NFS: ALLOCATE and DEALLOCATE improvemnts
Subsystems (2)
v4 cover NFS: ALLOCATE and DEALLOCATE improvemnts
These patches make a few tweaks that improve the performance of ALLOCATE and DEALLOCATE requests. The first patch attempts to update cache information after a (DE)ALLOCATE call, rather than clearing the cache and forcing the client to re-read the file later. The second patch changes when we lock the i_mutex to give NFS v4.0 and v4.1 a chance to exit without taking a lock.
Comments are appreciated!
Anna
Anna Schumaker (2):
NFS: Don't zap caches on fallocate()
NFS: Reduce time spent holding the i_mutex during fallocate()
fs/nfs/inode.c | 1 - fs/nfs/nfs42proc.c | 31 +++++++++++++++++++++++++++---- fs/nfs/nfs42xdr.c | 20 ++++++++++++++++---- fs/nfs/nfs4file.c | 10 ++-------- include/linux/nfs_xdr.h | 4 ++++ 5 files changed, 49 insertions(+), 17 deletions(-)
Signature · 1 line
2.3.3
PATCH v4 1/2 NFS: Don't zap caches on fallocate()
This patch adds a GETATTR to the end of ALLOCATE and DEALLOCATE operations so we can set the updated inode size and change attribute directly. DEALLOCATE will still need to release pagecache pages, so nfs42_proc_deallocate() now calls truncate_pagecache_range() before contacting the server.
--- fs/nfs/inode.c | 1 - fs/nfs/nfs42proc.c | 23 +++++++++++++++++++---- fs/nfs/nfs42xdr.c | 20 ++++++++++++++++---- fs/nfs/nfs4file.c | 1 - include/linux/nfs_xdr.h | 4 ++++ 5 files changed, 39 insertions(+), 10 deletions(-)
fs/nfs/inode.c +0 -1
@@ -192,7 +192,6 @@ void nfs_zap_caches(struct inode *inode)nfs_zap_caches_locked (inode);spin_unlock (&inode->i_lock); } -EXPORT_SYMBOL_GPL (nfs_zap_caches);void nfs_zap_mapping (struct inode *inode,struct address_space *mapping) {
fs/nfs/nfs42proc.c +19 -4
@@ -36,13 +36,16 @@ static int _nfs42_proc_fallocate(struct rpc_message *msg, struct file *filep, loff_t offset, loff_t len) {struct inode *inode =file_inode (filep); +struct nfs_server *server =NFS_SERVER (inode);struct nfs42_falloc_args args = { .falloc_fh =NFS_FH (inode), .falloc_offset = offset, .falloc_length = len, + .falloc_bitmask = server->cache_consistency_bitmask, + }; +struct nfs42_falloc_res res = { + .falloc_server = server, }; -struct nfs42_falloc_res res; -struct nfs_server *server =NFS_SERVER (inode);int status; msg->rpc_argp = &args;
@@ -52,8 +55,17 @@ static int _nfs42_proc_fallocate(struct rpc_message *msg, struct file *filep,if (status)return status; -return nfs4_call_sync (server->client, server, msg, - &args.seq_args, &res.seq_res,0 ); + res.falloc_fattr =nfs_alloc_fattr (); +if (!res.falloc_fattr) +return -ENOMEM ; + + status =nfs4_call_sync (server->client, server, msg, + &args.seq_args, &res.seq_res,0 ); +if (status ==0 ) + status =nfs_post_op_update_inode (inode, res.falloc_fattr); + +kfree (res.falloc_fattr); +return status; } static int nfs42_proc_fallocate(struct rpc_message *msg, struct file *filep,
@@ -101,7 +113,10 @@ int nfs42_proc_deallocate(struct file *filep, loff_t offset, loff_t len)if (!nfs_server_capable (inode,NFS_CAP_DEALLOCATE ))return -EOPNOTSUPP ; +nfs_wb_all (inode); err =nfs42_proc_fallocate (&msg, filep, offset, len); +if (err ==0 ) +truncate_pagecache_range (inode, offset, (offset + len) -1 );if (err == -EOPNOTSUPP )NFS_SERVER (inode)->caps &= ~NFS_CAP_DEALLOCATE ;return err;
fs/nfs/nfs42xdr.c +16 -4
@@ -25,16 +25,20 @@#define NFS4_enc_allocate_sz (compound_encode_hdr_maxsz + \ encode_putfh_maxsz + \ - encode_allocate_maxsz) + encode_allocate_maxsz + \ + encode_getattr_maxsz)#define NFS4_dec_allocate_sz (compound_decode_hdr_maxsz + \ decode_putfh_maxsz + \ - decode_allocate_maxsz) + decode_allocate_maxsz + \ + decode_getattr_maxsz)#define NFS4_enc_deallocate_sz (compound_encode_hdr_maxsz + \ encode_putfh_maxsz + \ - encode_deallocate_maxsz) + encode_deallocate_maxsz + \ + encode_getattr_maxsz)#define NFS4_dec_deallocate_sz (compound_decode_hdr_maxsz + \ decode_putfh_maxsz + \ - decode_deallocate_maxsz) + decode_deallocate_maxsz + \ + decode_getattr_maxsz)#define NFS4_enc_seek_sz (compound_encode_hdr_maxsz + \ encode_putfh_maxsz + \ encode_seek_maxsz)
@@ -92,6 +96,7 @@ static void nfs4_xdr_enc_allocate(struct rpc_rqst *req,encode_sequence (xdr, &args->seq_args, &hdr);encode_putfh (xdr, args->falloc_fh, &hdr);encode_allocate (xdr, args, &hdr); +encode_getfattr (xdr, args->falloc_bitmask, &hdr);encode_nops (&hdr); }
@@ -110,6 +115,7 @@ static void nfs4_xdr_enc_deallocate(struct rpc_rqst *req,encode_sequence (xdr, &args->seq_args, &hdr);encode_putfh (xdr, args->falloc_fh, &hdr);encode_deallocate (xdr, args, &hdr); +encode_getfattr (xdr, args->falloc_bitmask, &hdr);encode_nops (&hdr); }
@@ -183,6 +189,9 @@ static int nfs4_xdr_dec_allocate(struct rpc_rqst *rqstp,if (status) goto out; status =decode_allocate (xdr, res); +if (status) + goto out; +decode_getfattr (xdr, res->falloc_fattr, res->falloc_server); out:return status; }
@@ -207,6 +216,9 @@ static int nfs4_xdr_dec_deallocate(struct rpc_rqst *rqstp,if (status) goto out; status =decode_deallocate (xdr, res); +if (status) + goto out; +decode_getfattr (xdr, res->falloc_fattr, res->falloc_server); out:return status; }
fs/nfs/nfs4file.c +0 -1
@@ -159,7 +159,6 @@ static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t ret =nfs42_proc_allocate (filep, offset, len);mutex_unlock (&inode->i_mutex); -nfs_zap_caches (inode); return ret; }#endif /* CONFIG_NFS_V4_2 */
include/linux/nfs_xdr.h +4 -0
@@ -1273,11 +1273,15 @@ struct nfs42_falloc_args {nfs4_stateid falloc_stateid;u64 falloc_offset;u64 falloc_length; +const u32 *falloc_bitmask; };struct nfs42_falloc_res {struct nfs4_sequence_res seq_res;unsigned int status; + +struct nfs_fattr *falloc_fattr; +const struct nfs_server *falloc_server; };struct nfs42_seek_args {
Signature · 1 line
2.3.3
PATCH v4 2/2 NFS: Reduce time spent holding the i_mutex during fallocate()
In reply to Anna SchumakerAt the very least, we should not be taking the i_mutex until after checking if the server even supports ALLOCATE or DEALLOCATE, allowing v4.0 or v4.1 to exit without potentially waiting on a lock.
--- fs/nfs/nfs42proc.c | 8 ++++++++ fs/nfs/nfs4file.c | 9 ++------- 2 files changed, 10 insertions(+), 7 deletions(-)
fs/nfs/nfs42proc.c +8 -0
@@ -96,9 +96,13 @@ int nfs42_proc_allocate(struct file *filep, loff_t offset, loff_t len)if (!nfs_server_capable (inode,NFS_CAP_ALLOCATE ))return -EOPNOTSUPP ; +mutex_lock (&inode->i_mutex); + err =nfs42_proc_fallocate (&msg, filep, offset, len);if (err == -EOPNOTSUPP )NFS_SERVER (inode)->caps &= ~NFS_CAP_ALLOCATE ; + +mutex_unlock (&inode->i_mutex);return err; }
@@ -114,11 +118,15 @@ int nfs42_proc_deallocate(struct file *filep, loff_t offset, loff_t len)return -EOPNOTSUPP ;nfs_wb_all (inode); +mutex_lock (&inode->i_mutex); + err =nfs42_proc_fallocate (&msg, filep, offset, len);if (err ==0 )truncate_pagecache_range (inode, offset, (offset + len) -1 );if (err == -EOPNOTSUPP )NFS_SERVER (inode)->caps &= ~NFS_CAP_DEALLOCATE ; + +mutex_unlock (&inode->i_mutex);return err; }
fs/nfs/nfs4file.c +2 -7
@@ -152,14 +152,9 @@ static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_tif (ret <0 )return ret; -mutex_lock (&inode->i_mutex);if (mode &FALLOC_FL_PUNCH_HOLE ) - ret =nfs42_proc_deallocate (filep, offset, len); -else - ret =nfs42_proc_allocate (filep, offset, len); -mutex_unlock (&inode->i_mutex); - - return ret; +return nfs42_proc_deallocate (filep, offset, len); + return nfs42_proc_allocate(filep, offset, len); }#endif /* CONFIG_NFS_V4_2 */
Signature · 1 line
2.3.3