[PATCH 004/145] libxfs: backport kernel 4.6 changes
Darrick J. Wong
darrick.wong at oracle.com
Thu Jun 16 20:31:10 CDT 2016
Backport the changes from kernel 4.5 -> 4.6.
Signed-off-by: Darrick J. Wong <darrick.wong at oracle.com>
---
include/xfs_trans.h | 1
libxfs/libxfs_priv.h | 2 -
libxfs/trans.c | 1
libxfs/xfs_bmap.c | 170 +++++++++++++++++++++++++++++++++--------------
libxfs/xfs_bmap_btree.c | 4 +
libxfs/xfs_dir2_node.c | 4 +
libxfs/xfs_ialloc.c | 4 +
7 files changed, 129 insertions(+), 57 deletions(-)
diff --git a/include/xfs_trans.h b/include/xfs_trans.h
index 5467c7f..d7ee1fd 100644
--- a/include/xfs_trans.h
+++ b/include/xfs_trans.h
@@ -81,6 +81,7 @@ typedef struct xfs_trans {
long t_fdblocks_delta; /* superblock fdblocks chg */
long t_frextents_delta; /* superblock freextents chg */
struct list_head t_items; /* first log item desc chunk */
+ unsigned int t_blk_res;
} xfs_trans_t;
void xfs_trans_init(struct xfs_mount *);
diff --git a/libxfs/libxfs_priv.h b/libxfs/libxfs_priv.h
index ef9ff3b..ecd75e7 100644
--- a/libxfs/libxfs_priv.h
+++ b/libxfs/libxfs_priv.h
@@ -187,7 +187,7 @@ enum ce { CE_DEBUG, CE_CONT, CE_NOTE, CE_WARN, CE_ALERT, CE_PANIC };
*/
#define prandom_u32() 0
-#define PAGE_CACHE_SIZE getpagesize()
+#define PAGE_SIZE getpagesize()
static inline int __do_div(unsigned long long *n, unsigned base)
{
diff --git a/libxfs/trans.c b/libxfs/trans.c
index 0388950..18ea010 100644
--- a/libxfs/trans.c
+++ b/libxfs/trans.c
@@ -193,6 +193,7 @@ libxfs_trans_reserve(
if (blocks > 0) {
if (mpsb->sb_fdblocks < blocks)
return -ENOSPC;
+ tp->t_blk_res += blocks;
}
/* user space, don't need log/RT stuff (preserve the API though) */
return 0;
diff --git a/libxfs/xfs_bmap.c b/libxfs/xfs_bmap.c
index 40286e4..cbcfd72 100644
--- a/libxfs/xfs_bmap.c
+++ b/libxfs/xfs_bmap.c
@@ -469,10 +469,7 @@ xfs_bmap_check_leaf_extents(
}
block = XFS_BUF_TO_BLOCK(bp);
}
- if (bp_release) {
- bp_release = 0;
- xfs_trans_brelse(NULL, bp);
- }
+
return;
error0:
@@ -3737,11 +3734,11 @@ xfs_bmap_btalloc(
args.prod = align;
if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
args.mod = (xfs_extlen_t)(args.prod - args.mod);
- } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
+ } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
args.prod = 1;
args.mod = 0;
} else {
- args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
+ args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
args.mod = (xfs_extlen_t)(args.prod - args.mod);
}
@@ -4713,6 +4710,66 @@ error0:
}
/*
+ * When a delalloc extent is split (e.g., due to a hole punch), the original
+ * indlen reservation must be shared across the two new extents that are left
+ * behind.
+ *
+ * Given the original reservation and the worst case indlen for the two new
+ * extents (as calculated by xfs_bmap_worst_indlen()), split the original
+ * reservation fairly across the two new extents. If necessary, steal available
+ * blocks from a deleted extent to make up a reservation deficiency (e.g., if
+ * ores == 1). The number of stolen blocks is returned. The availability and
+ * subsequent accounting of stolen blocks is the responsibility of the caller.
+ */
+static xfs_filblks_t
+xfs_bmap_split_indlen(
+ xfs_filblks_t ores, /* original res. */
+ xfs_filblks_t *indlen1, /* ext1 worst indlen */
+ xfs_filblks_t *indlen2, /* ext2 worst indlen */
+ xfs_filblks_t avail) /* stealable blocks */
+{
+ xfs_filblks_t len1 = *indlen1;
+ xfs_filblks_t len2 = *indlen2;
+ xfs_filblks_t nres = len1 + len2; /* new total res. */
+ xfs_filblks_t stolen = 0;
+
+ /*
+ * Steal as many blocks as we can to try and satisfy the worst case
+ * indlen for both new extents.
+ */
+ while (nres > ores && avail) {
+ nres--;
+ avail--;
+ stolen++;
+ }
+
+ /*
+ * The only blocks available are those reserved for the original
+ * extent and what we can steal from the extent being removed.
+ * If this still isn't enough to satisfy the combined
+ * requirements for the two new extents, skim blocks off of each
+ * of the new reservations until they match what is available.
+ */
+ while (nres > ores) {
+ if (len1) {
+ len1--;
+ nres--;
+ }
+ if (nres == ores)
+ break;
+ if (len2) {
+ len2--;
+ nres--;
+ }
+ }
+
+ *indlen1 = len1;
+ *indlen2 = len2;
+
+ return stolen;
+}
+
+/*
* Called by xfs_bmapi to update file extent records and the btree
* after removing space (or undoing a delayed allocation).
*/
@@ -4976,28 +5033,29 @@ xfs_bmap_del_extent(
XFS_IFORK_NEXT_SET(ip, whichfork,
XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
} else {
+ xfs_filblks_t stolen;
ASSERT(whichfork == XFS_DATA_FORK);
- temp = xfs_bmap_worst_indlen(ip, temp);
+
+ /*
+ * Distribute the original indlen reservation across the
+ * two new extents. Steal blocks from the deleted extent
+ * if necessary. Stealing blocks simply fudges the
+ * fdblocks accounting in xfs_bunmapi().
+ */
+ temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
+ temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
+ stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
+ del->br_blockcount);
+ da_new = temp + temp2 - stolen;
+ del->br_blockcount -= stolen;
+
+ /*
+ * Set the reservation for each extent. Warn if either
+ * is zero as this can lead to delalloc problems.
+ */
+ WARN_ON_ONCE(!temp || !temp2);
xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
- temp2 = xfs_bmap_worst_indlen(ip, temp2);
new.br_startblock = nullstartblock((int)temp2);
- da_new = temp + temp2;
- while (da_new > da_old) {
- if (temp) {
- temp--;
- da_new--;
- xfs_bmbt_set_startblock(ep,
- nullstartblock((int)temp));
- }
- if (da_new == da_old)
- break;
- if (temp2) {
- temp2--;
- da_new--;
- new.br_startblock =
- nullstartblock((int)temp2);
- }
- }
}
trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
xfs_iext_insert(ip, *idx + 1, 1, &new, state);
@@ -5202,7 +5260,7 @@ xfs_bunmapi(
* This is better than zeroing it.
*/
ASSERT(del.br_state == XFS_EXT_NORM);
- ASSERT(xfs_trans_get_block_res(tp) > 0);
+ ASSERT(tp->t_blk_res > 0);
/*
* If this spans a realtime extent boundary,
* chop it back to the start of the one we end at.
@@ -5233,7 +5291,7 @@ xfs_bunmapi(
del.br_startblock += mod;
} else if ((del.br_startoff == start &&
(del.br_state == XFS_EXT_UNWRITTEN ||
- xfs_trans_get_block_res(tp) == 0)) ||
+ tp->t_blk_res == 0)) ||
!xfs_sb_version_hasextflgbit(&mp->m_sb)) {
/*
* Can't make it unwritten. There isn't
@@ -5288,9 +5346,37 @@ xfs_bunmapi(
goto nodelete;
}
}
+
+ /*
+ * If it's the case where the directory code is running
+ * with no block reservation, and the deleted block is in
+ * the middle of its extent, and the resulting insert
+ * of an extent would cause transformation to btree format,
+ * then reject it. The calling code will then swap
+ * blocks around instead.
+ * We have to do this now, rather than waiting for the
+ * conversion to btree format, since the transaction
+ * will be dirty.
+ */
+ if (!wasdel && tp->t_blk_res == 0 &&
+ XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
+ XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
+ XFS_IFORK_MAXEXT(ip, whichfork) &&
+ del.br_startoff > got.br_startoff &&
+ del.br_startoff + del.br_blockcount <
+ got.br_startoff + got.br_blockcount) {
+ error = -ENOSPC;
+ goto error0;
+ }
+
+ /*
+ * Unreserve quota and update realtime free space, if
+ * appropriate. If delayed allocation, update the inode delalloc
+ * counter now and wait to update the sb counters as
+ * xfs_bmap_del_extent() might need to borrow some blocks.
+ */
if (wasdel) {
ASSERT(startblockval(del.br_startblock) > 0);
- /* Update realtime/data freespace, unreserve quota */
if (isrt) {
xfs_filblks_t rtexts;
@@ -5301,8 +5387,6 @@ xfs_bunmapi(
ip, -((long)del.br_blockcount), 0,
XFS_QMOPT_RES_RTBLKS);
} else {
- xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount,
- false);
(void)xfs_trans_reserve_quota_nblks(NULL,
ip, -((long)del.br_blockcount), 0,
XFS_QMOPT_RES_REGBLKS);
@@ -5313,32 +5397,16 @@ xfs_bunmapi(
XFS_BTCUR_BPRV_WASDEL;
} else if (cur)
cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
- /*
- * If it's the case where the directory code is running
- * with no block reservation, and the deleted block is in
- * the middle of its extent, and the resulting insert
- * of an extent would cause transformation to btree format,
- * then reject it. The calling code will then swap
- * blocks around instead.
- * We have to do this now, rather than waiting for the
- * conversion to btree format, since the transaction
- * will be dirty.
- */
- if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
- XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
- XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
- XFS_IFORK_MAXEXT(ip, whichfork) &&
- del.br_startoff > got.br_startoff &&
- del.br_startoff + del.br_blockcount <
- got.br_startoff + got.br_blockcount) {
- error = -ENOSPC;
- goto error0;
- }
+
error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
&tmp_logflags, whichfork);
logflags |= tmp_logflags;
if (error)
goto error0;
+
+ if (!isrt && wasdel)
+ xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
+
bno = del.br_startoff - 1;
nodelete:
/*
diff --git a/libxfs/xfs_bmap_btree.c b/libxfs/xfs_bmap_btree.c
index a63379b..022d4b6 100644
--- a/libxfs/xfs_bmap_btree.c
+++ b/libxfs/xfs_bmap_btree.c
@@ -458,7 +458,7 @@ xfs_bmbt_alloc_block(
* reservation amount is insufficient then we may fail a
* block allocation here and corrupt the filesystem.
*/
- args.minleft = xfs_trans_get_block_res(args.tp);
+ args.minleft = args.tp->t_blk_res;
} else if (cur->bc_private.b.flist->xbf_low) {
args.type = XFS_ALLOCTYPE_START_BNO;
} else {
@@ -467,7 +467,7 @@ xfs_bmbt_alloc_block(
args.minlen = args.maxlen = args.prod = 1;
args.wasdel = cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL;
- if (!args.wasdel && xfs_trans_get_block_res(args.tp) == 0) {
+ if (!args.wasdel && args.tp->t_blk_res == 0) {
error = -ENOSPC;
goto error0;
}
diff --git a/libxfs/xfs_dir2_node.c b/libxfs/xfs_dir2_node.c
index 04fecf1..b75b432 100644
--- a/libxfs/xfs_dir2_node.c
+++ b/libxfs/xfs_dir2_node.c
@@ -2232,6 +2232,9 @@ xfs_dir2_node_trim_free(
dp = args->dp;
tp = args->trans;
+
+ *rvalp = 0;
+
/*
* Read the freespace block.
*/
@@ -2252,7 +2255,6 @@ xfs_dir2_node_trim_free(
*/
if (freehdr.nused > 0) {
xfs_trans_brelse(tp, bp);
- *rvalp = 0;
return 0;
}
/*
diff --git a/libxfs/xfs_ialloc.c b/libxfs/xfs_ialloc.c
index 72e9ff7..4f0e4ee 100644
--- a/libxfs/xfs_ialloc.c
+++ b/libxfs/xfs_ialloc.c
@@ -2396,8 +2396,8 @@ xfs_ialloc_compute_maxlevels(
maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
XFS_INODES_PER_CHUNK_LOG;
- minleafrecs = mp->m_alloc_mnr[0];
- minnoderecs = mp->m_alloc_mnr[1];
+ minleafrecs = mp->m_inobt_mnr[0];
+ minnoderecs = mp->m_inobt_mnr[1];
maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
for (level = 1; maxblocks > 1; level++)
maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
More information about the xfs
mailing list