diff options
author | mckusick <mckusick@FreeBSD.org> | 2015-05-28 00:11:36 +0000 |
---|---|---|
committer | mckusick <mckusick@FreeBSD.org> | 2015-05-28 00:11:36 +0000 |
commit | c94542bef304f991c0684e8a88dbcae364da6342 (patch) | |
tree | b4a35f5f10df4eaba01126a2ffe3ce3bacd636f7 /sys/ufs/ffs | |
parent | e12c5d1ed6e67d05e92e7300d4b38cff76215ae7 (diff) | |
download | FreeBSD-src-c94542bef304f991c0684e8a88dbcae364da6342.zip FreeBSD-src-c94542bef304f991c0684e8a88dbcae364da6342.tar.gz |
MFC of 269533:
Limit the number of cylinder groups that will be searched when
trying to build a cluster. The limit is tunable using the sysctl
vfs.ffs.maxclustersearch. The current limit is 10 cylinder groups
per block allocation. It was previously limited to the number of
cylinder groups in the filesystem per block allocation. When there
were no clusters of the needed size left, it repeatedly searched
the whole filesystem for a non-existent cluster on every block
allocation. The result was very slow filesystem allocation with
100% CPU utilization. The old behavior can be had by setting
vfs.ffs.maxclustersearch to a huge number (1,000,000).
This change affects only the layout policy routines so is not able
to interfere with the integrity of the filesystem.
Reported by: Dmitry Sivachenko (demon@)
Tested by: Dmitry Sivachenko (demon@)
Diffstat (limited to 'sys/ufs/ffs')
-rw-r--r-- | sys/ufs/ffs/ffs_alloc.c | 78 | ||||
-rw-r--r-- | sys/ufs/ffs/ffs_vfsops.c | 1 |
2 files changed, 63 insertions, 16 deletions
diff --git a/sys/ufs/ffs/ffs_alloc.c b/sys/ufs/ffs/ffs_alloc.c index 69f3e99..9982139 100644 --- a/sys/ufs/ffs/ffs_alloc.c +++ b/sys/ufs/ffs/ffs_alloc.c @@ -112,8 +112,7 @@ static void ffs_blkfree_trim_task(void *ctx, int pending __unused); #ifdef INVARIANTS static int ffs_checkblk(struct inode *, ufs2_daddr_t, long); #endif -static ufs2_daddr_t ffs_clusteralloc(struct inode *, u_int, ufs2_daddr_t, int, - int); +static ufs2_daddr_t ffs_clusteralloc(struct inode *, u_int, ufs2_daddr_t, int); static ino_t ffs_dirpref(struct inode *); static ufs2_daddr_t ffs_fragextend(struct inode *, u_int, ufs2_daddr_t, int, int); @@ -460,10 +459,16 @@ nospace: SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem"); static int doasyncfree = 1; -SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, ""); +SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, +"do not force synchronous writes when blocks are reallocated"); static int doreallocblks = 1; -SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, ""); +SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, +"enable block reallocation"); + +static int maxclustersearch = 10; +SYSCTL_INT(_vfs_ffs, OID_AUTO, maxclustersearch, CTLFLAG_RW, &maxclustersearch, +0, "max number of cylinder group to search for contigous blocks"); #ifdef DEBUG static volatile int prtrealloc = 0; @@ -510,7 +515,7 @@ ffs_reallocblks_ufs1(ap) ufs1_daddr_t soff, newblk, blkno; ufs2_daddr_t pref; struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp; - int i, len, start_lvl, end_lvl, ssize; + int i, cg, len, start_lvl, end_lvl, ssize; vp = ap->a_vp; ip = VTOI(vp); @@ -597,18 +602,39 @@ ffs_reallocblks_ufs1(ap) ebap = (ufs1_daddr_t *)ebp->b_data; } /* - * Find the preferred location for the cluster. + * Find the preferred location for the cluster. If we have not + * previously failed at this endeavor, then follow our standard + * preference calculation. If we have failed at it, then pick up + * where we last ended our search. */ UFS_LOCK(ump); - pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap); + if (ip->i_nextclustercg == -1) + pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap); + else + pref = cgdata(fs, ip->i_nextclustercg); /* * Search the block map looking for an allocation of the desired size. + * To avoid wasting too much time, we limit the number of cylinder + * groups that we will search. + */ + cg = dtog(fs, pref); + for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { + if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) + break; + cg += 1; + if (cg >= fs->fs_ncg) + cg = 0; + } + /* + * If we have failed in our search, record where we gave up for + * next time. Otherwise, fall back to our usual search citerion. */ - if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref, - len, len, ffs_clusteralloc)) == 0) { + if (newblk == 0) { + ip->i_nextclustercg = cg; UFS_UNLOCK(ump); goto fail; } + ip->i_nextclustercg = -1; /* * We have found a new contiguous block. * @@ -737,7 +763,7 @@ ffs_reallocblks_ufs2(ap) ufs_lbn_t start_lbn, end_lbn; ufs2_daddr_t soff, newblk, blkno, pref; struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp; - int i, len, start_lvl, end_lvl, ssize; + int i, cg, len, start_lvl, end_lvl, ssize; vp = ap->a_vp; ip = VTOI(vp); @@ -824,18 +850,39 @@ ffs_reallocblks_ufs2(ap) ebap = (ufs2_daddr_t *)ebp->b_data; } /* - * Find the preferred location for the cluster. + * Find the preferred location for the cluster. If we have not + * previously failed at this endeavor, then follow our standard + * preference calculation. If we have failed at it, then pick up + * where we last ended our search. */ UFS_LOCK(ump); - pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap); + if (ip->i_nextclustercg == -1) + pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap); + else + pref = cgdata(fs, ip->i_nextclustercg); /* * Search the block map looking for an allocation of the desired size. + * To avoid wasting too much time, we limit the number of cylinder + * groups that we will search. + */ + cg = dtog(fs, pref); + for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { + if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) + break; + cg += 1; + if (cg >= fs->fs_ncg) + cg = 0; + } + /* + * If we have failed in our search, record where we gave up for + * next time. Otherwise, fall back to our usual search citerion. */ - if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref, - len, len, ffs_clusteralloc)) == 0) { + if (newblk == 0) { + ip->i_nextclustercg = cg; UFS_UNLOCK(ump); goto fail; } + ip->i_nextclustercg = -1; /* * We have found a new contiguous block. * @@ -1785,12 +1832,11 @@ gotit: * take the first one that we find following bpref. */ static ufs2_daddr_t -ffs_clusteralloc(ip, cg, bpref, len, unused) +ffs_clusteralloc(ip, cg, bpref, len) struct inode *ip; u_int cg; ufs2_daddr_t bpref; int len; - int unused; { struct fs *fs; struct cg *cgp; diff --git a/sys/ufs/ffs/ffs_vfsops.c b/sys/ufs/ffs/ffs_vfsops.c index c0b6c9f..a7ed993 100644 --- a/sys/ufs/ffs/ffs_vfsops.c +++ b/sys/ufs/ffs/ffs_vfsops.c @@ -1687,6 +1687,7 @@ ffs_vgetf(mp, ino, flags, vpp, ffs_flags) ip->i_dev = dev; ip->i_number = ino; ip->i_ea_refs = 0; + ip->i_nextclustercg = -1; #ifdef QUOTA { int i; |