Skip to contentSkip to conversation

NFS: Add FATTR4_WORD1_SPACE_USED to the cache_consistency_bitmask

PATCH In review [email protected] 2 participants 4 messages Nov 8, 2019 – Nov 11, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 NFS: Add FATTR4_WORD1_SPACE_USED to the cache_consistency_bitmask

From: Anna Schumaker <[email protected]>

Changing a sparse file could have an effect not only on the file size, but also on the number of blocks used by the file in the underlying filesystem. Let's update the SPACE_USED attribute whenever we update SIZE to be as accurate as possible.

This patch fixes xfstests generic/568, which tests that fallocating an unaligned range allocates all blocks touched by that range. Without this patch, stat reports 0 bytes used immediately after the fallocate. Adding a sleep 5 to the test also catches the update, but it's better to just do it when we know something has changed.

---
 fs/nfs/nfs4proc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/nfs/nfs4proc.c +1 -1
index ac9063c06205..00a1f3ec7f22 100644
@@ -3775,7 +3775,7 @@ static int _nfs4_server_capabilities(struct nfs_server *server, struct nfs_fh *f
 
 		memcpy(server->cache_consistency_bitmask, res.attr_bitmask, sizeof(server->cache_consistency_bitmask));
 		server->cache_consistency_bitmask[0] &= FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE;
-		server->cache_consistency_bitmask[1] &= FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY;
+		server->cache_consistency_bitmask[1] &= FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY|FATTR4_WORD1_SPACE_USED;
 		server->cache_consistency_bitmask[2] = 0;
 
 		/* Avoid a regression due to buggy server */
Signature · 1 line
2.24.0
#1 lore raw

PATCH 2/2 NFS: Return -ETXTBSY when attempting to write to a swapfile

From: Anna Schumaker <[email protected]>

My understanding is that -EBUSY refers to the underlying device, and that -ETXTBSY is used when attempting to access a file in use by the kernel (like a swapfile). Changing this return code helps us pass xfstests generic/569

---
 fs/nfs/file.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/nfs/file.c +1 -1
index 95dc90570786..8eb731d9be3e 100644
@@ -649,7 +649,7 @@ ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
 
 out_swapfile:
 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
-	return -EBUSY;
+	return -ETXTBSY;
 }
 EXPORT_SYMBOL_GPL(nfs_file_write);
 
Signature · 1 line
2.24.0
#2 lore raw
In reply to [email protected]
Quoted 36 lines · On Fri, 2019-11-08 at 16:02 -0500, [email protected] wrote:

From: Anna Schumaker <[email protected]>

Changing a sparse file could have an effect not only on the file size,
but also on the number of blocks used by the file in the underlying filesystem. Let's update the SPACE_USED attribute whenever we update SIZE to be as accurate as possible.

This patch fixes xfstests generic/568, which tests that fallocating an
unaligned range allocates all blocks touched by that range. Without this
patch, stat reports 0 bytes used immediately after the fallocate. Adding a sleep 5 to the test also catches the update, but it's better
to just do it when we know something has changed.

---
 fs/nfs/nfs4proc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/nfs/nfs4proc.c +0 -0
index ac9063c06205..00a1f3ec7f22 100644
@@ -3775,7 +3775,7 @@ static int _nfs4_server_capabilities(struct

nfs_server *server, struct nfs_fh *f

memcpy(server->cache_consistency_bitmask,
res.attr_bitmask, sizeof(server->cache_consistency_bitmask));
server->cache_consistency_bitmask[0] &=
FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE;

fs/nfs/nfs4proc.c +0 -1
-		server->cache_consistency_bitmask[1] &=

FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY;

fs/nfs/nfs4proc.c +1 -0
+		server->cache_consistency_bitmask[1] &=
Quoted 2 lines

FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY|FATTR4_WORD1_SPAC E_USED;

I'd rather not do this. Space used is not a cache consistency attribute , as we do not use it to revalidate the cache and it can be rather expensive to retrieve on some platforms.

I'd therefore prefer that we just make sure we mark the cache validity with NFS_INO_INVALID_OTHER when we have a write succeed on a sparse file.

Quoted 3 lines
fs/nfs/nfs4proc.c +0 -0
 		server->cache_consistency_bitmask[2] = 0;
 
 		/* Avoid a regression due to buggy server */
Signature · 3 lines
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]
#3 lore raw

Hi Trond,

Quoted 30 lines · Trond Myklebust
Quoted 29 lines · On Fri, 2019-11-08 at 16:02 -0500, [email protected] wrote:

From: Anna Schumaker <[email protected]>

Changing a sparse file could have an effect not only on the file size,
but also on the number of blocks used by the file in the underlying filesystem. Let's update the SPACE_USED attribute whenever we update SIZE to be as accurate as possible.

This patch fixes xfstests generic/568, which tests that fallocating an
unaligned range allocates all blocks touched by that range. Without this
patch, stat reports 0 bytes used immediately after the fallocate. Adding a sleep 5 to the test also catches the update, but it's better
to just do it when we know something has changed.

---
 fs/nfs/nfs4proc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/nfs/nfs4proc.c +0 -0
index ac9063c06205..00a1f3ec7f22 100644
@@ -3775,7 +3775,7 @@ static int _nfs4_server_capabilities(struct

nfs_server *server, struct nfs_fh *f

Quoted 13 lines
Quoted 9 lines

memcpy(server->cache_consistency_bitmask,
res.attr_bitmask, sizeof(server->cache_consistency_bitmask));
server->cache_consistency_bitmask[0] &=
FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE;

fs/nfs/nfs4proc.c +0 -1
-		server->cache_consistency_bitmask[1] &=

FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY;

fs/nfs/nfs4proc.c +1 -0
+		server->cache_consistency_bitmask[1] &=

FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY|FATTR4_WORD1_SPAC E_USED;

I'd rather not do this. Space used is not a cache consistency attribute , as we do not use it to revalidate the cache and it can be rather expensive to retrieve on some platforms.

Okay, that makes sense.

Quoted 4 lines

I'd therefore prefer that we just make sure we mark the cache validity with NFS_INO_INVALID_OTHER when we have a write succeed on a sparse file.

It looks like setting falloc_bitmask to use the standard nfs4_fattr_bitmap instead of the cache_consistency_bitmask will also update the SPACE_USED attribute after a fallocate call.

I'll play around with NFS_INO_INVALID_OTHER next, but making use of the trailing getattr that ALLOCATE and DEALLOCATE already have seems more straightforward to me.

Anna

Quoted 10 lines
Quoted 4 lines
fs/nfs/nfs4proc.c +0 -0
 		server->cache_consistency_bitmask[2] = 0;
 
 		/* Avoid a regression due to buggy server */
Signature · 3 lines
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]