From owner-svn-src-head@FreeBSD.ORG Sat Nov 3 18:55:56 2012 Return-Path: Delivered-To: svn-src-head@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [69.147.83.52]) by hub.freebsd.org (Postfix) with ESMTP id 99CF3D41; Sat, 3 Nov 2012 18:55:56 +0000 (UTC) (envelope-from mckusick@FreeBSD.org) Received: from svn.freebsd.org (svn.freebsd.org [IPv6:2001:4f8:fff6::2c]) by mx1.freebsd.org (Postfix) with ESMTP id 816008FC15; Sat, 3 Nov 2012 18:55:56 +0000 (UTC) Received: from svn.freebsd.org (localhost [127.0.0.1]) by svn.freebsd.org (8.14.4/8.14.4) with ESMTP id qA3Itugk024812; Sat, 3 Nov 2012 18:55:56 GMT (envelope-from mckusick@svn.freebsd.org) Received: (from mckusick@localhost) by svn.freebsd.org (8.14.4/8.14.4/Submit) id qA3Ituxr024809; Sat, 3 Nov 2012 18:55:56 GMT (envelope-from mckusick@svn.freebsd.org) Message-Id: <201211031855.qA3Ituxr024809@svn.freebsd.org> From: Kirk McKusick Date: Sat, 3 Nov 2012 18:55:56 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r242520 - head/sys/ufs/ffs X-SVN-Group: head MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-head@freebsd.org X-Mailman-Version: 2.1.14 Precedence: list List-Id: SVN commit messages for the src tree for head/-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 03 Nov 2012 18:55:56 -0000 Author: mckusick Date: Sat Nov 3 18:55:55 2012 New Revision: 242520 URL: http://svn.freebsd.org/changeset/base/242520 Log: When a file is first being written, the dynamic block reallocation (implemented by ffs_reallocblks_ufs[12]) relocates the file's blocks so as to cluster them together into a contiguous set of blocks on the disk. When the cluster crosses the boundary into the first indirect block, the first indirect block is initially allocated in a position immediately following the last direct block. Block reallocation would usually destroy locality by moving the indirect block out of the way to keep the data blocks contiguous. This change compensates for this problem by noting that the first indirect block should be left immediately following the last direct block. It then tries to start a new cluster of contiguous blocks (referenced by the indirect block) immediately following the indirect block. We should also do this for other indirect block boundaries, but it is only important for the first one. Suggested by: Bruce Evans MFC: 2 weeks Modified: head/sys/ufs/ffs/ffs_alloc.c head/sys/ufs/ffs/ffs_balloc.c Modified: head/sys/ufs/ffs/ffs_alloc.c ============================================================================== --- head/sys/ufs/ffs/ffs_alloc.c Sat Nov 3 18:38:28 2012 (r242519) +++ head/sys/ufs/ffs/ffs_alloc.c Sat Nov 3 18:55:55 2012 (r242520) @@ -535,6 +535,18 @@ ffs_reallocblks_ufs1(ap) panic("ffs_reallocblks: non-physical cluster %d", i); #endif /* + * If the cluster crosses the boundary for the first indirect + * block, leave space for the indirect block. Indirect blocks + * are initially laid out in a position after the last direct + * block. Block reallocation would usually destroy locality by + * moving the indirect block out of the way to make room for + * data blocks if we didn't compensate here. We should also do + * this for other indirect block boundaries, but it is only + * important for the first one. + */ + if (start_lbn < NDADDR && end_lbn >= NDADDR) + return (ENOSPC); + /* * If the latest allocation is in a new cylinder group, assume that * the filesystem has decided to move and do not force it back to * the previous cylinder group. @@ -744,6 +756,18 @@ ffs_reallocblks_ufs2(ap) panic("ffs_reallocblks: non-physical cluster %d", i); #endif /* + * If the cluster crosses the boundary for the first indirect + * block, do not move anything in it. Indirect blocks are + * usually initially laid out in a position between the data + * blocks. Block reallocation would usually destroy locality by + * moving the indirect block out of the way to make room for + * data blocks if we didn't compensate here. We should also do + * this for other indirect block boundaries, but it is only + * important for the first one. + */ + if (start_lbn < NDADDR && end_lbn >= NDADDR) + return (ENOSPC); + /* * If the latest allocation is in a new cylinder group, assume that * the filesystem has decided to move and do not force it back to * the previous cylinder group. @@ -791,6 +815,15 @@ ffs_reallocblks_ufs2(ap) UFS_LOCK(ump); pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap); /* + * Skip a block for the first indirect block. Indirect blocks are + * usually initially laid out in a good position between the data + * blocks, but block reallocation would usually destroy locality by + * moving them out of the way to make room for data blocks if we + * didn't compensate here. + */ + if (start_lbn == NDADDR) + pref += fs->fs_frag; + /* * Search the block map looking for an allocation of the desired size. */ if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref, @@ -1185,9 +1218,25 @@ ffs_blkpref_ufs1(ip, lbn, indx, bap) struct fs *fs; u_int cg; u_int avgbfree, startcg; + ufs2_daddr_t pref; mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED); fs = ip->i_fs; + /* + * If we are allocating the first indirect block, try to place it + * immediately following the last direct block. + * + * If we are allocating the first data block in the first indirect + * block, try to place it immediately following the indirect block. + */ + if (lbn == NDADDR) { + pref = ip->i_din1->di_db[NDADDR - 1]; + if (bap == NULL && pref != 0) + return (pref + fs->fs_frag); + pref = ip->i_din1->di_ib[0]; + if (pref != 0) + return (pref + fs->fs_frag); + } if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { if (lbn < NDADDR + NINDIR(fs)) { cg = ino_to_cg(fs, ip->i_number); @@ -1235,9 +1284,25 @@ ffs_blkpref_ufs2(ip, lbn, indx, bap) struct fs *fs; u_int cg; u_int avgbfree, startcg; + ufs2_daddr_t pref; mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED); fs = ip->i_fs; + /* + * If we are allocating the first indirect block, try to place it + * immediately following the last direct block. + * + * If we are allocating the first data block in the first indirect + * block, try to place it immediately following the indirect block. + */ + if (lbn == NDADDR) { + pref = ip->i_din1->di_db[NDADDR - 1]; + if (bap == NULL && pref != 0) + return (pref + fs->fs_frag); + pref = ip->i_din1->di_ib[0]; + if (pref != 0) + return (pref + fs->fs_frag); + } if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { if (lbn < NDADDR + NINDIR(fs)) { cg = ino_to_cg(fs, ip->i_number); Modified: head/sys/ufs/ffs/ffs_balloc.c ============================================================================== --- head/sys/ufs/ffs/ffs_balloc.c Sat Nov 3 18:38:28 2012 (r242519) +++ head/sys/ufs/ffs/ffs_balloc.c Sat Nov 3 18:55:55 2012 (r242520) @@ -251,6 +251,7 @@ ffs_balloc_ufs1(struct vnode *vp, off_t curthread_pflags_restore(saved_inbdflush); return (error); } + pref = newb + fs->fs_frag; nb = newb; *allocblk++ = nb; *lbns_remfree++ = indirs[1].in_lbn; @@ -315,6 +316,7 @@ retry: } goto fail; } + pref = newb + fs->fs_frag; nb = newb; *allocblk++ = nb; *lbns_remfree++ = indirs[i].in_lbn; @@ -363,7 +365,9 @@ retry: */ if (nb == 0) { UFS_LOCK(ump); - pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off, &bap[0]); + if (pref == 0) + pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off, + &bap[0]); error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags | IO_BUFLOCKED, cred, &newb); if (error) { @@ -789,6 +793,7 @@ ffs_balloc_ufs2(struct vnode *vp, off_t curthread_pflags_restore(saved_inbdflush); return (error); } + pref = newb + fs->fs_frag; nb = newb; *allocblk++ = nb; *lbns_remfree++ = indirs[1].in_lbn; @@ -853,6 +858,7 @@ retry: } goto fail; } + pref = newb + fs->fs_frag; nb = newb; *allocblk++ = nb; *lbns_remfree++ = indirs[i].in_lbn; @@ -901,7 +907,9 @@ retry: */ if (nb == 0) { UFS_LOCK(ump); - pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off, &bap[0]); + if (pref == 0) + pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off, + &bap[0]); error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags | IO_BUFLOCKED, cred, &newb); if (error) {