Skip to contentSkip to conversation

Unaligned DIO read error path fix and clean ups

PATCH Superseded Gabriel Krisman Bertazi 2 participants 6 messages Sep 5, 2020 – Sep 7, 2020 VFS

lore
Share & tools
lore
#0 lore raw

v3 cover Unaligned DIO read error path fix and clean ups

This is v3 of Unaligned DIO read error path fix and clean ups. This version applies some small fixes to patch 1 suggested by Jan Kara (thank you!) and it was tested with xfstests aio group over f2fs and fio workloads.

Gabriel Krisman Bertazi (3):
direct-io: clean up error paths of do_blockdev_direct_IO
direct-io: don't force writeback for reads beyond EOF
direct-io: defer alignment check until after the EOF check

 fs/direct-io.c | 69 ++++++++++++++++++++++----------------------------
 1 file changed, 30 insertions(+), 39 deletions(-)
Signature · 1 line
2.28.0
#1 lore raw

PATCH v3 1/3 direct-io: clean up error paths of do_blockdev_direct_IO

In preparation to resort DIO checks, reduce code duplication of error handling in do_blockdev_direct_IO.

Changes since V1:
- Remove fail_dio_unlocked (Me)
- Ensure fail_dio won't call inode_unlock() for writes (Jan Kara)

---
 fs/direct-io.c | 35 ++++++++++++++---------------------
 1 file changed, 14 insertions(+), 21 deletions(-)
fs/direct-io.c +14 -21
index 183299892465..6c11db1cec27 100644
@@ -1170,7 +1170,7 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 			blkbits = blksize_bits(bdev_logical_block_size(bdev));
 		blocksize_mask = (1 << blkbits) - 1;
 		if (align & blocksize_mask)
-			goto out;
+			return -EINVAL;
 	}
 
 	/* watch out for a 0 len io from a tricksy fs */
@@ -1178,9 +1178,8 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 		return 0;
 
 	dio = kmem_cache_alloc(dio_cache, GFP_KERNEL);
-	retval = -ENOMEM;
 	if (!dio)
-		goto out;
+		return -ENOMEM;
 	/*
 	 * Believe it or not, zeroing out the page array caused a .5%
 	 * performance regression in a database benchmark.  So, we take
@@ -1199,22 +1198,16 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 
 			retval = filemap_write_and_wait_range(mapping, offset,
 							      end - 1);
-			if (retval) {
-				inode_unlock(inode);
-				kmem_cache_free(dio_cache, dio);
-				goto out;
-			}
+			if (retval)
+				goto fail_dio;
 		}
 	}
 
 	/* Once we sampled i_size check for reads beyond EOF */
 	dio->i_size = i_size_read(inode);
 	if (iov_iter_rw(iter) == READ && offset >= dio->i_size) {
-		if (dio->flags & DIO_LOCKING)
-			inode_unlock(inode);
-		kmem_cache_free(dio_cache, dio);
 		retval = 0;
-		goto out;
+		goto fail_dio;
 	}
 
 	/*
@@ -1258,14 +1251,8 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 			 */
 			retval = sb_init_dio_done_wq(dio->inode->i_sb);
 		}
-		if (retval) {
-			/*
-			 * We grab i_mutex only for reads so we don't have
-			 * to release it here
-			 */
-			kmem_cache_free(dio_cache, dio);
-			goto out;
-		}
+		if (retval)
+			goto fail_dio;
 	}
 
 	/*
@@ -1368,7 +1355,13 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 	} else
 		BUG_ON(retval != -EIOCBQUEUED);
 
-out:
+	return retval;
+
+fail_dio:
+	if (dio->flags & DIO_LOCKING && iov_iter_rw(iter) == READ)
+		inode_unlock(inode);
+
+	kmem_cache_free(dio_cache, dio);
 	return retval;
 }
 
Signature · 1 line
2.28.0
reviewer Reviewed-by on v3 1/3
#2 lore raw

Re: [PATCH v3 1/3] direct-io: clean up error paths of do_blockdev_direct_IO

Quoted 6 lines · Gabriel Krisman Bertazi

In preparation to resort DIO checks, reduce code duplication of error handling in do_blockdev_direct_IO.

Changes since V1:
- Remove fail_dio_unlocked (Me)
- Ensure fail_dio won't call inode_unlock() for writes (Jan Kara)

Please add the patch changelogs below the diffstat. That way they won't be in the final changelog (which is the right thing to do because they are mostly irrelevant for the final patch).

Otherwise the patch looks good to me so feel free to add:

Honza

Quoted 91 lines
---
 fs/direct-io.c | 35 ++++++++++++++---------------------
 1 file changed, 14 insertions(+), 21 deletions(-)
fs/direct-io.c +14 -21
index 183299892465..6c11db1cec27 100644
@@ -1170,7 +1170,7 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 			blkbits = blksize_bits(bdev_logical_block_size(bdev));
 		blocksize_mask = (1 << blkbits) - 1;
 		if (align & blocksize_mask)
-			goto out;
+			return -EINVAL;
 	}
 
 	/* watch out for a 0 len io from a tricksy fs */
@@ -1178,9 +1178,8 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 		return 0;
 
 	dio = kmem_cache_alloc(dio_cache, GFP_KERNEL);
-	retval = -ENOMEM;
 	if (!dio)
-		goto out;
+		return -ENOMEM;
 	/*
 	 * Believe it or not, zeroing out the page array caused a .5%
 	 * performance regression in a database benchmark.  So, we take
@@ -1199,22 +1198,16 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 
 			retval = filemap_write_and_wait_range(mapping, offset,
 							      end - 1);
-			if (retval) {
-				inode_unlock(inode);
-				kmem_cache_free(dio_cache, dio);
-				goto out;
-			}
+			if (retval)
+				goto fail_dio;
 		}
 	}
 
 	/* Once we sampled i_size check for reads beyond EOF */
 	dio->i_size = i_size_read(inode);
 	if (iov_iter_rw(iter) == READ && offset >= dio->i_size) {
-		if (dio->flags & DIO_LOCKING)
-			inode_unlock(inode);
-		kmem_cache_free(dio_cache, dio);
 		retval = 0;
-		goto out;
+		goto fail_dio;
 	}
 
 	/*
@@ -1258,14 +1251,8 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 			 */
 			retval = sb_init_dio_done_wq(dio->inode->i_sb);
 		}
-		if (retval) {
-			/*
-			 * We grab i_mutex only for reads so we don't have
-			 * to release it here
-			 */
-			kmem_cache_free(dio_cache, dio);
-			goto out;
-		}
+		if (retval)
+			goto fail_dio;
 	}
 
 	/*
@@ -1368,7 +1355,13 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 	} else
 		BUG_ON(retval != -EIOCBQUEUED);
 
-out:
+	return retval;
+
+fail_dio:
+	if (dio->flags & DIO_LOCKING && iov_iter_rw(iter) == READ)
+		inode_unlock(inode);
+
+	kmem_cache_free(dio_cache, dio);
 	return retval;
 }
 
Signature · 1 line
2.28.0
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#3 lore raw

PATCH v3 2/3 direct-io: don't force writeback for reads beyond EOF

In reply to Gabriel Krisman Bertazi

If a DIO read starts past EOF, the kernel won't attempt it, so we don't need to flush dirty pages before failing the syscall.

---
 fs/direct-io.c | 24 +++++++++++-------------
 1 file changed, 11 insertions(+), 13 deletions(-)
fs/direct-io.c +11 -13
index 6c11db1cec27..c17efe58f1c9 100644
@@ -1188,19 +1188,9 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 	memset(dio, 0, offsetof(struct dio, pages));
 
 	dio->flags = flags;
-	if (dio->flags & DIO_LOCKING) {
-		if (iov_iter_rw(iter) == READ) {
-			struct address_space *mapping =
-					iocb->ki_filp->f_mapping;
-
-			/* will be released by direct_io_worker */
-			inode_lock(inode);
-
-			retval = filemap_write_and_wait_range(mapping, offset,
-							      end - 1);
-			if (retval)
-				goto fail_dio;
-		}
+	if (dio->flags & DIO_LOCKING && iov_iter_rw(iter) == READ) {
+		/* will be released by direct_io_worker */
+		inode_lock(inode);
 	}
 
 	/* Once we sampled i_size check for reads beyond EOF */
@@ -1210,6 +1200,14 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 		goto fail_dio;
 	}
 
+	if (dio->flags & DIO_LOCKING && iov_iter_rw(iter) == READ) {
+		struct address_space *mapping = iocb->ki_filp->f_mapping;
+
+		retval = filemap_write_and_wait_range(mapping, offset, end - 1);
+		if (retval)
+			goto fail_dio;
+	}
+
 	/*
 	 * For file extending writes updating i_size before data writeouts
 	 * complete can expose uninitialized blocks in dumb filesystems.
Signature · 1 line
2.28.0
#4 lore raw

PATCH v3 3/3 direct-io: defer alignment check until after the EOF check

In reply to Gabriel Krisman Bertazi

Prior to commit 9fe55eea7e4b ("Fix race when checking i_size on direct i/o read"), an unaligned direct read past end of file would trigger EOF, since generic_file_aio_read detected this read-at-EOF condition and skipped the direct IO read entirely, returning 0. After that change, the read now reaches dio_generic, which detects the misalignment and returns EINVAL.

This consolidates the generic direct-io to follow the same behavior of filesystems. Apparently, this fix will only affect ocfs2 since other filesystems do this verification before calling do_blockdev_direct_IO, with the exception of f2fs, which has the same bug, but is fixed in the next patch.

it can be verified by a read loop on a file that does a partial read before EOF (On file that doesn't end at an aligned address). The following code fails on an unaligned file on filesystems without prior validation without this patch, but not on btrfs, ext4, and xfs.

  while (done < total) {
    ssize_t delta = pread(fd, buf + done, total - done, off + done);
    if (!delta)
      break;
    ...
  }

Fix this regression by moving the misalignment check to after the EOF check added by commit 74cedf9b6c60 ("direct-io: Fix negative return from dio read beyond eof").

Based on a patch by Jamie Liu.

---
 fs/direct-io.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)
fs/direct-io.c +8 -8
index c17efe58f1c9..82838cca934b 100644
@@ -1165,14 +1165,6 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 	 * the early prefetch in the caller enough time.
 	 */
 
-	if (align & blocksize_mask) {
-		if (bdev)
-			blkbits = blksize_bits(bdev_logical_block_size(bdev));
-		blocksize_mask = (1 << blkbits) - 1;
-		if (align & blocksize_mask)
-			return -EINVAL;
-	}
-
 	/* watch out for a 0 len io from a tricksy fs */
 	if (iov_iter_rw(iter) == READ && !count)
 		return 0;
@@ -1200,6 +1192,14 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 		goto fail_dio;
 	}
 
+	if (align & blocksize_mask) {
+		if (bdev)
+			blkbits = blksize_bits(bdev_logical_block_size(bdev));
+		blocksize_mask = (1 << blkbits) - 1;
+		if (align & blocksize_mask)
+			goto fail_dio;
+	}
+
 	if (dio->flags & DIO_LOCKING && iov_iter_rw(iter) == READ) {
 		struct address_space *mapping = iocb->ki_filp->f_mapping;
 
Signature · 1 line
2.28.0
reviewer Reviewed-by on v3 3/3
#5 lore raw

Re: [PATCH v3 3/3] direct-io: defer alignment check until after the EOF check

Quoted 25 lines · Gabriel Krisman Bertazi

Prior to commit 9fe55eea7e4b ("Fix race when checking i_size on direct i/o read"), an unaligned direct read past end of file would trigger EOF, since generic_file_aio_read detected this read-at-EOF condition and skipped the direct IO read entirely, returning 0. After that change, the read now reaches dio_generic, which detects the misalignment and returns EINVAL.

This consolidates the generic direct-io to follow the same behavior of filesystems. Apparently, this fix will only affect ocfs2 since other filesystems do this verification before calling do_blockdev_direct_IO, with the exception of f2fs, which has the same bug, but is fixed in the next patch.

it can be verified by a read loop on a file that does a partial read before EOF (On file that doesn't end at an aligned address). The following code fails on an unaligned file on filesystems without prior validation without this patch, but not on btrfs, ext4, and xfs.

  while (done < total) {
    ssize_t delta = pread(fd, buf + done, total - done, off + done);
    if (!delta)
      break;
    ...
  }
Quoted 8 lines

Fix this regression by moving the misalignment check to after the EOF check added by commit 74cedf9b6c60 ("direct-io: Fix negative return from dio read beyond eof").

Based on a patch by Jamie Liu.

The patch looks good to me. You can add:

Honza

Quoted 41 lines
---
 fs/direct-io.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)
fs/direct-io.c +8 -8
index c17efe58f1c9..82838cca934b 100644
@@ -1165,14 +1165,6 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 	 * the early prefetch in the caller enough time.
 	 */
 
-	if (align & blocksize_mask) {
-		if (bdev)
-			blkbits = blksize_bits(bdev_logical_block_size(bdev));
-		blocksize_mask = (1 << blkbits) - 1;
-		if (align & blocksize_mask)
-			return -EINVAL;
-	}
-
 	/* watch out for a 0 len io from a tricksy fs */
 	if (iov_iter_rw(iter) == READ && !count)
 		return 0;
@@ -1200,6 +1192,14 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 		goto fail_dio;
 	}
 
+	if (align & blocksize_mask) {
+		if (bdev)
+			blkbits = blksize_bits(bdev_logical_block_size(bdev));
+		blocksize_mask = (1 << blkbits) - 1;
+		if (align & blocksize_mask)
+			goto fail_dio;
+	}
+
 	if (dio->flags & DIO_LOCKING && iov_iter_rw(iter) == READ) {
 		struct address_space *mapping = iocb->ki_filp->f_mapping;
 
Signature · 1 line
2.28.0
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR