Set BLOCK_BIT_ERROR on bio submit failure during forced unmount

block_submit_bio will return -ENOLINK if called during a forced
shutdown, the bio is never submitted, and thus no completion callback
will fire to set BLOCK_BIT_ERROR. Any other task waiting for this
specific bp will end up waiting forever.

To fix, fall through to the existing block_end_io call on the
error path instead of returning directly.  That means moving
the forcing_unmount check past the setup calls so block_end_io's
bookkeeping stays balanced. block_end_io then sets BLOCK_BIT_ERROR
and wakes up waiters just as it would on a failed async completion.

Signed-off-by: Auke Kok <auke.kok@versity.com>

kmod/src/block.c

@@ -467,9 +467,6 @@ static int block_submit_bio(struct super_block *sb, struct block_private *bp,
 	sector_t sector;
 	int ret = 0;
 
-	if (scoutfs_forcing_unmount(sb))
-		return -ENOLINK;
-
 	sector = bp->bl.blkno << (SCOUTFS_BLOCK_LG_SHIFT - 9);
 
 	WARN_ON_ONCE(bp->bl.blkno == U64_MAX);
@@ -480,6 +477,17 @@ static int block_submit_bio(struct super_block *sb, struct block_private *bp,
 	set_bit(BLOCK_BIT_IO_BUSY, &bp->bits);
 	block_get(bp);
 
+	/*
+	 * A second thread may already be waiting on this block's completion
+	 * after this thread won the race to submit the block.  We exit through
+	 * the block_end_io error path which sets BLOCK_BIT_ERROR and assures
+	 * that other callers in the waitq get woken up.
+	 */
+	if (scoutfs_forcing_unmount(sb)) {
+		ret = -ENOLINK;
+		goto end_io;
+	}
+
 	blk_start_plug(&plug);
 
 	for (off = 0; off < SCOUTFS_BLOCK_LG_SIZE; off += PAGE_SIZE) {
@@ -517,6 +525,7 @@ static int block_submit_bio(struct super_block *sb, struct block_private *bp,
 
 	blk_finish_plug(&plug);
 
+end_io:
 	/* let racing end_io know we're done */
 	block_end_io(sb, opf, bp, ret);
 

kmod/src/data.c

@@ -384,12 +384,6 @@ static inline u64 ext_last(struct scoutfs_extent *ext)
  * This can waste a lot of space for small or sparse files but is
  * reasonable when a file population is known to be large and dense but
  * known to be written with non-streaming write patterns.
- *
- * In either strategy, preallocation ramps up proportionally with the
- * file's online block count rather than jumping to the full prealloc
- * size.  This avoids overallocation for small files in mixed-size
- * workloads while still allowing large files to benefit from full
- * preallocation.
  */
 static int alloc_block(struct super_block *sb, struct inode *inode,
 		       struct scoutfs_extent *ext, u64 iblock,
@@ -406,7 +400,6 @@ static int alloc_block(struct super_block *sb, struct inode *inode,
 	struct scoutfs_extent found;
 	struct scoutfs_extent pre = {0,};
 	bool undo_pre = false;
-	bool have_onoff = false;
 	u64 blkno = 0;
 	u64 online;
 	u64 offline;
@@ -452,7 +445,6 @@ static int alloc_block(struct super_block *sb, struct inode *inode,
 		 * blocks.
 		 */
 		scoutfs_inode_get_onoff(inode, &online, &offline);
-		have_onoff = true;
 		if (iblock > 1 && iblock == online) {
 			ret = scoutfs_ext_next(sb, &data_ext_ops, &args,
 					       iblock, 1, &found);
@@ -499,16 +491,6 @@ static int alloc_block(struct super_block *sb, struct inode *inode,
 	/* overall prealloc limit */
 	count = min_t(u64, count, opts.data_prealloc_blocks);
 
-	/*
-	 * Ramp preallocation up proportionally with the file's online
-	 * block count rather than jumping to the full prealloc size.
-	 */
-	if (!ext->len) {
-		if (!have_onoff)
-			scoutfs_inode_get_onoff(inode, &online, &offline);
-		count = max_t(u64, 1, min(count, online));
-	}
-
 	ret = scoutfs_alloc_data(sb, datinf->alloc, datinf->wri,
 				 &datinf->dalloc, count, &blkno, &count);
 	if (ret < 0)

tests/golden/data-prealloc

@@ -8,10 +8,10 @@
 /mnt/test/test/data-prealloc/file-1: extents: 32
 /mnt/test/test/data-prealloc/file-2: extents: 32
 == any writes to region prealloc get full extents
-/mnt/test/test/data-prealloc/file-1: extents: 8
-/mnt/test/test/data-prealloc/file-2: extents: 8
-/mnt/test/test/data-prealloc/file-1: extents: 8
-/mnt/test/test/data-prealloc/file-2: extents: 8
+/mnt/test/test/data-prealloc/file-1: extents: 4
+/mnt/test/test/data-prealloc/file-2: extents: 4
+/mnt/test/test/data-prealloc/file-1: extents: 4
+/mnt/test/test/data-prealloc/file-2: extents: 4
 == streaming offline writes get full extents either way
 /mnt/test/test/data-prealloc/file-1: extents: 4
 /mnt/test/test/data-prealloc/file-2: extents: 4
@@ -20,8 +20,8 @@
 == goofy preallocation amounts work
 /mnt/test/test/data-prealloc/file-1: extents: 6
 /mnt/test/test/data-prealloc/file-2: extents: 6
-/mnt/test/test/data-prealloc/file-1: extents: 10
-/mnt/test/test/data-prealloc/file-2: extents: 10
+/mnt/test/test/data-prealloc/file-1: extents: 6
+/mnt/test/test/data-prealloc/file-2: extents: 6
 /mnt/test/test/data-prealloc/file-1: extents: 3
 /mnt/test/test/data-prealloc/file-2: extents: 3
 == block writes into region allocs hole