr->s); + afake->af_levels += inc; +} + +/* + * Initialize a AG-rooted btree cursor with the given AG btree fake root. The + * btree cursor's @bc_ops will be overridden as needed to make the staging + * functionality work. If @new_ops is not NULL, these new ops will be passed + * out to the caller for further overriding. + */ +void +xfs_btree_stage_afakeroot( + struct xfs_btree_cur *cur, + struct xbtree_afakeroot *afake, + struct xfs_btree_ops **new_ops) +{ + struct xfs_btree_ops *nops; + + ASSERT(!(cur->bc_flags & XFS_BTREE_STAGING)); + ASSERT(!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)); + + nops = kmem_alloc(sizeof(struct xfs_btree_ops), KM_NOFS); + memcpy(nops, cur->bc_ops, sizeof(struct xfs_btree_ops)); + nops->alloc_block = xfs_btree_fakeroot_alloc_block; + nops->free_block = xfs_btree_fakeroot_free_block; + nops->init_ptr_from_cur = xfs_btree_fakeroot_init_ptr_from_cur; + nops->set_root = xfs_btree_afakeroot_set_root; + nops->dup_cursor = xfs_btree_fakeroot_dup_cursor; + + cur->bc_private.a.afake = afake; + cur->bc_nlevels = afake->af_levels; + cur->bc_ops = nops; + cur->bc_flags |= XFS_BTREE_STAGING; + + if (new_ops) + *new_ops = nops; +} + +/* + * Transform an AG-rooted staging btree cursor back into a regular cursor by + * substituting a real btree root for the fake one and restoring normal btree + * cursor ops. The caller must log the btree root change prior to calling + * this. + */ +void +xfs_btree_commit_afakeroot( + struct xfs_btree_cur *cur, + struct xfs_buf *agbp, + const struct xfs_btree_ops *ops) +{ + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + + trace_xfs_btree_commit_afakeroot(cur); + + kmem_free((void *)cur->bc_ops); + cur->bc_private.a.agbp = agbp; + cur->bc_ops = ops; + cur->bc_flags &= ~XFS_BTREE_STAGING; +} diff --git a/fs/xfs/libxfs/xfs_btree.h b/fs/xfs/libxfs/xfs_btree.h index bac15668aeee..39ca3f0a3bf8 100644 --- a/fs/xfs/libxfs/xfs_btree.h +++ b/fs/xfs/libxfs/xfs_btree.h @@ -185,6 +185,16 @@ union xfs_btree_cur_private { } refc; }; +/* Private information for a AG-rooted btree. */ +struct xfs_btree_priv_ag { /* needed for BNO, CNT, INO */ + union { + struct xfs_buf *agbp; /* agf/agi buffer pointer */ + struct xbtree_afakeroot *afake; /* fake ag header root */ + }; + xfs_agnumber_t agno; /* ag number */ + union xfs_btree_cur_private priv; +}; + /* * Btree cursor structure. * This collects all information needed by the btree code in one place. @@ -206,11 +216,7 @@ typedef struct xfs_btree_cur xfs_btnum_t bc_btnum; /* identifies which btree type */ int bc_statoff; /* offset of btre stats array */ union { - struct { /* needed for BNO, CNT, INO */ - struct xfs_buf *agbp; /* agf/agi buffer pointer */ - xfs_agnumber_t agno; /* ag number */ - union xfs_btree_cur_private priv; - } a; + struct xfs_btree_priv_ag a; struct { /* needed for BMAP */ struct xfs_inode *ip; /* pointer to our inode */ int allocated; /* count of alloced */ @@ -229,6 +235,12 @@ typedef struct xfs_btree_cur #define XFS_BTREE_LASTREC_UPDATE (1<<2) /* track last rec externally */ #define XFS_BTREE_CRC_BLOCKS (1<<3) /* uses extended btree blocks */ #define XFS_BTREE_OVERLAPPING (1<<4) /* overlapping intervals */ +/* + * The root of this btree is a fakeroot structure so that we can stage a btree + * rebuild without leaving it accessible via primary metadata. The ops struct + * is dynamically allocated and must be freed when the cursor is deleted. + */ +#define XFS_BTREE_STAGING (1<<5) #define XFS_BTREE_NOERROR 0 @@ -521,4 +533,24 @@ int xfs_btree_has_record(struct xfs_btree_cur *cur, union xfs_btree_irec *low, union xfs_btree_irec *high, bool *exists); bool xfs_btree_has_more_records(struct xfs_btree_cur *cur); +/* Fake root for an AG-rooted btree. */ +struct xbtree_afakeroot { + /* AG block number of the new btree root. */ + xfs_agblock_t af_root; + + /* Height of the new btree. */ + unsigned int af_levels; + + /* Number of blocks used by the btree. */ + unsigned int af_blocks; +}; + +/* Cursor interactions with with fake roots for AG-rooted btrees. */ +void xfs_btree_stage_afakeroot(struct xfs_btree_cur *cur, + struct xbtree_afakeroot *afake, + struct xfs_btree_ops **new_ops); +void xfs_btree_commit_afakeroot(struct xfs_btree_cur *cur, + struct xfs_buf *agbp, + const struct xfs_btree_ops *ops); + #endif /* __XFS_BTREE_H__ */ diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h index eaae275ed430..e04ed5324696 100644 --- a/fs/xfs/xfs_trace.h +++ b/fs/xfs/xfs_trace.h @@ -3609,6 +3609,34 @@ DEFINE_KMEM_EVENT(kmem_alloc_large); DEFINE_KMEM_EVENT(kmem_realloc); DEFINE_KMEM_EVENT(kmem_zone_alloc); +TRACE_EVENT(xfs_btree_commit_afakeroot, + TP_PROTO(struct xfs_btree_cur *cur), + TP_ARGS(cur), + TP_STRUCT__entry( + __field(dev_t, dev) + __field(xfs_btnum_t, btnum) + __field(xfs_agnumber_t, agno) + __field(xfs_agblock_t, agbno) + __field(unsigned int, levels) + __field(unsigned int, blocks) + ), + TP_fast_assign( + __entry->dev = cur->bc_mp->m_super->s_dev; + __entry->btnum = cur->bc_btnum; + __entry->agno = cur->bc_private.a.agno; + __entry->agbno = cur->bc_private.a.afake->af_root; + __entry->levels = cur->bc_private.a.afake->af_levels; + __entry->blocks = cur->bc_private.a.afake->af_blocks; + ), + TP_printk("dev %d:%d btree %s ag %u levels %u blocks %u root %u", + MAJOR(__entry->dev), MINOR(__entry->dev), + __print_symbolic(__entry->btnum, XFS_BTNUM_STRINGS), + __entry->agno, + __entry->levels, + __entry->blocks, + __entry->agbno) +) + #endif /* _TRACE_XFS_H */ #undef TRACE_INCLUDE_PATH From patchwork Tue Oct 29 23:30:15 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Darrick J. Wong" X-Patchwork-Id: 11218811 Return-Path: Received: from mail.kernel.org (pdx-korg-mail-1.web.codeaurora.org [172.30.200.123]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id E7A241390 for ; Tue, 29 Oct 2019 23:32:20 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id BB6F92083E for ; Tue, 29 Oct 2019 23:32:20 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=pass (2048-bit key) header.d=oracle.com header.i=@oracle.com header.b="dFFW0nX6" Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726028AbfJ2XcU (ORCPT ); Tue, 29 Oct 2019 19:32:20 -0400 Received: from userp2120.oracle.com ([156.151.31.85]:49890 "EHLO userp2120.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726084AbfJ2XcU (ORCPT ); Tue, 29 Oct 2019 19:32:20 -0400 Received: from pps.filterd (userp2120.oracle.com [127.0.0.1]) by userp2120.oracle.com (8.16.0.27/8.16.0.27) with SMTP id x9TNSjru010477 for ; Tue, 29 Oct 2019 23:32:18 GMT DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=oracle.com; h=subject : from : to : cc : date : message-id : in-reply-to : references : mime-version : content-type : content-transfer-encoding; s=corp-2019-08-05; bh=12T758YK4pNRpgtjt/siwwGT2UFHn7jBP1SDjBHtIC0=; b=dFFW0nX6/kXEIgSzY4Eqmx9DYtTAMtVcIsO8G9TS/QLsHSDCxjZ3/UWpGNm4hHqhxMVo Ir1OgLxrBfkf5nZzxOP+RA5wfcdt6NuDdGLN9O2UFFqgQ7OBE0oPkJIJIx3LKuEotqDr v0xa0u9pi3Yx5JOy4OzZgpPnGIq8/9fNCmB7FuR+a5+DLHAFAbnp5hiMWdngVj8ZUAZj Lu6pAG2N/kIyAAdXKfhkvdp0WI1HLtU5gxpxbwpIBMm+tdEHNbsc0Afxprjg/KEcdkn6 NXFFZVSfx2YxWuanVmiiB666GDY9hJGtUscjFX17sjBXbtx26HJk+xrJQGmY48n2dWPf FQ== Received: from aserp3020.oracle.com (aserp3020.oracle.com [141.146.126.70]) by userp2120.oracle.com with ESMTP id 2vxwhfgb0f-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Tue, 29 Oct 2019 23:32:18 +0000 Received: from pps.filterd (aserp3020.oracle.com [127.0.0.1]) by aserp3020.oracle.com (8.16.0.27/8.16.0.27) with SMTP id x9TNS6OS178629 for ; Tue, 29 Oct 2019 23:30:17 GMT Received: from aserv0122.oracle.com (aserv0122.oracle.com [141.146.126.236]) by aserp3020.oracle.com with ESMTP id 2vxwj54vxt-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Tue, 29 Oct 2019 23:30:17 +0000 Received: from abhmp0013.oracle.com (abhmp0013.oracle.com [141.146.116.19]) by aserv0122.oracle.com (8.14.4/8.14.4) with ESMTP id x9TNUHqF023684 for ; Tue, 29 Oct 2019 23:30:17 GMT Received: from localhost (/67.169.218.210) by default (Oracle Beehive Gateway v4.0) with ESMTP ; Tue, 29 Oct 2019 16:30:16 -0700 Subject: [PATCH 2/4] xfs: introduce fake roots for inode-rooted btrees From: "Darrick J. Wong" To: darrick.wong@oracle.com Cc: linux-xfs@vger.kernel.org Date: Tue, 29 Oct 2019 16:30:15 -0700 Message-ID: <157239181588.1266747.11813424979199195366.stgit@magnolia> In-Reply-To: <157239180345.1266747.1586648446090389492.stgit@magnolia> References: <157239180345.1266747.1586648446090389492.stgit@magnolia> User-Agent: StGit/0.17.1-dirty MIME-Version: 1.0 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9425 signatures=668685 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 suspectscore=3 malwarescore=0 phishscore=0 bulkscore=0 spamscore=0 mlxscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1908290000 definitions=main-1910290206 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9425 signatures=668685 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 priorityscore=1501 malwarescore=0 suspectscore=3 phishscore=0 bulkscore=0 spamscore=0 clxscore=1015 lowpriorityscore=0 mlxscore=0 impostorscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1908290000 definitions=main-1910290206 Sender: linux-xfs-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-xfs@vger.kernel.org From: Darrick J. Wong Create an in-core fake root for inode-rooted btree types so that callers can generate a whole new btree using the upcoming btree bulk load function without making the new tree accessible from the rest of the filesystem. It is up to the individual btree type to provide a function to create a staged cursor (presumably with the appropriate callouts to update the fakeroot) and then commit the staged root back into the filesystem. Signed-off-by: Darrick J. Wong --- fs/xfs/libxfs/xfs_btree.c | 82 +++++++++++++++++++++++++++++++++++++++++++-- fs/xfs/libxfs/xfs_btree.h | 52 ++++++++++++++++++++++++----- fs/xfs/xfs_trace.h | 33 ++++++++++++++++++ 3 files changed, 154 insertions(+), 13 deletions(-) diff --git a/fs/xfs/libxfs/xfs_btree.c b/fs/xfs/libxfs/xfs_btree.c index ab0d459c10ea..5949ba3b0825 100644 --- a/fs/xfs/libxfs/xfs_btree.c +++ b/fs/xfs/libxfs/xfs_btree.c @@ -646,6 +646,17 @@ xfs_btree_ptr_addr( ((char *)block + xfs_btree_ptr_offset(cur, n, level)); } +struct xfs_ifork * +xfs_btree_ifork_ptr( + struct xfs_btree_cur *cur) +{ + ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE); + + if (cur->bc_flags & XFS_BTREE_STAGING) + return cur->bc_private.b.ifake->if_fork; + return XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork); +} + /* * Get the root block which is stored in the inode. * @@ -656,9 +667,8 @@ STATIC struct xfs_btree_block * xfs_btree_get_iroot( struct xfs_btree_cur *cur) { - struct xfs_ifork *ifp; + struct xfs_ifork *ifp = xfs_btree_ifork_ptr(cur); - ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork); return (struct xfs_btree_block *)ifp->if_broot; } @@ -4980,8 +4990,17 @@ xfs_btree_fakeroot_init_ptr_from_cur( ASSERT(cur->bc_flags & XFS_BTREE_STAGING); - afake = cur->bc_private.a.afake; - ptr->s = cpu_to_be32(afake->af_root); + if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) { + /* + * The root block lives in the inode core, so we zero the + * pointer (like the bmbt code does) to make it obvious if + * anyone ever tries to use this pointer. + */ + ptr->l = cpu_to_be64(0); + } else { + afake = cur->bc_private.a.afake; + ptr->s = cpu_to_be32(afake->af_root); + } } /* Set the root block when our tree has a fakeroot. */ @@ -5053,3 +5072,58 @@ xfs_btree_commit_afakeroot( cur->bc_ops = ops; cur->bc_flags &= ~XFS_BTREE_STAGING; } +/* + * Initialize an inode-rooted btree cursor with the given inode btree fake + * root. The btree cursor's @bc_ops will be overridden as needed to make the + * staging functionality work. If @new_ops is not NULL, these new ops will be + * passed out to the caller for further overriding. + */ +void +xfs_btree_stage_ifakeroot( + struct xfs_btree_cur *cur, + struct xbtree_ifakeroot *ifake, + struct xfs_btree_ops **new_ops) +{ + struct xfs_btree_ops *nops; + + ASSERT(!(cur->bc_flags & XFS_BTREE_STAGING)); + ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE); + + nops = kmem_alloc(sizeof(struct xfs_btree_ops), KM_NOFS); + memcpy(nops, cur->bc_ops, sizeof(struct xfs_btree_ops)); + nops->alloc_block = xfs_btree_fakeroot_alloc_block; + nops->free_block = xfs_btree_fakeroot_free_block; + nops->init_ptr_from_cur = xfs_btree_fakeroot_init_ptr_from_cur; + nops->dup_cursor = xfs_btree_fakeroot_dup_cursor; + + cur->bc_private.b.ifake = ifake; + cur->bc_nlevels = ifake->if_levels; + cur->bc_ops = nops; + cur->bc_flags |= XFS_BTREE_STAGING; + + if (new_ops) + *new_ops = nops; +} + +/* + * Transform an inode-rooted staging btree cursor back into a regular cursor by + * substituting a real btree root for the fake one and restoring normal btree + * cursor ops. The caller must log the btree root change prior to calling + * this. + */ +void +xfs_btree_commit_ifakeroot( + struct xfs_btree_cur *cur, + int whichfork, + const struct xfs_btree_ops *ops) +{ + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + + trace_xfs_btree_commit_ifakeroot(cur); + + kmem_free((void *)cur->bc_ops); + cur->bc_private.b.ifake = NULL; + cur->bc_private.b.whichfork = whichfork; + cur->bc_ops = ops; + cur->bc_flags &= ~XFS_BTREE_STAGING; +} diff --git a/fs/xfs/libxfs/xfs_btree.h b/fs/xfs/libxfs/xfs_btree.h index 39ca3f0a3bf8..d1df48d3d1ee 100644 --- a/fs/xfs/libxfs/xfs_btree.h +++ b/fs/xfs/libxfs/xfs_btree.h @@ -10,6 +10,7 @@ struct xfs_buf; struct xfs_inode; struct xfs_mount; struct xfs_trans; +struct xfs_ifork; extern kmem_zone_t *xfs_btree_cur_zone; @@ -195,6 +196,18 @@ struct xfs_btree_priv_ag { /* needed for BNO, CNT, INO */ union xfs_btree_cur_private priv; }; +/* Private information for an inode-rooted btree. */ +struct xfs_btree_priv_inode { /* needed for BMAP */ + struct xfs_inode *ip; /* pointer to our inode */ + struct xbtree_ifakeroot *ifake; /* fake inode fork */ + int allocated; /* count of alloced */ + short forksize; /* fork's inode space */ + char whichfork; /* data or attr fork */ + char flags; /* flags */ +#define XFS_BTCUR_BPRV_WASDEL (1<<0) /* was delayed */ +#define XFS_BTCUR_BPRV_INVALID_OWNER (1<<1) /* for ext swap */ +}; + /* * Btree cursor structure. * This collects all information needed by the btree code in one place. @@ -217,15 +230,7 @@ typedef struct xfs_btree_cur int bc_statoff; /* offset of btre stats array */ union { struct xfs_btree_priv_ag a; - struct { /* needed for BMAP */ - struct xfs_inode *ip; /* pointer to our inode */ - int allocated; /* count of alloced */ - short forksize; /* fork's inode space */ - char whichfork; /* data or attr fork */ - char flags; /* flags */ -#define XFS_BTCUR_BPRV_WASDEL (1<<0) /* was delayed */ -#define XFS_BTCUR_BPRV_INVALID_OWNER (1<<1) /* for ext swap */ - } b; + struct xfs_btree_priv_inode b; } bc_private; /* per-btree type data */ } xfs_btree_cur_t; @@ -532,6 +537,7 @@ union xfs_btree_key *xfs_btree_high_key_from_key(struct xfs_btree_cur *cur, int xfs_btree_has_record(struct xfs_btree_cur *cur, union xfs_btree_irec *low, union xfs_btree_irec *high, bool *exists); bool xfs_btree_has_more_records(struct xfs_btree_cur *cur); +struct xfs_ifork *xfs_btree_ifork_ptr(struct xfs_btree_cur *cur); /* Fake root for an AG-rooted btree. */ struct xbtree_afakeroot { @@ -553,4 +559,32 @@ void xfs_btree_commit_afakeroot(struct xfs_btree_cur *cur, struct xfs_buf *agbp, const struct xfs_btree_ops *ops); +/* Fake root for an inode-rooted btree. */ +struct xbtree_ifakeroot { + /* Fake inode fork. */ + struct xfs_ifork *if_fork; + + /* Number of blocks used by the btree. */ + int64_t if_blocks; + + /* Height of the new btree. */ + unsigned int if_levels; + + /* Number of bytes available for this fork in the inode. */ + unsigned int if_fork_size; + + /* Fork format. */ + unsigned int if_format; + + /* Number of records. */ + unsigned int if_extents; +}; + +/* Cursor interactions with with fake roots for inode-rooted btrees. */ +void xfs_btree_stage_ifakeroot(struct xfs_btree_cur *cur, + struct xbtree_ifakeroot *ifake, + struct xfs_btree_ops **new_ops); +void xfs_btree_commit_ifakeroot(struct xfs_btree_cur *cur, int whichfork, + const struct xfs_btree_ops *ops); + #endif /* __XFS_BTREE_H__ */ diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h index e04ed5324696..a78055521fcd 100644 --- a/fs/xfs/xfs_trace.h +++ b/fs/xfs/xfs_trace.h @@ -3637,6 +3637,39 @@ TRACE_EVENT(xfs_btree_commit_afakeroot, __entry->agbno) ) +TRACE_EVENT(xfs_btree_commit_ifakeroot, + TP_PROTO(struct xfs_btree_cur *cur), + TP_ARGS(cur), + TP_STRUCT__entry( + __field(dev_t, dev) + __field(xfs_btnum_t, btnum) + __field(xfs_agnumber_t, agno) + __field(xfs_agino_t, agino) + __field(unsigned int, levels) + __field(unsigned int, blocks) + __field(int, whichfork) + ), + TP_fast_assign( + __entry->dev = cur->bc_mp->m_super->s_dev; + __entry->btnum = cur->bc_btnum; + __entry->agno = XFS_INO_TO_AGNO(cur->bc_mp, + cur->bc_private.b.ip->i_ino); + __entry->agino = XFS_INO_TO_AGINO(cur->bc_mp, + cur->bc_private.b.ip->i_ino); + __entry->levels = cur->bc_private.b.ifake->if_levels; + __entry->blocks = cur->bc_private.b.ifake->if_blocks; + __entry->whichfork = cur->bc_private.b.whichfork; + ), + TP_printk("dev %d:%d btree %s ag %u agino %u whichfork %s levels %u blocks %u", + MAJOR(__entry->dev), MINOR(__entry->dev), + __print_symbolic(__entry->btnum, XFS_BTNUM_STRINGS), + __entry->agno, + __entry->agino, + __entry->whichfork == XFS_ATTR_FORK ? "attr" : "data", + __entry->levels, + __entry->blocks) +) + #endif /* _TRACE_XFS_H */ #undef TRACE_INCLUDE_PATH From patchwork Tue Oct 29 23:30:22 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Darrick J. Wong" X-Patchwork-Id: 11218777 Return-Path: Received: from mail.kernel.org (pdx-korg-mail-1.web.codeaurora.org [172.30.200.123]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 73EED139A for ; Tue, 29 Oct 2019 23:30:29 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 3461021906 for ; Tue, 29 Oct 2019 23:30:29 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=pass (2048-bit key) header.d=oracle.com header.i=@oracle.com header.b="anLKup2Q" Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726076AbfJ2Xa2 (ORCPT ); Tue, 29 Oct 2019 19:30:28 -0400 Received: from aserp2120.oracle.com ([141.146.126.78]:50954 "EHLO aserp2120.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726028AbfJ2Xa2 (ORCPT ); Tue, 29 Oct 2019 19:30:28 -0400 Received: from pps.filterd (aserp2120.oracle.com [127.0.0.1]) by aserp2120.oracle.com (8.16.0.27/8.16.0.27) with SMTP id x9TNT3Na017448 for ; Tue, 29 Oct 2019 23:30:24 GMT DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=oracle.com; h=subject : from : to : cc : date : message-id : in-reply-to : references : mime-version : content-type : content-transfer-encoding; s=corp-2019-08-05; bh=Pl3wPM1iD+15m2x8uyaqfOIoDhm3xz0XcNJCLS5mRlw=; b=anLKup2Q36npImsy8U/HElHK6CGo0FgJ/ULF9BJ0OHePgE9RRQGkkEgbRqwjblTkSnzp agRimDs5u8XBAx3HEyuCwhCyXHLwWP1fybhTUASXuZuDAA/yfMdaSKZ6uKvBiAagE3JY s46blys4rjh+l3Gh9+Xt0M5hdOshHsd7uT1ICGuAd2lhuojrfbdZDXv6/mlckDAUcm6H MY+Em8EzqDBhskXG9ab2VajcI77VFDSWraknyIoxhNWK6qcuKVgR48TIgUkS8nGVgHp2 UatGpz9T8N2c+M/qr9o+XkPuClIqBT2Zp677J/x80nVafVVx88B2jUxiht7NIndPwTib lA== Received: from aserp3020.oracle.com (aserp3020.oracle.com [141.146.126.70]) by aserp2120.oracle.com with ESMTP id 2vxwhf8b1d-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Tue, 29 Oct 2019 23:30:24 +0000 Received: from pps.filterd (aserp3020.oracle.com [127.0.0.1]) by aserp3020.oracle.com (8.16.0.27/8.16.0.27) with SMTP id x9TNS5Ir178613 for ; Tue, 29 Oct 2019 23:30:24 GMT Received: from aserv0121.oracle.com (aserv0121.oracle.com [141.146.126.235]) by aserp3020.oracle.com with ESMTP id 2vxwj54w2u-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Tue, 29 Oct 2019 23:30:23 +0000 Received: from abhmp0023.oracle.com (abhmp0023.oracle.com [141.146.116.29]) by aserv0121.oracle.com (8.14.4/8.13.8) with ESMTP id x9TNUNxK011147 for ; Tue, 29 Oct 2019 23:30:23 GMT Received: from localhost (/67.169.218.210) by default (Oracle Beehive Gateway v4.0) with ESMTP ; Tue, 29 Oct 2019 23:30:22 +0000 Subject: [PATCH 3/4] xfs: support bulk loading of staged btrees From: "Darrick J. Wong" To: darrick.wong@oracle.com Cc: linux-xfs@vger.kernel.org Date: Tue, 29 Oct 2019 16:30:22 -0700 Message-ID: <157239182199.1266747.9716632567665840722.stgit@magnolia> In-Reply-To: <157239180345.1266747.1586648446090389492.stgit@magnolia> References: <157239180345.1266747.1586648446090389492.stgit@magnolia> User-Agent: StGit/0.17.1-dirty MIME-Version: 1.0 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9425 signatures=668685 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 suspectscore=3 malwarescore=0 phishscore=0 bulkscore=0 spamscore=0 mlxscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1908290000 definitions=main-1910290206 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9425 signatures=668685 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 priorityscore=1501 malwarescore=0 suspectscore=3 phishscore=0 bulkscore=0 spamscore=0 clxscore=1015 lowpriorityscore=0 mlxscore=0 impostorscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1908290000 definitions=main-1910290206 Sender: linux-xfs-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-xfs@vger.kernel.org From: Darrick J. Wong Add a new btree function that enables us to bulk load a btree cursor. This will be used by the upcoming online repair patches to generate new btrees. This avoids the programmatic inefficiency of calling xfs_btree_insert in a loop (which generates a lot of log traffic) in favor of stamping out new btree blocks with ordered buffers, and then committing both the new root and scheduling the removal of the old btree blocks in a single transaction commit. The design of this new generic code is based off the btree rebuilding code in xfs_repair's phase 5 code, with the explicit goal of enabling us to share that code between scrub and repair. It has the additional feature of being able to control btree block loading factors. Signed-off-by: Darrick J. Wong --- fs/xfs/libxfs/xfs_btree.c | 581 +++++++++++++++++++++++++++++++++++++++++++++ fs/xfs/libxfs/xfs_btree.h | 46 ++++ fs/xfs/xfs_trace.c | 1 fs/xfs/xfs_trace.h | 85 +++++++ 4 files changed, 712 insertions(+), 1 deletion(-) diff --git a/fs/xfs/libxfs/xfs_btree.c b/fs/xfs/libxfs/xfs_btree.c index 5949ba3b0825..39d0ebc86816 100644 --- a/fs/xfs/libxfs/xfs_btree.c +++ b/fs/xfs/libxfs/xfs_btree.c @@ -1382,7 +1382,7 @@ STATIC void xfs_btree_copy_ptrs( struct xfs_btree_cur *cur, union xfs_btree_ptr *dst_ptr, - union xfs_btree_ptr *src_ptr, + const union xfs_btree_ptr *src_ptr, int numptrs) { ASSERT(numptrs >= 0); @@ -5127,3 +5127,582 @@ xfs_btree_commit_ifakeroot( cur->bc_ops = ops; cur->bc_flags &= ~XFS_BTREE_STAGING; } + +/* + * Bulk Loading of Staged Btrees + * ============================= + * + * This interface is used with a staged btree cursor to create a totally new + * btree with a large number of records (i.e. more than what would fit in a + * single block). When the creation is complete, the new root can be linked + * atomically into the filesystem by committing the staged cursor. + * + * The first step for the caller is to construct a fake btree root structure + * and a staged btree cursor. A staging cursor contains all the geometry + * information for the btree type but will fail all operations that could have + * side effects in the filesystem (e.g. btree shape changes). Regular + * operations will not work unless the staging cursor is committed and becomes + * a regular cursor. + * + * For a btree rooted in an AG header, use an xbtree_afakeroot structure. + * This should be initialized to zero. For a btree rooted in an inode fork, + * use an xbtree_ifakeroot structure. @if_fork_size field should be set to + * the number of bytes available to the fork in the inode; @if_fork should + * point to a freshly allocated xfs_inode_fork; and @if_format should be set + * to the appropriate fork type (e.g. XFS_DINODE_FMT_BTREE). + * + * The next step for the caller is to initialize a struct xfs_btree_bload + * context. The @nr_records field is the number of records that are to be + * loaded into the btree. The @leaf_slack and @node_slack fields are the + * number of records (or key/ptr) slots to leave empty in new btree blocks. + * If a caller sets a slack value to -1, the slack value will be computed to + * fill the block halfway between minrecs and maxrecs items per block. + * + * The number of items placed in each btree block is computed via the following + * algorithm: For leaf levels, the number of items for the level is nr_records. + * For node levels, the number of items for the level is the number of blocks + * in the next lower level of the tree. For each level, the desired number of + * items per block is defined as: + * + * desired = max(minrecs, maxrecs - slack factor) + * + * The number of blocks for the level is defined to be: + * + * blocks = nr_items / desired + * + * Note this is rounded down so that the npb calculation below will never fall + * below minrecs. The number of items that will actually be loaded into each + * btree block is defined as: + * + * npb = nr_items / blocks + * + * Some of the leftmost blocks in the level will contain one extra record as + * needed to handle uneven division. If the number of records in any block + * would exceed maxrecs for that level, blocks is incremented and npb is + * recalculated. + * + * In other words, we compute the number of blocks needed to satisfy a given + * loading level, then spread the items as evenly as possible. + * + * To complete this step, call xfs_btree_bload_compute_geometry, which uses + * those settings to compute the height of the btree and the number of blocks + * that will be needed to construct the btree. These values are stored in the + * @btree_height and @nr_blocks fields. + * + * At this point, the caller must allocate @nr_blocks blocks and save them for + * later. If space is to be allocated transactionally, the staging cursor + * must be deleted before and recreated after, which is why computing the + * geometry is a separate step. + * + * The fourth step in the bulk loading process is to set the function pointers + * in the bload context structure. @get_data will be called for each record + * that will be loaded into the btree; it should set the cursor's bc_rec + * field, which will be converted to on-disk format and copied into the + * appropriate record slot. @alloc_block should supply one of the blocks + * allocated in the previous step. For btrees which are rooted in an inode + * fork, @iroot_size is called to compute the size of the incore btree root + * block. Call xfs_btree_bload to start constructing the btree. + * + * The final step is to commit the staging cursor, which logs the new btree + * root and turns the btree into a regular btree cursor, and free the fake + * roots. + */ + +/* + * Put a btree block that we're loading onto the ordered list and release it. + * The btree blocks will be written when the final transaction swapping the + * btree roots is committed. + */ +static void +xfs_btree_bload_drop_buf( + struct xfs_btree_bload *bbl, + struct xfs_trans *tp, + struct xfs_buf **bpp) +{ + if (*bpp == NULL) + return; + + xfs_buf_delwri_queue(*bpp, &bbl->buffers_list); + xfs_trans_brelse(tp, *bpp); + *bpp = NULL; +} + +/* Allocate and initialize one btree block for bulk loading. */ +STATIC int +xfs_btree_bload_prep_block( + struct xfs_btree_cur *cur, + struct xfs_btree_bload *bbl, + unsigned int level, + unsigned int nr_this_block, + union xfs_btree_ptr *ptrp, + struct xfs_buf **bpp, + struct xfs_btree_block **blockp, + void *priv) +{ + union xfs_btree_ptr new_ptr; + struct xfs_buf *new_bp; + struct xfs_btree_block *new_block; + int ret; + + if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) && + level == cur->bc_nlevels - 1) { + struct xfs_ifork *ifp = cur->bc_private.b.ifake->if_fork; + size_t new_size; + + /* Allocate a new incore btree root block. */ + new_size = bbl->iroot_size(cur, nr_this_block, priv); + ifp->if_broot = kmem_zalloc(new_size, 0); + ifp->if_broot_bytes = (int)new_size; + ifp->if_flags |= XFS_IFBROOT; + + /* Initialize it and send it out. */ + xfs_btree_init_block_int(cur->bc_mp, ifp->if_broot, + XFS_BUF_DADDR_NULL, cur->bc_btnum, level, + nr_this_block, cur->bc_private.b.ip->i_ino, + cur->bc_flags); + + *bpp = NULL; + *blockp = ifp->if_broot; + xfs_btree_set_ptr_null(cur, ptrp); + return 0; + } + + /* Allocate a new leaf block. */ + ret = bbl->alloc_block(cur, &new_ptr, priv); + if (ret) + return ret; + + ASSERT(!xfs_btree_ptr_is_null(cur, &new_ptr)); + + ret = xfs_btree_get_buf_block(cur, &new_ptr, &new_block, &new_bp); + if (ret) + return ret; + + /* Initialize the btree block. */ + xfs_btree_init_block_cur(cur, new_bp, level, nr_this_block); + if (*blockp) + xfs_btree_set_sibling(cur, *blockp, &new_ptr, XFS_BB_RIGHTSIB); + xfs_btree_set_sibling(cur, new_block, ptrp, XFS_BB_LEFTSIB); + xfs_btree_set_numrecs(new_block, nr_this_block); + + /* Release the old block and set the out parameters. */ + xfs_btree_bload_drop_buf(bbl, cur->bc_tp, bpp); + *blockp = new_block; + *bpp = new_bp; + xfs_btree_copy_ptrs(cur, ptrp, &new_ptr, 1); + return 0; +} + +/* Load one leaf block. */ +STATIC int +xfs_btree_bload_leaf( + struct xfs_btree_cur *cur, + unsigned int recs_this_block, + xfs_btree_bload_get_fn get_data, + struct xfs_btree_block *block, + void *priv) +{ + unsigned int j; + int ret; + + /* Fill the leaf block with records. */ + for (j = 1; j <= recs_this_block; j++) { + union xfs_btree_rec *block_recs; + + ret = get_data(cur, priv); + if (ret) + return ret; + block_recs = xfs_btree_rec_addr(cur, j, block); + cur->bc_ops->init_rec_from_cur(cur, block_recs); + } + + return 0; +} + +/* Load one node block. */ +STATIC int +xfs_btree_bload_node( + struct xfs_btree_cur *cur, + unsigned int recs_this_block, + union xfs_btree_ptr *child_ptr, + struct xfs_btree_block *block) +{ + unsigned int j; + int ret; + + /* Fill the node block with keys and pointers. */ + for (j = 1; j <= recs_this_block; j++) { + union xfs_btree_key child_key; + union xfs_btree_ptr *block_ptr; + union xfs_btree_key *block_key; + struct xfs_btree_block *child_block; + struct xfs_buf *child_bp; + + ASSERT(!xfs_btree_ptr_is_null(cur, child_ptr)); + + ret = xfs_btree_get_buf_block(cur, child_ptr, &child_block, + &child_bp); + if (ret) + return ret; + + xfs_btree_get_keys(cur, child_block, &child_key); + + block_ptr = xfs_btree_ptr_addr(cur, j, block); + xfs_btree_copy_ptrs(cur, block_ptr, child_ptr, 1); + + block_key = xfs_btree_key_addr(cur, j, block); + xfs_btree_copy_keys(cur, block_key, &child_key, 1); + + xfs_btree_get_sibling(cur, child_block, child_ptr, + XFS_BB_RIGHTSIB); + xfs_trans_brelse(cur->bc_tp, child_bp); + } + + return 0; +} + +/* + * Compute the maximum number of records (or keyptrs) per block that we want to + * install at this level in the btree. Caller is responsible for having set + * @cur->bc_private.b.forksize to the desired fork size, if appropriate. + */ +STATIC unsigned int +xfs_btree_bload_max_npb( + struct xfs_btree_cur *cur, + struct xfs_btree_bload *bbl, + unsigned int level) +{ + unsigned int ret; + + if (level == cur->bc_nlevels - 1 && cur->bc_ops->get_dmaxrecs) + return cur->bc_ops->get_dmaxrecs(cur, level); + + ret = cur->bc_ops->get_maxrecs(cur, level); + if (level == 0) + ret -= bbl->leaf_slack; + else + ret -= bbl->node_slack; + return ret; +} + +/* + * Compute the desired number of records (or keyptrs) per block that we want to + * install at this level in the btree, which must be somewhere between minrecs + * and max_npb. The caller is free to install fewer records per block. + */ +STATIC unsigned int +xfs_btree_bload_desired_npb( + struct xfs_btree_cur *cur, + struct xfs_btree_bload *bbl, + unsigned int level) +{ + unsigned int npb = xfs_btree_bload_max_npb(cur, bbl, level); + + /* Root blocks are not subject to minrecs rules. */ + if (level == cur->bc_nlevels - 1) + return max(1U, npb); + + return max_t(unsigned int, cur->bc_ops->get_minrecs(cur, level), npb); +} + +/* + * Compute the number of records to be stored in each block at this level and + * the number of blocks for this level. For leaf levels, we must populate an + * empty root block even if there are no records, so we have to have at least + * one block. + */ +STATIC void +xfs_btree_bload_level_geometry( + struct xfs_btree_cur *cur, + struct xfs_btree_bload *bbl, + unsigned int level, + uint64_t nr_this_level, + unsigned int *avg_per_block, + uint64_t *blocks, + uint64_t *blocks_with_extra) +{ + uint64_t npb; + uint64_t dontcare; + unsigned int desired_npb; + unsigned int maxnr; + + maxnr = cur->bc_ops->get_maxrecs(cur, level); + + /* + * Compute the number of blocks we need to fill each block with the + * desired number of records/keyptrs per block. Because desired_npb + * could be minrecs, we use regular integer division (which rounds + * the block count down) so that in the next step the effective # of + * items per block will never be less than desired_npb. + */ + desired_npb = xfs_btree_bload_desired_npb(cur, bbl, level); + *blocks = div64_u64_rem(nr_this_level, desired_npb, &dontcare); + *blocks = max(1ULL, *blocks); + + /* + * Compute the number of records that we will actually put in each + * block, assuming that we want to spread the records evenly between + * the blocks. Take care that the effective # of items per block (npb) + * won't exceed maxrecs even for the blocks that get an extra record, + * since desired_npb could be maxrecs, and in the previous step we + * rounded the block count down. + */ + npb = div64_u64_rem(nr_this_level, *blocks, blocks_with_extra); + if (npb > maxnr || (npb == maxnr && *blocks_with_extra > 0)) { + (*blocks)++; + npb = div64_u64_rem(nr_this_level, *blocks, blocks_with_extra); + } + + *avg_per_block = min_t(uint64_t, npb, nr_this_level); + + trace_xfs_btree_bload_level_geometry(cur, level, nr_this_level, + *avg_per_block, desired_npb, *blocks, + *blocks_with_extra); +} + +/* + * Ensure a slack value is appropriate for the btree. + * + * If the slack value is negative, set slack so that we fill the block to + * halfway between minrecs and maxrecs. Make sure the slack is never so large + * that we can underflow minrecs. + */ +static void +xfs_btree_bload_ensure_slack( + struct xfs_btree_cur *cur, + int *slack, + int level) +{ + int maxr; + int minr; + + /* + * We only care about slack for btree blocks, so set the btree nlevels + * to 3 so that level 0 is a leaf block and level 1 is a node block. + * Avoid straying into inode roots, since we don't do slack there. + */ + cur->bc_nlevels = 3; + maxr = cur->bc_ops->get_maxrecs(cur, level); + minr = cur->bc_ops->get_minrecs(cur, level); + + /* + * If slack is negative, automatically set slack so that we load the + * btree block approximately halfway between minrecs and maxrecs. + * Generally, this will net us 75% loading. + */ + if (*slack < 0) + *slack = maxr - ((maxr + minr) >> 1); + + *slack = min(*slack, maxr - minr); +} + +/* + * Prepare a btree cursor for a bulk load operation by computing the geometry + * fields in @bbl. Caller must ensure that the btree cursor is a staging + * cursor. This function can be called multiple times. + */ +int +xfs_btree_bload_compute_geometry( + struct xfs_btree_cur *cur, + struct xfs_btree_bload *bbl, + uint64_t nr_records) +{ + uint64_t nr_blocks = 0; + uint64_t nr_this_level; + + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + + xfs_btree_bload_ensure_slack(cur, &bbl->leaf_slack, 0); + xfs_btree_bload_ensure_slack(cur, &bbl->node_slack, 1); + + bbl->nr_records = nr_this_level = nr_records; + for (cur->bc_nlevels = 1; cur->bc_nlevels < XFS_BTREE_MAXLEVELS;) { + uint64_t level_blocks; + uint64_t dontcare64; + unsigned int level = cur->bc_nlevels - 1; + unsigned int avg_per_block; + + /* + * If all the things we want to store at this level would fit + * in a single root block, then we have our btree root and are + * done. Note that bmap btrees do not allow records in the + * root. + */ + if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) || level != 0) { + xfs_btree_bload_level_geometry(cur, bbl, level, + nr_this_level, &avg_per_block, + &level_blocks, &dontcare64); + if (nr_this_level <= avg_per_block) { + nr_blocks++; + break; + } + } + + /* + * Otherwise, we have to store all the records for this level + * in blocks and therefore need another level of btree to point + * to those blocks. Increase the number of levels and + * recompute the number of records we can store at this level + * because that can change depending on whether or not a level + * is the root level. + */ + cur->bc_nlevels++; + xfs_btree_bload_level_geometry(cur, bbl, level, nr_this_level, + &avg_per_block, &level_blocks, &dontcare64); + nr_blocks += level_blocks; + nr_this_level = level_blocks; + } + + if (cur->bc_nlevels == XFS_BTREE_MAXLEVELS) + return -EOVERFLOW; + + bbl->btree_height = cur->bc_nlevels; + if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) + bbl->nr_blocks = nr_blocks - 1; + else + bbl->nr_blocks = nr_blocks; + return 0; +} + +/* + * Bulk load a btree. + * + * Load @bbl->nr_records quantity of records into a btree using the supplied + * empty and staging btree cursor @cur and a @bbl that has been filled out by + * the xfs_btree_bload_compute_geometry function. + * + * The @bbl->get_data function must populate the cursor's bc_rec every time it + * is called. The @bbl->alloc_block function will be used to allocate new + * btree blocks. @priv is passed to both functions. + * + * Caller must ensure that @cur is a staging cursor. Any existing btree rooted + * in the fakeroot will be lost, so do not call this function twice. + */ +int +xfs_btree_bload( + struct xfs_btree_cur *cur, + struct xfs_btree_bload *bbl, + void *priv) +{ + union xfs_btree_ptr child_ptr; + union xfs_btree_ptr ptr; + struct xfs_buf *bp = NULL; + struct xfs_btree_block *block = NULL; + uint64_t nr_this_level = bbl->nr_records; + uint64_t blocks; + uint64_t i; + uint64_t blocks_with_extra; + uint64_t total_blocks = 0; + unsigned int avg_per_block; + unsigned int level = 0; + int ret; + + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + + INIT_LIST_HEAD(&bbl->buffers_list); + cur->bc_nlevels = bbl->btree_height; + xfs_btree_set_ptr_null(cur, &child_ptr); + xfs_btree_set_ptr_null(cur, &ptr); + + xfs_btree_bload_level_geometry(cur, bbl, level, nr_this_level, + &avg_per_block, &blocks, &blocks_with_extra); + + /* Load each leaf block. */ + for (i = 0; i < blocks; i++) { + unsigned int nr_this_block = avg_per_block; + + if (i < blocks_with_extra) + nr_this_block++; + + ret = xfs_btree_bload_prep_block(cur, bbl, level, + nr_this_block, &ptr, &bp, &block, priv); + if (ret) + return ret; + + trace_xfs_btree_bload_block(cur, level, i, blocks, &ptr, + nr_this_block); + + ret = xfs_btree_bload_leaf(cur, nr_this_block, bbl->get_data, + block, priv); + if (ret) + goto out; + + /* Record the leftmost pointer to start the next level. */ + if (i == 0) + xfs_btree_copy_ptrs(cur, &child_ptr, &ptr, 1); + } + total_blocks += blocks; + xfs_btree_bload_drop_buf(bbl, cur->bc_tp, &bp); + + /* Populate the internal btree nodes. */ + for (level = 1; level < cur->bc_nlevels; level++) { + union xfs_btree_ptr first_ptr; + + nr_this_level = blocks; + block = NULL; + xfs_btree_set_ptr_null(cur, &ptr); + + xfs_btree_bload_level_geometry(cur, bbl, level, nr_this_level, + &avg_per_block, &blocks, &blocks_with_extra); + + /* Load each node block. */ + for (i = 0; i < blocks; i++) { + unsigned int nr_this_block = avg_per_block; + + if (i < blocks_with_extra) + nr_this_block++; + + ret = xfs_btree_bload_prep_block(cur, bbl, level, + nr_this_block, &ptr, &bp, &block, + priv); + if (ret) + return ret; + + trace_xfs_btree_bload_block(cur, level, i, blocks, + &ptr, nr_this_block); + + ret = xfs_btree_bload_node(cur, nr_this_block, + &child_ptr, block); + if (ret) + goto out; + + /* + * Record the leftmost pointer to start the next level. + */ + if (i == 0) + xfs_btree_copy_ptrs(cur, &first_ptr, &ptr, 1); + } + total_blocks += blocks; + xfs_btree_bload_drop_buf(bbl, cur->bc_tp, &bp); + xfs_btree_copy_ptrs(cur, &child_ptr, &first_ptr, 1); + } + + /* Initialize the new root. */ + if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) { + ASSERT(xfs_btree_ptr_is_null(cur, &ptr)); + cur->bc_private.b.ifake->if_levels = cur->bc_nlevels; + cur->bc_private.b.ifake->if_blocks = total_blocks - 1; + } else { + cur->bc_private.a.afake->af_root = be32_to_cpu(ptr.s); + cur->bc_private.a.afake->af_levels = cur->bc_nlevels; + cur->bc_private.a.afake->af_blocks = total_blocks; + } + + /* + * Write the new blocks to disk. If the ordered list isn't empty after + * that, then something went wrong and we have to fail. This should + * never happen, but we'll check anyway. + */ + ret = xfs_buf_delwri_submit(&bbl->buffers_list); + if (ret) + goto out; + if (!list_empty(&bbl->buffers_list)) { + ASSERT(list_empty(&bbl->buffers_list)); + ret = -EIO; + } +out: + xfs_buf_delwri_cancel(&bbl->buffers_list); + if (bp) + xfs_trans_brelse(cur->bc_tp, bp); + return ret; +} diff --git a/fs/xfs/libxfs/xfs_btree.h b/fs/xfs/libxfs/xfs_btree.h index d1df48d3d1ee..fc2ff4447b53 100644 --- a/fs/xfs/libxfs/xfs_btree.h +++ b/fs/xfs/libxfs/xfs_btree.h @@ -587,4 +587,50 @@ void xfs_btree_stage_ifakeroot(struct xfs_btree_cur *cur, void xfs_btree_commit_ifakeroot(struct xfs_btree_cur *cur, int whichfork, const struct xfs_btree_ops *ops); +typedef int (*xfs_btree_bload_get_fn)(struct xfs_btree_cur *cur, void *priv); +typedef int (*xfs_btree_bload_alloc_block_fn)(struct xfs_btree_cur *cur, + union xfs_btree_ptr *ptr, void *priv); +typedef size_t (*xfs_btree_bload_iroot_size_fn)(struct xfs_btree_cur *cur, + unsigned int nr_this_level, void *priv); + +/* Bulk loading of staged btrees. */ +struct xfs_btree_bload { + /* Buffer list for delwri_queue. */ + struct list_head buffers_list; + + /* Function to store a record in the cursor. */ + xfs_btree_bload_get_fn get_data; + + /* Function to allocate a block for the btree. */ + xfs_btree_bload_alloc_block_fn alloc_block; + + /* Function to compute the size of the in-core btree root block. */ + xfs_btree_bload_iroot_size_fn iroot_size; + + /* Number of records the caller wants to store. */ + uint64_t nr_records; + + /* Number of btree blocks needed to store those records. */ + uint64_t nr_blocks; + + /* + * Number of free records to leave in each leaf block. If this (or + * any of the slack values) are negative, this will be computed to + * be halfway between maxrecs and minrecs. This typically leaves the + * block 75% full. + */ + int leaf_slack; + + /* Number of free keyptrs to leave in each node block. */ + int node_slack; + + /* Computed btree height. */ + unsigned int btree_height; +}; + +int xfs_btree_bload_compute_geometry(struct xfs_btree_cur *cur, + struct xfs_btree_bload *bbl, uint64_t nr_records); +int xfs_btree_bload(struct xfs_btree_cur *cur, struct xfs_btree_bload *bbl, + void *priv); + #endif /* __XFS_BTREE_H__ */ diff --git a/fs/xfs/xfs_trace.c b/fs/xfs/xfs_trace.c index bc85b89f88ca..9b5e58a92381 100644 --- a/fs/xfs/xfs_trace.c +++ b/fs/xfs/xfs_trace.c @@ -6,6 +6,7 @@ #include "xfs.h" #include "xfs_fs.h" #include "xfs_shared.h" +#include "xfs_bit.h" #include "xfs_format.h" #include "xfs_log_format.h" #include "xfs_trans_resv.h" diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h index a78055521fcd..6d7ba64b7a0f 100644 --- a/fs/xfs/xfs_trace.h +++ b/fs/xfs/xfs_trace.h @@ -35,6 +35,7 @@ struct xfs_icreate_log; struct xfs_owner_info; struct xfs_trans_res; struct xfs_inobt_rec_incore; +union xfs_btree_ptr; DECLARE_EVENT_CLASS(xfs_attr_list_class, TP_PROTO(struct xfs_attr_list_context *ctx), @@ -3670,6 +3671,90 @@ TRACE_EVENT(xfs_btree_commit_ifakeroot, __entry->blocks) ) +TRACE_EVENT(xfs_btree_bload_level_geometry, + TP_PROTO(struct xfs_btree_cur *cur, unsigned int level, + uint64_t nr_this_level, unsigned int nr_per_block, + unsigned int desired_npb, uint64_t blocks, + uint64_t blocks_with_extra), + TP_ARGS(cur, level, nr_this_level, nr_per_block, desired_npb, blocks, + blocks_with_extra), + TP_STRUCT__entry( + __field(dev_t, dev) + __field(xfs_btnum_t, btnum) + __field(unsigned int, level) + __field(unsigned int, nlevels) + __field(uint64_t, nr_this_level) + __field(unsigned int, nr_per_block) + __field(unsigned int, desired_npb) + __field(unsigned long long, blocks) + __field(unsigned long long, blocks_with_extra) + ), + TP_fast_assign( + __entry->dev = cur->bc_mp->m_super->s_dev; + __entry->btnum = cur->bc_btnum; + __entry->level = level; + __entry->nlevels = cur->bc_nlevels; + __entry->nr_this_level = nr_this_level; + __entry->nr_per_block = nr_per_block; + __entry->desired_npb = desired_npb; + __entry->blocks = blocks; + __entry->blocks_with_extra = blocks_with_extra; + ), + TP_printk("dev %d:%d btree %s level %u/%u nr_this_level %llu nr_per_block %u desired_npb %u blocks %llu blocks_with_extra %llu", + MAJOR(__entry->dev), MINOR(__entry->dev), + __print_symbolic(__entry->btnum, XFS_BTNUM_STRINGS), + __entry->level, + __entry->nlevels, + __entry->nr_this_level, + __entry->nr_per_block, + __entry->desired_npb, + __entry->blocks, + __entry->blocks_with_extra) +) + +TRACE_EVENT(xfs_btree_bload_block, + TP_PROTO(struct xfs_btree_cur *cur, unsigned int level, + uint64_t block_idx, uint64_t nr_blocks, + union xfs_btree_ptr *ptr, unsigned int nr_records), + TP_ARGS(cur, level, block_idx, nr_blocks, ptr, nr_records), + TP_STRUCT__entry( + __field(dev_t, dev) + __field(xfs_btnum_t, btnum) + __field(unsigned int, level) + __field(unsigned long long, block_idx) + __field(unsigned long long, nr_blocks) + __field(xfs_agnumber_t, agno) + __field(xfs_agblock_t, agbno) + __field(unsigned int, nr_records) + ), + TP_fast_assign( + __entry->dev = cur->bc_mp->m_super->s_dev; + __entry->btnum = cur->bc_btnum; + __entry->level = level; + __entry->block_idx = block_idx; + __entry->nr_blocks = nr_blocks; + if (cur->bc_flags & XFS_BTREE_LONG_PTRS) { + xfs_fsblock_t fsb = be64_to_cpu(ptr->l); + + __entry->agno = XFS_FSB_TO_AGNO(cur->bc_mp, fsb); + __entry->agbno = XFS_FSB_TO_AGBNO(cur->bc_mp, fsb); + } else { + __entry->agno = cur->bc_private.a.agno; + __entry->agbno = be32_to_cpu(ptr->s); + } + __entry->nr_records = nr_records; + ), + TP_printk("dev %d:%d btree %s level %u block %llu/%llu fsb (%u/%u) recs %u", + MAJOR(__entry->dev), MINOR(__entry->dev), + __print_symbolic(__entry->btnum, XFS_BTNUM_STRINGS), + __entry->level, + __entry->block_idx, + __entry->nr_blocks, + __entry->agno, + __entry->agbno, + __entry->nr_records) +) + #endif /* _TRACE_XFS_H */ #undef TRACE_INCLUDE_PATH From patchwork Tue Oct 29 23:30:28 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Darrick J. Wong" X-Patchwork-Id: 11218781 Return-Path: Received: from mail.kernel.org (pdx-korg-mail-1.web.codeaurora.org [172.30.200.123]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id D39721390 for ; Tue, 29 Oct 2019 23:30:35 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 9E1ED222C5 for ; Tue, 29 Oct 2019 23:30:35 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=pass (2048-bit key) header.d=oracle.com header.i=@oracle.com header.b="hWRBzn7S" Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726088AbfJ2Xaf (ORCPT ); Tue, 29 Oct 2019 19:30:35 -0400 Received: from userp2130.oracle.com ([156.151.31.86]:41868 "EHLO userp2130.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726028AbfJ2Xae (ORCPT ); Tue, 29 Oct 2019 19:30:34 -0400 Received: from pps.filterd (userp2130.oracle.com [127.0.0.1]) by userp2130.oracle.com (8.16.0.27/8.16.0.27) with SMTP id x9TNShfM038166 for ; Tue, 29 Oct 2019 23:30:32 GMT DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=oracle.com; h=subject : from : to : cc : date : message-id : in-reply-to : references : mime-version : content-type : content-transfer-encoding; s=corp-2019-08-05; bh=TdUPsrdKWhum8pMbugdYLFZreQd50lGLiVyFDpuc+Io=; b=hWRBzn7SEf1qo3kF545adypoNR9Aw2aO93+qEEd3g3nGqYcNis8kDTWyNQnqUdV8sLPI zGhA3UUBG3KRiEfR+TKzSePf0i5QNKpP9dHTNrcpXTcEPb+ojUi9+28TWaNExJMJbAvr dLm54dS0efcv/31T8IWTQwBdeiRvhUIjPtbZH3FkHuNx01XeujhCZviLPaFg5veluwVg 0PW+x5L0WggTEnZryobCRKvZEGlUQVMGes8lws0ywcKQunyBHZeQk+DhZlIEs7mm76r7 2cqFB8ZdAR76zWazCve+yNUPemtAEQFSJFiw/hOk97rIRWV2EA5ORoe0KXFPSpBOM3ys UA== Received: from aserp3020.oracle.com (aserp3020.oracle.com [141.146.126.70]) by userp2130.oracle.com with ESMTP id 2vxwhfgb2f-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Tue, 29 Oct 2019 23:30:31 +0000 Received: from pps.filterd (aserp3020.oracle.com [127.0.0.1]) by aserp3020.oracle.com (8.16.0.27/8.16.0.27) with SMTP id x9TNS6uk178630 for ; Tue, 29 Oct 2019 23:30:31 GMT Received: from userv0121.oracle.com (userv0121.oracle.com [156.151.31.72]) by aserp3020.oracle.com with ESMTP id 2vxwj54w6x-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Tue, 29 Oct 2019 23:30:30 +0000 Received: from abhmp0004.oracle.com (abhmp0004.oracle.com [141.146.116.10]) by userv0121.oracle.com (8.14.4/8.13.8) with ESMTP id x9TNUTQB006725 for ; Tue, 29 Oct 2019 23:30:29 GMT Received: from localhost (/67.169.218.210) by default (Oracle Beehive Gateway v4.0) with ESMTP ; Tue, 29 Oct 2019 16:30:29 -0700 Subject: [PATCH 4/4] xfs: support staging cursors for per-AG btree types From: "Darrick J. Wong" To: darrick.wong@oracle.com Cc: linux-xfs@vger.kernel.org Date: Tue, 29 Oct 2019 16:30:28 -0700 Message-ID: <157239182846.1266747.3874603601415674314.stgit@magnolia> In-Reply-To: <157239180345.1266747.1586648446090389492.stgit@magnolia> References: <157239180345.1266747.1586648446090389492.stgit@magnolia> User-Agent: StGit/0.17.1-dirty MIME-Version: 1.0 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9425 signatures=668685 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 suspectscore=3 malwarescore=0 phishscore=0 bulkscore=0 spamscore=0 mlxscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1908290000 definitions=main-1910290206 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9425 signatures=668685 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 priorityscore=1501 malwarescore=0 suspectscore=3 phishscore=0 bulkscore=0 spamscore=0 clxscore=1015 lowpriorityscore=0 mlxscore=0 impostorscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1908290000 definitions=main-1910290206 Sender: linux-xfs-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-xfs@vger.kernel.org From: Darrick J. Wong Add support for btree staging cursors for the per-AG btree types. This is needed both for online repair and also to convert xfs_repair to use btree bulk loading. Signed-off-by: Darrick J. Wong --- fs/xfs/libxfs/xfs_alloc_btree.c | 97 ++++++++++++++++++++++++++++++------ fs/xfs/libxfs/xfs_alloc_btree.h | 7 +++ fs/xfs/libxfs/xfs_ialloc_btree.c | 84 +++++++++++++++++++++++++++---- fs/xfs/libxfs/xfs_ialloc_btree.h | 6 ++ fs/xfs/libxfs/xfs_refcount_btree.c | 69 ++++++++++++++++++++++---- fs/xfs/libxfs/xfs_refcount_btree.h | 7 +++ fs/xfs/libxfs/xfs_rmap_btree.c | 66 +++++++++++++++++++++--- fs/xfs/libxfs/xfs_rmap_btree.h | 6 ++ 8 files changed, 294 insertions(+), 48 deletions(-) diff --git a/fs/xfs/libxfs/xfs_alloc_btree.c b/fs/xfs/libxfs/xfs_alloc_btree.c index 2a94543857a1..4d6968437c29 100644 --- a/fs/xfs/libxfs/xfs_alloc_btree.c +++ b/fs/xfs/libxfs/xfs_alloc_btree.c @@ -471,6 +471,40 @@ static const struct xfs_btree_ops xfs_cntbt_ops = { .recs_inorder = xfs_cntbt_recs_inorder, }; +/* Allocate most of a new allocation btree cursor. */ +STATIC struct xfs_btree_cur * +xfs_allocbt_init_common( + struct xfs_mount *mp, + struct xfs_trans *tp, + xfs_agnumber_t agno, + xfs_btnum_t btnum) +{ + struct xfs_btree_cur *cur; + + ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT); + + cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS); + + cur->bc_tp = tp; + cur->bc_mp = mp; + cur->bc_btnum = btnum; + cur->bc_blocklog = mp->m_sb.sb_blocklog; + cur->bc_private.a.agno = agno; + + if (btnum == XFS_BTNUM_CNT) { + cur->bc_ops = &xfs_cntbt_ops; + cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2); + } else { + cur->bc_ops = &xfs_bnobt_ops; + cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2); + } + + if (xfs_sb_version_hascrc(&mp->m_sb)) + cur->bc_flags |= XFS_BTREE_CRC_BLOCKS; + + return cur; +} + /* * Allocate a new allocation btree cursor. */ @@ -485,35 +519,64 @@ xfs_allocbt_init_cursor( struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); struct xfs_btree_cur *cur; - ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT); - - cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS); - - cur->bc_tp = tp; - cur->bc_mp = mp; - cur->bc_btnum = btnum; - cur->bc_blocklog = mp->m_sb.sb_blocklog; - + cur = xfs_allocbt_init_common(mp, tp, agno, btnum); if (btnum == XFS_BTNUM_CNT) { - cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtc_2); - cur->bc_ops = &xfs_cntbt_ops; cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]); - cur->bc_flags = XFS_BTREE_LASTREC_UPDATE; + cur->bc_flags |= XFS_BTREE_LASTREC_UPDATE; } else { - cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_abtb_2); - cur->bc_ops = &xfs_bnobt_ops; cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]); } cur->bc_private.a.agbp = agbp; - cur->bc_private.a.agno = agno; - if (xfs_sb_version_hascrc(&mp->m_sb)) - cur->bc_flags |= XFS_BTREE_CRC_BLOCKS; + return cur; +} + +/* Create a free space btree cursor with a fake root for staging. */ +struct xfs_btree_cur * +xfs_allocbt_stage_cursor( + struct xfs_mount *mp, + struct xfs_trans *tp, + struct xbtree_afakeroot *afake, + xfs_agnumber_t agno, + xfs_btnum_t btnum) +{ + struct xfs_btree_cur *cur; + cur = xfs_allocbt_init_common(mp, tp, agno, btnum); + if (btnum == XFS_BTNUM_BNO) + xfs_btree_stage_afakeroot(cur, afake, NULL); + else + xfs_btree_stage_afakeroot(cur, afake, NULL); return cur; } +/* + * Install a new free space btree root. Caller is responsible for invalidating + * and freeing the old btree blocks. + */ +void +xfs_allocbt_commit_staged_btree( + struct xfs_btree_cur *cur, + struct xfs_buf *agbp) +{ + struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); + struct xbtree_afakeroot *afake = cur->bc_private.a.afake; + + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + + agf->agf_roots[cur->bc_btnum] = cpu_to_be32(afake->af_root); + agf->agf_levels[cur->bc_btnum] = cpu_to_be32(afake->af_levels); + xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS); + + if (cur->bc_btnum == XFS_BTNUM_BNO) { + xfs_btree_commit_afakeroot(cur, agbp, &xfs_bnobt_ops); + } else { + cur->bc_flags |= XFS_BTREE_LASTREC_UPDATE; + xfs_btree_commit_afakeroot(cur, agbp, &xfs_cntbt_ops); + } +} + /* * Calculate number of records in an alloc btree block. */ diff --git a/fs/xfs/libxfs/xfs_alloc_btree.h b/fs/xfs/libxfs/xfs_alloc_btree.h index c9305ebb69f6..dde324609a89 100644 --- a/fs/xfs/libxfs/xfs_alloc_btree.h +++ b/fs/xfs/libxfs/xfs_alloc_btree.h @@ -13,6 +13,7 @@ struct xfs_buf; struct xfs_btree_cur; struct xfs_mount; +struct xbtree_afakeroot; /* * Btree block header size depends on a superblock flag. @@ -48,8 +49,14 @@ struct xfs_mount; extern struct xfs_btree_cur *xfs_allocbt_init_cursor(struct xfs_mount *, struct xfs_trans *, struct xfs_buf *, xfs_agnumber_t, xfs_btnum_t); +struct xfs_btree_cur *xfs_allocbt_stage_cursor(struct xfs_mount *mp, + struct xfs_trans *tp, struct xbtree_afakeroot *afake, + xfs_agnumber_t agno, xfs_btnum_t btnum); extern int xfs_allocbt_maxrecs(struct xfs_mount *, int, int); extern xfs_extlen_t xfs_allocbt_calc_size(struct xfs_mount *mp, unsigned long long len); +void xfs_allocbt_commit_staged_btree(struct xfs_btree_cur *cur, + struct xfs_buf *agbp); + #endif /* __XFS_ALLOC_BTREE_H__ */ diff --git a/fs/xfs/libxfs/xfs_ialloc_btree.c b/fs/xfs/libxfs/xfs_ialloc_btree.c index b82992f795aa..15d8ec692a6e 100644 --- a/fs/xfs/libxfs/xfs_ialloc_btree.c +++ b/fs/xfs/libxfs/xfs_ialloc_btree.c @@ -400,32 +400,27 @@ static const struct xfs_btree_ops xfs_finobt_ops = { }; /* - * Allocate a new inode btree cursor. + * Initialize a new inode btree cursor. */ -struct xfs_btree_cur * /* new inode btree cursor */ -xfs_inobt_init_cursor( +static struct xfs_btree_cur * +xfs_inobt_init_common( struct xfs_mount *mp, /* file system mount point */ struct xfs_trans *tp, /* transaction pointer */ - struct xfs_buf *agbp, /* buffer for agi structure */ xfs_agnumber_t agno, /* allocation group number */ xfs_btnum_t btnum) /* ialloc or free ino btree */ { - struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); struct xfs_btree_cur *cur; cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS); - cur->bc_tp = tp; cur->bc_mp = mp; cur->bc_btnum = btnum; if (btnum == XFS_BTNUM_INO) { - cur->bc_nlevels = be32_to_cpu(agi->agi_level); - cur->bc_ops = &xfs_inobt_ops; cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_ibt_2); + cur->bc_ops = &xfs_inobt_ops; } else { - cur->bc_nlevels = be32_to_cpu(agi->agi_free_level); - cur->bc_ops = &xfs_finobt_ops; cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_fibt_2); + cur->bc_ops = &xfs_finobt_ops; } cur->bc_blocklog = mp->m_sb.sb_blocklog; @@ -433,12 +428,79 @@ xfs_inobt_init_cursor( if (xfs_sb_version_hascrc(&mp->m_sb)) cur->bc_flags |= XFS_BTREE_CRC_BLOCKS; - cur->bc_private.a.agbp = agbp; cur->bc_private.a.agno = agno; + return cur; +} +/* Create an inode btree cursor. */ +struct xfs_btree_cur * +xfs_inobt_init_cursor( + struct xfs_mount *mp, + struct xfs_trans *tp, + struct xfs_buf *agbp, + xfs_agnumber_t agno, + xfs_btnum_t btnum) +{ + struct xfs_btree_cur *cur; + struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); + + cur = xfs_inobt_init_common(mp, tp, agno, btnum); + if (btnum == XFS_BTNUM_INO) + cur->bc_nlevels = be32_to_cpu(agi->agi_level); + else + cur->bc_nlevels = be32_to_cpu(agi->agi_free_level); + cur->bc_private.a.agbp = agbp; return cur; } +/* Create an inode btree cursor with a fake root for staging. */ +struct xfs_btree_cur * +xfs_inobt_stage_cursor( + struct xfs_mount *mp, + struct xfs_trans *tp, + struct xbtree_afakeroot *afake, + xfs_agnumber_t agno, + xfs_btnum_t btnum) +{ + struct xfs_btree_cur *cur; + + cur = xfs_inobt_init_common(mp, tp, agno, btnum); + if (btnum == XFS_BTNUM_INO) + xfs_btree_stage_afakeroot(cur, afake, NULL); + else + xfs_btree_stage_afakeroot(cur, afake, NULL); + return cur; +} + +/* + * Install a new inobt btree root. Caller is responsible for invalidating + * and freeing the old btree blocks. + */ +void +xfs_inobt_commit_staged_btree( + struct xfs_btree_cur *cur, + struct xfs_buf *agbp) +{ + struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp); + struct xbtree_afakeroot *afake = cur->bc_private.a.afake; + + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + + if (cur->bc_btnum == XFS_BTNUM_INO) { + agi->agi_root = cpu_to_be32(afake->af_root); + agi->agi_level = cpu_to_be32(afake->af_levels); + xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_ROOT | + XFS_AGI_LEVEL); + xfs_btree_commit_afakeroot(cur, agbp, &xfs_inobt_ops); + } else { + agi->agi_free_root = cpu_to_be32(afake->af_root); + agi->agi_free_level = cpu_to_be32(afake->af_levels); + xfs_ialloc_log_agi(cur->bc_tp, agbp, XFS_AGI_FREE_ROOT | + XFS_AGI_FREE_LEVEL); + xfs_btree_commit_afakeroot(cur, agbp, &xfs_finobt_ops); + } +} + /* * Calculate number of records in an inobt btree block. */ diff --git a/fs/xfs/libxfs/xfs_ialloc_btree.h b/fs/xfs/libxfs/xfs_ialloc_btree.h index 951305ecaae1..9265b3e08c69 100644 --- a/fs/xfs/libxfs/xfs_ialloc_btree.h +++ b/fs/xfs/libxfs/xfs_ialloc_btree.h @@ -48,6 +48,9 @@ struct xfs_mount; extern struct xfs_btree_cur *xfs_inobt_init_cursor(struct xfs_mount *, struct xfs_trans *, struct xfs_buf *, xfs_agnumber_t, xfs_btnum_t); +struct xfs_btree_cur *xfs_inobt_stage_cursor(struct xfs_mount *mp, + struct xfs_trans *tp, struct xbtree_afakeroot *afake, + xfs_agnumber_t agno, xfs_btnum_t btnum); extern int xfs_inobt_maxrecs(struct xfs_mount *, int, int); /* ir_holemask to inode allocation bitmap conversion */ @@ -68,4 +71,7 @@ int xfs_inobt_cur(struct xfs_mount *mp, struct xfs_trans *tp, xfs_agnumber_t agno, xfs_btnum_t btnum, struct xfs_btree_cur **curpp, struct xfs_buf **agi_bpp); +void xfs_inobt_commit_staged_btree(struct xfs_btree_cur *cur, + struct xfs_buf *agbp); + #endif /* __XFS_IALLOC_BTREE_H__ */ diff --git a/fs/xfs/libxfs/xfs_refcount_btree.c b/fs/xfs/libxfs/xfs_refcount_btree.c index 38529dbacd55..9034b40bd5cf 100644 --- a/fs/xfs/libxfs/xfs_refcount_btree.c +++ b/fs/xfs/libxfs/xfs_refcount_btree.c @@ -311,41 +311,90 @@ static const struct xfs_btree_ops xfs_refcountbt_ops = { }; /* - * Allocate a new refcount btree cursor. + * Initialize a new refcount btree cursor. */ -struct xfs_btree_cur * -xfs_refcountbt_init_cursor( +static struct xfs_btree_cur * +xfs_refcountbt_init_common( struct xfs_mount *mp, struct xfs_trans *tp, - struct xfs_buf *agbp, xfs_agnumber_t agno) { - struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); struct xfs_btree_cur *cur; ASSERT(agno != NULLAGNUMBER); ASSERT(agno < mp->m_sb.sb_agcount); - cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS); + cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS); cur->bc_tp = tp; cur->bc_mp = mp; cur->bc_btnum = XFS_BTNUM_REFC; cur->bc_blocklog = mp->m_sb.sb_blocklog; - cur->bc_ops = &xfs_refcountbt_ops; cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_refcbt_2); - cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level); - - cur->bc_private.a.agbp = agbp; cur->bc_private.a.agno = agno; cur->bc_flags |= XFS_BTREE_CRC_BLOCKS; cur->bc_private.a.priv.refc.nr_ops = 0; cur->bc_private.a.priv.refc.shape_changes = 0; + cur->bc_ops = &xfs_refcountbt_ops; + return cur; +} + +/* Create a btree cursor. */ +struct xfs_btree_cur * +xfs_refcountbt_init_cursor( + struct xfs_mount *mp, + struct xfs_trans *tp, + struct xfs_buf *agbp, + xfs_agnumber_t agno) +{ + struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); + struct xfs_btree_cur *cur; + cur = xfs_refcountbt_init_common(mp, tp, agno); + cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level); + cur->bc_private.a.agbp = agbp; return cur; } +/* Create a btree cursor with a fake root for staging. */ +struct xfs_btree_cur * +xfs_refcountbt_stage_cursor( + struct xfs_mount *mp, + struct xfs_trans *tp, + struct xbtree_afakeroot *afake, + xfs_agnumber_t agno) +{ + struct xfs_btree_cur *cur; + + cur = xfs_refcountbt_init_common(mp, tp, agno); + xfs_btree_stage_afakeroot(cur, afake, NULL); + return cur; +} + +/* + * Swap in the new btree root. Once we pass this point the newly rebuilt btree + * is in place and we have to kill off all the old btree blocks. + */ +void +xfs_refcountbt_commit_staged_btree( + struct xfs_btree_cur *cur, + struct xfs_buf *agbp) +{ + struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); + struct xbtree_afakeroot *afake = cur->bc_private.a.afake; + + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + + agf->agf_refcount_root = cpu_to_be32(afake->af_root); + agf->agf_refcount_level = cpu_to_be32(afake->af_levels); + agf->agf_refcount_blocks = cpu_to_be32(afake->af_blocks); + xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS | + XFS_AGF_REFCOUNT_ROOT | + XFS_AGF_REFCOUNT_LEVEL); + xfs_btree_commit_afakeroot(cur, agbp, &xfs_refcountbt_ops); +} + /* * Calculate the number of records in a refcount btree block. */ diff --git a/fs/xfs/libxfs/xfs_refcount_btree.h b/fs/xfs/libxfs/xfs_refcount_btree.h index ba416f71c824..978b714be9f4 100644 --- a/fs/xfs/libxfs/xfs_refcount_btree.h +++ b/fs/xfs/libxfs/xfs_refcount_btree.h @@ -13,6 +13,7 @@ struct xfs_buf; struct xfs_btree_cur; struct xfs_mount; +struct xbtree_afakeroot; /* * Btree block header size @@ -46,6 +47,9 @@ struct xfs_mount; extern struct xfs_btree_cur *xfs_refcountbt_init_cursor(struct xfs_mount *mp, struct xfs_trans *tp, struct xfs_buf *agbp, xfs_agnumber_t agno); +struct xfs_btree_cur *xfs_refcountbt_stage_cursor(struct xfs_mount *mp, + struct xfs_trans *tp, struct xbtree_afakeroot *afake, + xfs_agnumber_t agno); extern int xfs_refcountbt_maxrecs(int blocklen, bool leaf); extern void xfs_refcountbt_compute_maxlevels(struct xfs_mount *mp); @@ -58,4 +62,7 @@ extern int xfs_refcountbt_calc_reserves(struct xfs_mount *mp, struct xfs_trans *tp, xfs_agnumber_t agno, xfs_extlen_t *ask, xfs_extlen_t *used); +void xfs_refcountbt_commit_staged_btree(struct xfs_btree_cur *cur, + struct xfs_buf *agbp); + #endif /* __XFS_REFCOUNT_BTREE_H__ */ diff --git a/fs/xfs/libxfs/xfs_rmap_btree.c b/fs/xfs/libxfs/xfs_rmap_btree.c index fc78efa52c94..062aeaaa7a8c 100644 --- a/fs/xfs/libxfs/xfs_rmap_btree.c +++ b/fs/xfs/libxfs/xfs_rmap_btree.c @@ -448,17 +448,12 @@ static const struct xfs_btree_ops xfs_rmapbt_ops = { .recs_inorder = xfs_rmapbt_recs_inorder, }; -/* - * Allocate a new allocation btree cursor. - */ -struct xfs_btree_cur * -xfs_rmapbt_init_cursor( +static struct xfs_btree_cur * +xfs_rmapbt_init_common( struct xfs_mount *mp, struct xfs_trans *tp, - struct xfs_buf *agbp, xfs_agnumber_t agno) { - struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); struct xfs_btree_cur *cur; cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS); @@ -468,16 +463,67 @@ xfs_rmapbt_init_cursor( cur->bc_btnum = XFS_BTNUM_RMAP; cur->bc_flags = XFS_BTREE_CRC_BLOCKS | XFS_BTREE_OVERLAPPING; cur->bc_blocklog = mp->m_sb.sb_blocklog; - cur->bc_ops = &xfs_rmapbt_ops; - cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]); cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_rmap_2); + cur->bc_private.a.agno = agno; + cur->bc_ops = &xfs_rmapbt_ops; + return cur; +} + +/* Create a new reverse mapping btree cursor. */ +struct xfs_btree_cur * +xfs_rmapbt_init_cursor( + struct xfs_mount *mp, + struct xfs_trans *tp, + struct xfs_buf *agbp, + xfs_agnumber_t agno) +{ + struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); + struct xfs_btree_cur *cur; + + cur = xfs_rmapbt_init_common(mp, tp, agno); + cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]); cur->bc_private.a.agbp = agbp; - cur->bc_private.a.agno = agno; + return cur; +} + +/* Create a new reverse mapping btree cursor with a fake root for staging. */ +struct xfs_btree_cur * +xfs_rmapbt_stage_cursor( + struct xfs_mount *mp, + struct xfs_trans *tp, + struct xbtree_afakeroot *afake, + xfs_agnumber_t agno) +{ + struct xfs_btree_cur *cur; + cur = xfs_rmapbt_init_common(mp, tp, agno); + xfs_btree_stage_afakeroot(cur, afake, NULL); return cur; } +/* + * Install a new reverse mapping btree root. Caller is responsible for + * invalidating and freeing the old btree blocks. + */ +void +xfs_rmapbt_commit_staged_btree( + struct xfs_btree_cur *cur, + struct xfs_buf *agbp) +{ + struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); + struct xbtree_afakeroot *afake = cur->bc_private.a.afake; + + ASSERT(cur->bc_flags & XFS_BTREE_STAGING); + + agf->agf_roots[cur->bc_btnum] = cpu_to_be32(afake->af_root); + agf->agf_levels[cur->bc_btnum] = cpu_to_be32(afake->af_levels); + agf->agf_rmap_blocks = cpu_to_be32(afake->af_blocks); + xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS | + XFS_AGF_RMAP_BLOCKS); + xfs_btree_commit_afakeroot(cur, agbp, &xfs_rmapbt_ops); +} + /* * Calculate number of records in an rmap btree block. */ diff --git a/fs/xfs/libxfs/xfs_rmap_btree.h b/fs/xfs/libxfs/xfs_rmap_btree.h index 820d668b063d..c6785c7851a8 100644 --- a/fs/xfs/libxfs/xfs_rmap_btree.h +++ b/fs/xfs/libxfs/xfs_rmap_btree.h @@ -9,6 +9,7 @@ struct xfs_buf; struct xfs_btree_cur; struct xfs_mount; +struct xbtree_afakeroot; /* rmaps only exist on crc enabled filesystems */ #define XFS_RMAP_BLOCK_LEN XFS_BTREE_SBLOCK_CRC_LEN @@ -43,6 +44,11 @@ struct xfs_mount; struct xfs_btree_cur *xfs_rmapbt_init_cursor(struct xfs_mount *mp, struct xfs_trans *tp, struct xfs_buf *bp, xfs_agnumber_t agno); +struct xfs_btree_cur *xfs_rmapbt_stage_cursor(struct xfs_mount *mp, + struct xfs_trans *tp, struct xbtree_afakeroot *afake, + xfs_agnumber_t agno); +void xfs_rmapbt_commit_staged_btree(struct xfs_btree_cur *cur, + struct xfs_buf *agbp); int xfs_rmapbt_maxrecs(int blocklen, int leaf); extern void xfs_rmapbt_compute_maxlevels(struct xfs_mount *mp);