RFC: pnfs: LAYOUTGET/DEVINFO submission, v2
Subsystems (3)
cover RFC: pnfs: LAYOUTGET/DEVINFO submission, v2
This fixes various issue brought up in comments to the previous version.
Among the more substantial changes:
find_pnfs_driver now grabs module reference
driver init/uninit_mountpoint argument changed to struct nfs_server io_ops moved into pnfs_layoutdriver_type
handling of pnfs_layout_from_open_stateid changed
file driver no longer uses vmalloc
deviceid moved from general to driver layer CONFIG_PNFS_FILE_LAYOUT removed, just uses CONFIG_NFS_V4_1 error handling of LAYOUTGET and GETDEVICEINFO rpc calls changed
For those interested, the patches can also be found atgit://linux-nfs.org/~iisaman/linux-pnfs.git,
under the branch devinfo-submit-rc4-trond
A rebase of Benny's pnfs-submit branch onto these patches is under the branch bennys-pnfs-submit-on-devinfo-rc4
Fred
This is the start of code implementing pnfs, based on RFC 5661. Since sending the whole thing at once would be overwhelming, we are trying to break it into bite sized chunks. This chunk implements the mount/umount infrastructure, as well as sending the LAYOUTGET and GETDEVTICEINFO calls on io (but not actually using the information for io). Note that two major simplifications to the protocol will be made throughout the initial submission process: only the file layout driver is considered, and only whole file layouts are requested.
These patches apply against Trond's nfs-for-2.6.37 branch.
patches 01-08 implement the mount/umount hooks patches 09-12 implement LAYOUTGET and GETDEVICEINFO
PATCH 1/12 NFSD: remove duplicate NFS4_STATEID_SIZE
From: Andy Adamson <[email protected]>
Already accepted by Bruce
--- fs/nfsd/nfs4callback.c | 1 - 1 files changed, 0 insertions(+), 1 deletions(-)
fs/nfsd/nfs4callback.c +0 -1
@@ -41,7 +41,6 @@#define NFSPROC4_CB_NULL 0#define NFSPROC4_CB_COMPOUND 1 -#define NFS4_STATEID_SIZE 16/* Index of predefined Linux callback client operations */
Signature · 1 line
1.7.2.1
PATCH 2/12 SUNRPC: define xdr_decode_opaque_fixed
In reply to Fred IsamanFrom: Benny Halevy <[email protected]>
A helper for decoding a fixed length opaque value. Returns a pointer to the next item in the xdr stream.
--- include/linux/sunrpc/xdr.h | 7 +++++++ 1 files changed, 7 insertions(+), 0 deletions(-)
include/linux/sunrpc/xdr.h +7 -0
@@ -131,6 +131,13 @@ xdr_decode_hyper(__be32 *p, __u64 *valp)return p +2 ; } +static inline __be32 * +xdr_decode_opaque_fixed (__be32 *p,void *ptr,unsigned int len) +{ +memcpy (ptr, p, len); +return p +XDR_QUADLEN (len); +} +/* * Adjust kvec to reflect end of xdr'ed data (RPC client XDR) */
Signature · 1 line
1.7.2.1
PATCH 3/12 RFC: pnfsd, pnfs: protocol level pnfs constants
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
Use only layoutreturn constant for both returns and recalls. (return_* works better for recall_type rather the other way around)
--- include/linux/nfs4.h | 39 +++++++++++++++++++++++++++++++++++++++ 1 files changed, 39 insertions(+), 0 deletions(-)
include/linux/nfs4.h +39 -0
@@ -471,6 +471,8 @@ enum lock_type4 {#define FATTR4_WORD1_TIME_MODIFY (1UL << 21)#define FATTR4_WORD1_TIME_MODIFY_SET (1UL << 22)#define FATTR4_WORD1_MOUNTED_ON_FILEID (1UL << 23) +#define FATTR4_WORD1_FS_LAYOUT_TYPES (1UL << 30) +#define FATTR4_WORD2_LAYOUT_BLKSIZE (1UL << 1)#define NFSPROC4_NULL 0#define NFSPROC4_COMPOUND 1
@@ -550,6 +552,43 @@ enum state_protect_how4 {SP4_SSV =2 }; +enum pnfs_layouttype { +LAYOUT_NFSV4_1_FILES =1 , +LAYOUT_OSD2_OBJECTS =2 , +LAYOUT_BLOCK_VOLUME =3 , +}; + +/* used for both layout return and recall */ +enum pnfs_layoutreturn_type { +RETURN_FILE =1 , +RETURN_FSID =2 , +RETURN_ALL =3 +}; + +enum pnfs_iomode { +IOMODE_READ =1 , +IOMODE_RW =2 , +IOMODE_ANY =3 , +}; + +enum pnfs_notify_deviceid_type4 { +NOTIFY_DEVICEID4_CHANGE =1 <<1 , +NOTIFY_DEVICEID4_DELETE =1 <<2 , +}; + +#define NFL4_UFLG_MASK 0x0000003F +#define NFL4_UFLG_DENSE 0x00000001 +#define NFL4_UFLG_COMMIT_THRU_MDS 0x00000002 +#define NFL4_UFLG_STRIPE_UNIT_SIZE_MASK 0xFFFFFFC0 + +/* Encoded in the loh_body field of type layouthint4 */ +enum filelayout_hint_care4 { +NFLH4_CARE_DENSE =NFL4_UFLG_DENSE , +NFLH4_CARE_COMMIT_THRU_MDS =NFL4_UFLG_COMMIT_THRU_MDS , +NFLH4_CARE_STRIPE_UNIT_SIZE =0x00000040 , +NFLH4_CARE_STRIPE_COUNT =0x00000080 +}; +#endif #endif
Signature · 1 line
1.7.2.1
PATCH 4/12 RFC: nfs: change stateid to be a union
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
In NFSv4.1 the stateid consists of the other and seqid fields. For layout processing we need to numerically compare the seqid value of layout stateids. To do so, introduce a union to nfs4_stateid to swtich between opaque(16 bytes) and opaque(12 bytes) / __be32
--- fs/nfs/callback_proc.c | 8 ++++---- include/linux/nfs4.h | 15 +++++++++++++-- 2 files changed, 17 insertions(+), 6 deletions(-)
fs/nfs/callback_proc.c +4 -4
@@ -118,11 +118,11 @@ int nfs41_validate_delegation_stateid(struct nfs_delegation *delegation, const nif (delegation ==NULL )return 0 ; -/* seqid is 4-bytes long */ -if (((u32 *) &stateid->data)[0 ] !=0 ) +if (stateid->stateid.seqid !=0 )return 0 ; -if (memcmp (&delegation->stateid.data[4 ], &stateid->data[4 ], -sizeof (stateid->data)-4 )) +if (memcmp (&delegation->stateid.stateid.other, + &stateid->stateid.other, +NFS4_STATEID_OTHER_SIZE ))return 0 ;return 1 ;
include/linux/nfs4.h +13 -2
@@ -17,7 +17,9 @@#define NFS4_BITMAP_SIZE 2#define NFS4_VERIFIER_SIZE 8 -#define NFS4_STATEID_SIZE 16 +#define NFS4_STATEID_SEQID_SIZE 4 +#define NFS4_STATEID_OTHER_SIZE 12 +#define NFS4_STATEID_SIZE (NFS4_STATEID_SEQID_SIZE + NFS4_STATEID_OTHER_SIZE)#define NFS4_FHSIZE 128#define NFS4_MAXPATHLEN PATH_MAX#define NFS4_MAXNAMLEN NAME_MAX
@@ -167,7 +169,16 @@ struct nfs4_acl { };typedef struct {char data[NFS4_VERIFIER_SIZE ]; }nfs4_verifier ; -typedef struct {char data[NFS4_STATEID_SIZE ]; }nfs4_stateid ; + +struct nfs41_stateid { +__be32 seqid; +char other[NFS4_STATEID_OTHER_SIZE ]; +} __attribute__ ((packed)); + +typedef union { +char data[NFS4_STATEID_SIZE ]; +struct nfs41_stateid stateid; +}nfs4_stateid ;enum nfs_opnum4 {OP_ACCESS =3 ,
Signature · 1 line
1.7.2.1
PATCH 5/12 RFC: nfs: ask for layouttypes during fsinfo call
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
This information will be used to determine which layout driver, if any, to use for subsequent IO on this filesystem. Each driver is assigned an integer id, with 0 reserved to indicate no driver.
The server can in theory return multiple ids. However, our current client implementation only notes the first entry and ignores the rest.
--- fs/nfs/nfs4proc.c | 2 +- fs/nfs/nfs4xdr.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++ include/linux/nfs_xdr.h | 1 + 3 files changed, 59 insertions(+), 1 deletions(-)
fs/nfs/nfs4proc.c +1 -1
@@ -129,7 +129,7 @@ const u32 nfs4_fsinfo_bitmap[2] = { FATTR4_WORD0_MAXFILESIZE | FATTR4_WORD0_MAXREAD | FATTR4_WORD0_MAXWRITE | FATTR4_WORD0_LEASE_TIME, - 0 + FATTR4_WORD1_FS_LAYOUT_TYPES };const u32 nfs4_fs_locations_bitmap[2 ] = {
fs/nfs/nfs4xdr.c +57 -0
69 lines not on this page · raw message
include/linux/nfs_xdr.h +1 -0
@@ -113,6 +113,7 @@ struct nfs_fsinfo {__u32 dtpref;/* pref. readdir transfer size */ __u64 maxfilesize;__u32 lease_time;/* in seconds */ +__u32 layouttype;/* supported pnfs layout driver */ };struct nfs_fsstat {
Signature · 1 line
1.7.2.1
Re: [PATCH 05/12] RFC: nfs: ask for layouttypes during fsinfo call
Quoted 22 lines · Fred Isaman
From: The pNFS Team <[email protected]>
This information will be used to determine which layout driver, if any, to use for subsequent IO on this filesystem. Each driver is assigned an integer id, with 0 reserved to indicate no driver.
The server can in theory return multiple ids. However, our current client implementation only notes the first entry and ignores the rest.
Signed-off-by TBD - melding/reorganization of several patches--- fs/nfs/nfs4proc.c | 2 +- fs/nfs/nfs4xdr.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++ include/linux/nfs_xdr.h | 1 + 3 files changed, 59 insertions(+), 1 deletions(-)fs/nfs/nfs4proc.c +0
-0
fs/nfs/nfs4proc.c const u32 nfs4_fsinfo_bitmap[2] = { FATTR4_WORD0_MAXFILESIZE · line 132
Quoted 5 lines
fs/nfs/nfs4proc.c +1
-1@@ -129,7 +129,7 @@ const u32 nfs4_fsinfo_bitmap[2] = { FATTR4_WORD0_MAXFILESIZE | FATTR4_WORD0_MAXREAD |FATTR4_WORD0_MAXWRITE |FATTR4_WORD0_LEASE_TIME , -0 +FATTR4_WORD1_FS_LAYOUT_TYPES
Note that originally (9170cf5 pnfs_submit: set and unset pnfs layoutdriver modules), we set this only under CONFIG_NFS_V4_1, but it should be ok to ask for it anyhow.
Benny
Quoted 90 lines
fs/nfs/nfs4proc.c +0
-0};const u32 nfs4_fs_locations_bitmap[2 ] = {fs/nfs/nfs4xdr.c +57
-0@@ -3868,6 +3868,60 @@ xdr_error:return status; } +/* +* Decode potentially multiple layout types. Currently we only support +* one layout driver per file system. +*/ +static int decode_first_pnfs_layout_type (struct xdr_stream *xdr, +uint32_t *layouttype) +{ +uint32_t *p; +int num; + + p =xdr_inline_decode (xdr,4 ); +if (unlikely (!p)) + goto out_overflow; + num =be32_to_cpup (p); + +/* pNFS is not supported by the underlying file system */ +if (num ==0 ) { + *layouttype =0 ; +return 0 ; + } +if (num >1 ) +printk (KERN_INFO "%s: Warning: Multiple pNFS layout drivers " +"per filesystem not supported\n" , __func__); + +/* Decode and set first layout type, move xdr->p past unused types */ + p =xdr_inline_decode (xdr, num *4 ); +if (unlikely (!p)) + goto out_overflow; + *layouttype =be32_to_cpup (p); +return 0 ; +out_overflow: +print_overflow_msg (__func__, xdr); +return -EIO ; +} + +/* +* The type of file system exported. +* Note we must ensure that layouttype is set in any non-error case. +*/ +static int decode_attr_pnfstype (struct xdr_stream *xdr,uint32_t *bitmap, +uint32_t *layouttype) +{ +int status =0 ; + +dprintk ("%s: bitmap is %x\n" , __func__, bitmap[1 ]); +if (unlikely (bitmap[1 ] & (FATTR4_WORD1_FS_LAYOUT_TYPES -1U ))) +return -EIO ; +if (bitmap[1 ] &FATTR4_WORD1_FS_LAYOUT_TYPES ) { + status =decode_first_pnfs_layout_type (xdr, layouttype); + bitmap[1 ] &= ~FATTR4_WORD1_FS_LAYOUT_TYPES ; + }else + *layouttype =0 ; +return status; +}static int decode_fsinfo (struct xdr_stream *xdr,struct nfs_fsinfo *fsinfo) {@@ -3894,6 +3948,9 @@ static int decode_fsinfo(struct xdr_stream *xdr, struct nfs_fsinfo *fsinfo)if ((status =decode_attr_maxwrite (xdr, bitmap, &fsinfo->wtmax)) !=0 ) goto xdr_error; fsinfo->wtpref = fsinfo->wtmax; + status =decode_attr_pnfstype (xdr, bitmap, &fsinfo->layouttype); +if (status) + goto xdr_error; status =verify_attr_len (xdr, savep, attrlen); xdr_error:include/linux/nfs_xdr.h +1
-0@@ -113,6 +113,7 @@ struct nfs_fsinfo {__u32 dtpref;/* pref. readdir transfer size */ __u64 maxfilesize;__u32 lease_time;/* in seconds */ +__u32 layouttype;/* supported pnfs layout driver */ };struct nfs_fsstat {
Re: [PATCH 05/12] RFC: nfs: ask for layouttypes during fsinfo call
Quoted 33 lines · Benny Halevy
Quoted 25 lines · Fred Isaman
From: The pNFS Team <[email protected]>
This information will be used to determine which layout driver, if any, to use for subsequent IO on this filesystem. Each driver is assigned an integer id, with 0 reserved to indicate no driver.
The server can in theory return multiple ids. However, our current client implementation only notes the first entry and ignores the rest.
Signed-off-by TBD - melding/reorganization of several patches---
fs/nfs/nfs4proc.c | 2 +-
fs/nfs/nfs4xdr.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++include/linux/nfs_xdr.h| 1 +3 files changed, 59 insertions(+), 1 deletions(-)fs/nfs/nfs4proc.c +0
-0@@ -129,7 +129,7 @@ const u32 nfs4_fsinfo_bitmap[2] = { FATTR4_WORD0_MAXFILESIZE| FATTR4_WORD0_MAXREAD | FATTR4_WORD0_MAXWRITE | FATTR4_WORD0_LEASE_TIME,Quoted 2 lines
- 0
+FATTR4_WORD1_FS_LAYOUT_TYPESNote that originally (9170cf5
pnfs_submit: set and unset pnfs layoutdriver modules), we set this only underCONFIG_NFS_V4_1, but it should be ok to ask for it anyhow.
Quoted 2 lines
Benny
Yes.
Fred
Quoted 95 lines
Quoted 89 lines
};
const u32 nfs4_fs_locations_bitmap[2] = {
fs/nfs/nfs4xdr.c +57
-0@@ -3868,6 +3868,60 @@ xdr_error: return status; } +/* +* Decode potentially multiple layout types. Currently we only support +* one layout driver per file system. +*/ +static int decode_first_pnfs_layout_type (struct xdr_stream *xdr, + uint32_t *layouttype) +{ + uint32_t *p; + int num; + + p = xdr_inline_decode(xdr, 4); + if (unlikely(!p)) + goto out_overflow; + num = be32_to_cpup(p); + + /* pNFS is not supported by the underlying file system */ + if (num == 0) { + *layouttype = 0; + return 0; + } + if (num > 1) + printk(KERN_INFO "%s: Warning: Multiple pNFS layout drivers " + "per filesystem not supported\n", __func__); + + /* Decode and set first layout type, move xdr->p past unused types */ + p = xdr_inline_decode(xdr, num * 4); + if (unlikely(!p)) + goto out_overflow; + *layouttype = be32_to_cpup(p); + return 0; +out_overflow: + print_overflow_msg(__func__, xdr); + return -EIO; +} + +/* +* The type of file system exported. +* Note we must ensure that layouttype is set in any non-error case. +*/ +static int decode_attr_pnfstype (struct xdr_stream *xdr,uint32_t *bitmap, + uint32_t *layouttype) +{ + int status = 0; + + dprintk("%s: bitmap is %x\n", __func__, bitmap[1]); + if (unlikely(bitmap[1] & (FATTR4_WORD1_FS_LAYOUT_TYPES - 1U))) + return -EIO; + if (bitmap[1] & FATTR4_WORD1_FS_LAYOUT_TYPES) { + status = decode_first_pnfs_layout_type(xdr, layouttype); + bitmap[1] &= ~FATTR4_WORD1_FS_LAYOUT_TYPES; + } else + *layouttype = 0; + return status; +} static int decode_fsinfo(struct xdr_stream *xdr, struct nfs_fsinfo *fsinfo) {@@ -3894,6 +3948,9 @@ static int decode_fsinfo(struct xdr_stream *xdr, struct nfs_fsinfo *fsinfo) if ((status = decode_attr_maxwrite(xdr, bitmap, &fsinfo->wtmax)) != 0) goto xdr_error; fsinfo->wtpref = fsinfo->wtmax; + status = decode_attr_pnfstype(xdr, bitmap, &fsinfo->layouttype); + if (status) + goto xdr_error; status = verify_attr_len(xdr, savep, attrlen);xdr_error:
include/linux/nfs_xdr.h +0
-0@@ -113,6 +113,7 @@ struct nfs_fsinfo {
__u32dtpref; /* pref. readdir transfer size */__u64maxfilesize;
__u32lease_time; /* in seconds */
+__u32layouttype; /* supported pnfs layout driver */ };Quoted 1 line
struct
nfs_fsstat{--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
PATCH 6/12 RFC: nfs: set layout driver
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
Put in the infrastructure that uses information returned from the server at mount to select a layout driver module.
In this patch, a stub is used that always returns "no driver found".
--- fs/nfs/Makefile | 1 + fs/nfs/client.c | 4 ++ fs/nfs/pnfs.c | 78 +++++++++++++++++++++++++++++++++++++++++++++ fs/nfs/pnfs.h | 56 ++++++++++++++++++++++++++++++++ include/linux/nfs_fs.h | 1 + include/linux/nfs_fs_sb.h | 1 + 6 files changed, 141 insertions(+), 0 deletions(-) create mode 100644 fs/nfs/pnfs.c create mode 100644 fs/nfs/pnfs.h
fs/nfs/Makefile +1 -0
@@ -15,5 +15,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ delegation.o idmap.o \ callback.o callback_xdr.o callback_proc.o \ nfs4namespace.o +nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o
fs/nfs/client.c +4 -0
@@ -48,6 +48,7 @@#include "iostat.h" #include "internal.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_CLIENT
@@ -898,6 +899,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo *if (server->wsize >NFS_MAX_FILE_IO_SIZE ) server->wsize =NFS_MAX_FILE_IO_SIZE ; server->wpages = (server->wsize +PAGE_CACHE_SIZE -1 ) >>PAGE_CACHE_SHIFT ; +set_pnfs_layoutdriver (server, fsinfo->layouttype); + server->wtmult =nfs_block_bits (fsinfo->wtmult,NULL ); server->dtsize =nfs_block_size (fsinfo->dtpref,NULL );
@@ -1017,6 +1020,7 @@ void nfs_free_server(struct nfs_server *server) {dprintk ("--> nfs_free_server()\n" ); +unset_pnfs_layoutdriver (server);spin_lock (&nfs_client_lock);list_del (&server->client_link);list_del (&server->master_link);
fs/nfs/pnfs.c new file +78 -0
78 lines not on this page · raw message
fs/nfs/pnfs.h new file +56 -0
56 lines not on this page · raw message
include/linux/nfs_fs.h +1 -0
@@ -615,6 +615,7 @@ nfs_fileid_to_ino_t(u64 fileid)#define NFSDBG_CLIENT 0x0200#define NFSDBG_MOUNT 0x0400#define NFSDBG_FSCACHE 0x0800 +#define NFSDBG_PNFS 0x1000#define NFSDBG_ALL 0xFFFF#ifdef __KERNEL__
include/linux/nfs_fs_sb.h +1 -0
@@ -144,6 +144,7 @@ struct nfs_server {u32 acl_bitmask;/* V4 bitmask representing the ACEs that are supported on this filesystem */ +struct pnfs_layoutdriver_type *pnfs_curr_ld;/* Active layout driver */ #endif void (*destroy)(struct nfs_server *);
Signature · 1 line
1.7.2.1
Re: [PATCH 06/12] RFC: nfs: set layout driver
Quoted 43 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Put in the infrastructure that uses information returned from the server at mount to select a layout driver module.
In this patch, a stub is used that always returns "no driver found".
Signed-off-by TBD - melding/reorganization of several patches--- fs/nfs/Makefile | 1 + fs/nfs/client.c | 4 ++ fs/nfs/pnfs.c | 78 +++++++++++++++++++++++++++++++++++++++++++++ fs/nfs/pnfs.h | 56 ++++++++++++++++++++++++++++++++ include/linux/nfs_fs.h | 1 + include/linux/nfs_fs_sb.h | 1 + 6 files changed, 141 insertions(+), 0 deletions(-) create mode 100644 fs/nfs/pnfs.c create mode 100644 fs/nfs/pnfs.hfs/nfs/Makefile +1
-0@@ -15,5 +15,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ delegation.o idmap.o \ callback.o callback_xdr.o callback_proc.o \ nfs4namespace.o +nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.ofs/nfs/client.c +1
-0@@ -48,6 +48,7 @@#include "iostat.h" #include "internal.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_CLIENT
fs/nfs/client.c static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * · line 903
Quoted 5 lines
fs/nfs/client.c +2
-0@@ -898,6 +899,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo *if (server->wsize >NFS_MAX_FILE_IO_SIZE ) server->wsize =NFS_MAX_FILE_IO_SIZE ; server->wpages = (server->wsize +PAGE_CACHE_SIZE -1 ) >>PAGE_CACHE_SHIFT ; +set_pnfs_layoutdriver (server, fsinfo->layouttype); +
Originally (9170cf5 pnfs_submit: set and unset pnfs layoutdriver modules), nfs4_init_pnfs called set_pnfs_layoutdriver under the following conditions:
if (nfs4_has_session(clp) && (clp->cl_exchange_flags & EXCHGID4_FLAG_USE_PNFS_MDS)) set_pnfs_layoutdriver(server, fsinfo->layouttype);
So in the DS only case we don't want to set server->pnfs_curr_ld. did you test your code with a standalone DS to make there are no ill side effects in this case?
Quoted 174 lines
fs/nfs/client.c +1
-0server->wtmult =nfs_block_bits (fsinfo->wtmult,NULL ); server->dtsize =nfs_block_size (fsinfo->dtpref,NULL );@@ -1017,6 +1020,7 @@ void nfs_free_server(struct nfs_server *server) {dprintk ("--> nfs_free_server()\n" ); +unset_pnfs_layoutdriver (server);spin_lock (&nfs_client_lock);list_del (&server->client_link);list_del (&server->master_link);fs/nfs/pnfs.c new file +78
-0@@ -0,0 +1,78 @@ +/* +* pNFS functions to call and manage layout drivers. +* +* Copyright (c) 2002 [year of first publication] +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#include <linux/nfs_fs.h> +#include "pnfs.h" + +#define NFSDBG_FACILITY NFSDBG_PNFS + +/* STUB that returns the equivalent of "no module found" */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver (u32 id) { +return NULL ; +} + +/* Unitialize a mountpoint in a layout driver */ +void +unset_pnfs_layoutdriver (struct nfs_server *nfss) +{ + nfss->pnfs_curr_ld =NULL ; +} + +/* +* Try to set the server's pnfs module to the pnfs layout type specified by id. +* Currently only one pNFS layout driver per filesystem is supported. +* +* @id layout type. Zero (illegal layout type) indicates pNFS not in use. +*/ +void +set_pnfs_layoutdriver (struct nfs_server *server,u32 id) +{ +struct pnfs_layoutdriver_type *ld_type =NULL ; + +if (id ==0 ) + goto out_no_driver; + ld_type =find_pnfs_driver (id); +if (!ld_type) { +request_module ("%s-%u" ,LAYOUT_NFSV4_1_MODULE_PREFIX , id); + ld_type =find_pnfs_driver (id); +if (!ld_type) { +dprintk ("%s: No pNFS module found for %u.\n" , + __func__, id); + goto out_no_driver; + } + } + server->pnfs_curr_ld = ld_type; +dprintk ("%s: pNFS module for %u set\n" , __func__, id); +return ; + +out_no_driver: +dprintk ("%s: Using NFSv4 I/O\n" , __func__); + server->pnfs_curr_ld =NULL ; +}fs/nfs/pnfs.h new file +56
-0@@ -0,0 +1,56 @@ +/* +* pNFS client data structures. +* +* Copyright (c) 2002 +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#ifndef FS_NFS_PNFS_H +#define FS_NFS_PNFS_H + +#ifdef CONFIG_NFS_V4_1 + +#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4" + +/* Per-layout driver specific registration structure */ +struct pnfs_layoutdriver_type { +}; + +void set_pnfs_layoutdriver (struct nfs_server *,u32 id); +void unset_pnfs_layoutdriver (struct nfs_server *); + +#else /* CONFIG_NFS_V4_1 */ + +static inline void set_pnfs_layoutdriver (struct nfs_server *s,u32 id) +{ +} + +static inline void unset_pnfs_layoutdriver (struct nfs_server *s) +{ +} + +#endif /* CONFIG_NFS_V4_1 */ + +#endif /* FS_NFS_PNFS_H */ include/linux/nfs_fs.h +1
-0@@ -615,6 +615,7 @@ nfs_fileid_to_ino_t(u64 fileid)#define NFSDBG_CLIENT 0x0200#define NFSDBG_MOUNT 0x0400#define NFSDBG_FSCACHE 0x0800 +#define NFSDBG_PNFS 0x1000#define NFSDBG_ALL 0xFFFF#ifdef __KERNEL__include/linux/nfs_fs_sb.h +0
-0
include/linux/nfs_fs_sb.h struct nfs_server { · line 147
Quoted 4 lines
include/linux/nfs_fs_sb.h +1
-0@@ -144,6 +144,7 @@ struct nfs_server {u32 acl_bitmask;/* V4 bitmask representing the ACEs that are supported on this filesystem */ +struct pnfs_layoutdriver_type *pnfs_curr_ld;/* Active layout driver */
This is under #ifdef CONFIG_NFS_V4
Why not only under CONFIG_NFS_V4_1?
(see 9170cf5 pnfs_submit: set and unset pnfs layoutdriver modules)
Benny
Quoted 3 lines
include/linux/nfs_fs_sb.h +0
-0#endif void (*destroy)(struct nfs_server *);
Re: [PATCH 06/12] RFC: nfs: set layout driver
Quoted 53 lines · Benny Halevy
Quoted 43 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Put in the infrastructure that uses information returned from the server at mount to select a layout driver module.
In this patch, a stub is used that always returns "no driver found".
Signed-off-by TBD - melding/reorganization of several patches---
fs/nfs/Makefile | 1 +
fs/nfs/client.c | 4 ++
fs/nfs/pnfs.c | 78 +++++++++++++++++++++++++++++++++++++++++++++fs/nfs/pnfs.h| 56 ++++++++++++++++++++++++++++++++include/linux/nfs_fs.h| 1 +
include/linux/nfs_fs_sb.h | 1 +6 files changed, 141 insertions(+), 0 deletions(-)create mode 100644
fs/nfs/pnfs.c
create mode 100644fs/nfs/pnfs.hfs/nfs/Makefile +0
-0@@ -15,5 +15,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ delegation.o idmap.o \ callback.o callback_xdr.o callback_proc.o \nfs4namespace.o +nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o
fs/nfs/client.c +1
-0@@ -48,6 +48,7 @@ #include "iostat.h" #include "internal.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_CLIENTfs/nfs/client.c static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * · line 903
Quoted 5 lines
fs/nfs/client.c +2
-0@@ -898,6 +899,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * if (server->wsize > NFS_MAX_FILE_IO_SIZE) server->wsize = NFS_MAX_FILE_IO_SIZE; server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; + set_pnfs_layoutdriver(server, fsinfo->layouttype); +Originally (9170cf5
pnfs_submit: set and unset pnfs layoutdriver modules),nfs4_init_pnfscalledset_pnfs_layoutdriverunder the following conditions:
Quoted 8 lines
if (
nfs4_has_session(clp)&& (clp->cl_exchange_flags&EXCHGID4_FLAG_USE_PNFS_MDS))set_pnfs_layoutdriver(server, fsinfo->layouttype);So in the DS only case we don't want to set
server->pnfs_curr_ld. did you test your code with a standalone DS to make there are no ill side effects in this case?
This is called from the mount code. It is never called for a DS (unless the DS is also a MDS, in which case it will be called in the MDS role).
Quoted 168 lines
Quoted 168 lines
fs/nfs/client.c +1
-0server->wtmult = nfs_block_bits(fsinfo->wtmult, NULL); server->dtsize = nfs_block_size(fsinfo->dtpref, NULL);@@ -1017,6 +1020,7 @@ void nfs_free_server(struct nfs_server *server) { dprintk("--> nfs_free_server()\n"); + unset_pnfs_layoutdriver(server); spin_lock(&nfs_client_lock); list_del(&server->client_link); list_del(&server->master_link);fs/nfs/pnfs.c new file +78
-0@@ -0,0 +1,78 @@ +/* +* pNFS functions to call and manage layout drivers. +* +* Copyright (c) 2002 [year of first publication] +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#include <linux/nfs_fs.h> +#include "pnfs.h" + +#define NFSDBG_FACILITY NFSDBG_PNFS + +/* STUB that returns the equivalent of "no module found" */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver (u32 id) { + return NULL; +} + +/* Unitialize a mountpoint in a layout driver */ +void +unset_pnfs_layoutdriver (struct nfs_server *nfss) +{ + nfss->pnfs_curr_ld = NULL; +} + +/* +* Try to set the server's pnfs module to the pnfs layout type specified by id. +* Currently only one pNFS layout driver per filesystem is supported. +* +* @id layout type. Zero (illegal layout type) indicates pNFS not in use. +*/ +void +set_pnfs_layoutdriver (struct nfs_server *server,u32 id) +{ + struct pnfs_layoutdriver_type *ld_type = NULL; + + if (id == 0) + goto out_no_driver; + ld_type = find_pnfs_driver(id); + if (!ld_type) { + request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); + ld_type = find_pnfs_driver(id); + if (!ld_type) { + dprintk("%s: No pNFS module found for %u.\n", + __func__, id); + goto out_no_driver; + } + } + server->pnfs_curr_ld = ld_type; + dprintk("%s: pNFS module for %u set\n", __func__, id); + return; + +out_no_driver: + dprintk("%s: Using NFSv4 I/O\n", __func__); + server->pnfs_curr_ld = NULL; +}fs/nfs/pnfs.h new file +56
-0@@ -0,0 +1,56 @@ +/* +* pNFS client data structures. +* +* Copyright (c) 2002 +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#ifndef FS_NFS_PNFS_H +#define FS_NFS_PNFS_H + +#ifdef CONFIG_NFS_V4_1 + +#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4" + +/* Per-layout driver specific registration structure */ +struct pnfs_layoutdriver_type { +}; + +void set_pnfs_layoutdriver (struct nfs_server *,u32 id); +void unset_pnfs_layoutdriver (struct nfs_server *); + +#else /* CONFIG_NFS_V4_1 */ + +static inline void set_pnfs_layoutdriver (struct nfs_server *s,u32 id) +{ +} + +static inline void unset_pnfs_layoutdriver (struct nfs_server *s) +{ +} + +#endif /* CONFIG_NFS_V4_1 */ + +#endif /* FS_NFS_PNFS_H */ include/linux/nfs_fs.h +1
-0@@ -615,6 +615,7 @@ nfs_fileid_to_ino_t(u64 fileid) #define NFSDBG_CLIENT 0x0200 #define NFSDBG_MOUNT 0x0400 #define NFSDBG_FSCACHE 0x0800 +#define NFSDBG_PNFS 0x1000 #define NFSDBG_ALL 0xFFFF
Quoted 14 lines
Quoted 10 lines
include/linux/nfs_fs.h +0
-0#ifdef __KERNEL__include/linux/nfs_fs_sb.h +0
-0@@ -144,6 +144,7 @@ struct nfs_server {u32
acl_bitmask; /* V4 bitmask representing the ACEs that are supported on this filesystem */
+ structpnfs_layoutdriver_type*pnfs_curr_ld; /* Active layout driver */This is under #ifdef
CONFIG_NFS_V4
Why not only underCONFIG_NFS_V4_1?
(see 9170cf5pnfs_submit: set and unset pnfs layoutdriver modules)
That was part of the campaign to remove as many ifdef statements as possible. I have no preference one way or the other.
Fred
Quoted 11 lines
Benny
Quoted 3 lines
#endif
void (*destroy)(structnfs_server*);--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 06/12] RFC: nfs: set layout driver
Quoted 251 lines · Fred Isaman
Quoted 53 lines · Benny Halevy
Quoted 43 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Put in the infrastructure that uses information returned from the server at mount to select a layout driver module.
In this patch, a stub is used that always returns "no driver found".
Signed-off-by TBD - melding/reorganization of several patches--- fs/nfs/Makefile | 1 + fs/nfs/client.c | 4 ++ fs/nfs/pnfs.c | 78 +++++++++++++++++++++++++++++++++++++++++++++ fs/nfs/pnfs.h | 56 ++++++++++++++++++++++++++++++++ include/linux/nfs_fs.h | 1 + include/linux/nfs_fs_sb.h | 1 + 6 files changed, 141 insertions(+), 0 deletions(-) create mode 100644 fs/nfs/pnfs.c create mode 100644 fs/nfs/pnfs.hfs/nfs/Makefile +1
-0@@ -15,5 +15,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ delegation.o idmap.o \ callback.o callback_xdr.o callback_proc.o \ nfs4namespace.o +nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.ofs/nfs/client.c +1
-0@@ -48,6 +48,7 @@#include "iostat.h" #include "internal.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_CLIENTfs/nfs/client.c static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * · line 903
Quoted 5 lines
fs/nfs/client.c +2
-0@@ -898,6 +899,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo *if (server->wsize >NFS_MAX_FILE_IO_SIZE ) server->wsize =NFS_MAX_FILE_IO_SIZE ; server->wpages = (server->wsize +PAGE_CACHE_SIZE -1 ) >>PAGE_CACHE_SHIFT ; +set_pnfs_layoutdriver (server, fsinfo->layouttype); +Originally (9170cf5
pnfs_submit: set and unset pnfs layoutdriver modules),nfs4_init_pnfscalledset_pnfs_layoutdriverunder the following conditions:Quoted 8 lines
if (nfs4_has_session(clp) && (clp->cl_exchange_flags & EXCHGID4_FLAG_USE_PNFS_MDS)) set_pnfs_layoutdriver(server, fsinfo->layouttype);So in the DS only case we don't want to set
server->pnfs_curr_ld. did you test your code with a standalone DS to make there are no ill side effects in this case?This is called from the mount code. It is never called for a DS (unless the DS is also a MDS, in which case it will be called in the MDS role).
Quoted 168 lines
Quoted 168 lines
fs/nfs/client.c +1
-0server->wtmult =nfs_block_bits (fsinfo->wtmult,NULL ); server->dtsize =nfs_block_size (fsinfo->dtpref,NULL );@@ -1017,6 +1020,7 @@ void nfs_free_server(struct nfs_server *server) {dprintk ("--> nfs_free_server()\n" ); +unset_pnfs_layoutdriver (server);spin_lock (&nfs_client_lock);list_del (&server->client_link);list_del (&server->master_link);fs/nfs/pnfs.c new file +78
-0@@ -0,0 +1,78 @@ +/* +* pNFS functions to call and manage layout drivers. +* +* Copyright (c) 2002 [year of first publication] +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#include <linux/nfs_fs.h> +#include "pnfs.h" + +#define NFSDBG_FACILITY NFSDBG_PNFS + +/* STUB that returns the equivalent of "no module found" */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver (u32 id) { +return NULL ; +} + +/* Unitialize a mountpoint in a layout driver */ +void +unset_pnfs_layoutdriver (struct nfs_server *nfss) +{ + nfss->pnfs_curr_ld =NULL ; +} + +/* +* Try to set the server's pnfs module to the pnfs layout type specified by id. +* Currently only one pNFS layout driver per filesystem is supported. +* +* @id layout type. Zero (illegal layout type) indicates pNFS not in use. +*/ +void +set_pnfs_layoutdriver (struct nfs_server *server,u32 id) +{ +struct pnfs_layoutdriver_type *ld_type =NULL ; + +if (id ==0 ) + goto out_no_driver; + ld_type =find_pnfs_driver (id); +if (!ld_type) { +request_module ("%s-%u" ,LAYOUT_NFSV4_1_MODULE_PREFIX , id); + ld_type =find_pnfs_driver (id); +if (!ld_type) { +dprintk ("%s: No pNFS module found for %u.\n" , + __func__, id); + goto out_no_driver; + } + } + server->pnfs_curr_ld = ld_type; +dprintk ("%s: pNFS module for %u set\n" , __func__, id); +return ; + +out_no_driver: +dprintk ("%s: Using NFSv4 I/O\n" , __func__); + server->pnfs_curr_ld =NULL ; +}fs/nfs/pnfs.h new file +56
-0@@ -0,0 +1,56 @@ +/* +* pNFS client data structures. +* +* Copyright (c) 2002 +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#ifndef FS_NFS_PNFS_H +#define FS_NFS_PNFS_H + +#ifdef CONFIG_NFS_V4_1 + +#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4" + +/* Per-layout driver specific registration structure */ +struct pnfs_layoutdriver_type { +}; + +void set_pnfs_layoutdriver (struct nfs_server *,u32 id); +void unset_pnfs_layoutdriver (struct nfs_server *); + +#else /* CONFIG_NFS_V4_1 */ + +static inline void set_pnfs_layoutdriver (struct nfs_server *s,u32 id) +{ +} + +static inline void unset_pnfs_layoutdriver (struct nfs_server *s) +{ +} + +#endif /* CONFIG_NFS_V4_1 */ + +#endif /* FS_NFS_PNFS_H */ include/linux/nfs_fs.h +1
-0@@ -615,6 +615,7 @@ nfs_fileid_to_ino_t(u64 fileid)#define NFSDBG_CLIENT 0x0200#define NFSDBG_MOUNT 0x0400#define NFSDBG_FSCACHE 0x0800 +#define NFSDBG_PNFS 0x1000#define NFSDBG_ALL 0xFFFFQuoted 14 lines
include/linux/nfs_fs_sb.h struct nfs_server { · line 147
Quoted 10 lines
include/linux/nfs_fs.h +0
-0#ifdef __KERNEL__include/linux/nfs_fs_sb.h +1
-0@@ -144,6 +144,7 @@ struct nfs_server {u32 acl_bitmask;/* V4 bitmask representing the ACEs that are supported on this filesystem */ +struct pnfs_layoutdriver_type *pnfs_curr_ld;/* Active layout driver */ This is under #ifdef
CONFIG_NFS_V4
Why not only underCONFIG_NFS_V4_1?
(see 9170cf5pnfs_submit: set and unset pnfs layoutdriver modules)
Quoted 2 lines
That was part of the campaign to remove as many ifdef statements as possible. I have no preference one way or the other.
I'll defer this to Trond then.
In any case, I think that this patch can be merged with the following one - "full mount/umount infrastructure" for completeness.
Benny
Quoted 14 lines
Fred
Quoted 11 lines
Benny
include/linux/nfs_fs_sb.h struct nfs_server { · line 150
Quoted 3 lines
include/linux/nfs_fs_sb.h +0
-0#endif void (*destroy)(struct nfs_server *);--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 06/12] RFC: nfs: set layout driver
Quoted 255 lines · Benny Halevy
Quoted 251 lines · Fred Isaman
Quoted 53 lines · Benny Halevy
Quoted 43 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Put in the infrastructure that uses information returned from the server at mount to select a layout driver module.
In this patch, a stub is used that always returns "no driver found".
Signed-off-by TBD - melding/reorganization of several patches---
fs/nfs/Makefile | 1 +
fs/nfs/client.c | 4 ++
fs/nfs/pnfs.c | 78 +++++++++++++++++++++++++++++++++++++++++++++fs/nfs/pnfs.h| 56 ++++++++++++++++++++++++++++++++include/linux/nfs_fs.h| 1 +
include/linux/nfs_fs_sb.h | 1 +6 files changed, 141 insertions(+), 0 deletions(-)create mode 100644
fs/nfs/pnfs.c
create mode 100644fs/nfs/pnfs.hfs/nfs/Makefile +0
-0@@ -15,5 +15,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ delegation.o idmap.o \ callback.o callback_xdr.o callback_proc.o \nfs4namespace.o +nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o
fs/nfs/client.c +1
-0@@ -48,6 +48,7 @@ #include "iostat.h" #include "internal.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_CLIENTfs/nfs/client.c static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * · line 903
Quoted 5 lines
fs/nfs/client.c +2
-0@@ -898,6 +899,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * if (server->wsize > NFS_MAX_FILE_IO_SIZE) server->wsize = NFS_MAX_FILE_IO_SIZE; server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; + set_pnfs_layoutdriver(server, fsinfo->layouttype); +Originally (9170cf5
pnfs_submit: set and unset pnfs layoutdriver modules),nfs4_init_pnfscalledset_pnfs_layoutdriverunder the following conditions:Quoted 8 lines
if (
nfs4_has_session(clp)&& (clp->cl_exchange_flags&EXCHGID4_FLAG_USE_PNFS_MDS))set_pnfs_layoutdriver(server, fsinfo->layouttype);So in the DS only case we don't want to set
server->pnfs_curr_ld. did you test your code with a standalone DS to make there are no ill side effects in this case?This is called from the mount code. It is never called for a DS (unless the DS is also a MDS, in which case it will be called in the MDS role).
Quoted 168 lines
Quoted 168 lines
fs/nfs/client.c +1
-0server->wtmult = nfs_block_bits(fsinfo->wtmult, NULL); server->dtsize = nfs_block_size(fsinfo->dtpref, NULL);@@ -1017,6 +1020,7 @@ void nfs_free_server(struct nfs_server *server) { dprintk("--> nfs_free_server()\n"); + unset_pnfs_layoutdriver(server); spin_lock(&nfs_client_lock); list_del(&server->client_link); list_del(&server->master_link);fs/nfs/pnfs.c new file +78
-0@@ -0,0 +1,78 @@ +/* +* pNFS functions to call and manage layout drivers. +* +* Copyright (c) 2002 [year of first publication] +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#include <linux/nfs_fs.h> +#include "pnfs.h" + +#define NFSDBG_FACILITY NFSDBG_PNFS + +/* STUB that returns the equivalent of "no module found" */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver (u32 id) { + return NULL; +} + +/* Unitialize a mountpoint in a layout driver */ +void +unset_pnfs_layoutdriver (struct nfs_server *nfss) +{ + nfss->pnfs_curr_ld = NULL; +} + +/* +* Try to set the server's pnfs module to the pnfs layout type specified by id. +* Currently only one pNFS layout driver per filesystem is supported. +* +* @id layout type. Zero (illegal layout type) indicates pNFS not in use. +*/ +void +set_pnfs_layoutdriver (struct nfs_server *server,u32 id) +{ + struct pnfs_layoutdriver_type *ld_type = NULL; + + if (id == 0) + goto out_no_driver; + ld_type = find_pnfs_driver(id); + if (!ld_type) { + request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); + ld_type = find_pnfs_driver(id); + if (!ld_type) { + dprintk("%s: No pNFS module found for %u.\n", + __func__, id); + goto out_no_driver; + } + } + server->pnfs_curr_ld = ld_type; + dprintk("%s: pNFS module for %u set\n", __func__, id); + return; + +out_no_driver: + dprintk("%s: Using NFSv4 I/O\n", __func__); + server->pnfs_curr_ld = NULL; +}fs/nfs/pnfs.h new file +56
-0@@ -0,0 +1,56 @@ +/* +* pNFS client data structures. +* +* Copyright (c) 2002 +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#ifndef FS_NFS_PNFS_H +#define FS_NFS_PNFS_H + +#ifdef CONFIG_NFS_V4_1 + +#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4" + +/* Per-layout driver specific registration structure */ +struct pnfs_layoutdriver_type { +}; + +void set_pnfs_layoutdriver (struct nfs_server *,u32 id); +void unset_pnfs_layoutdriver (struct nfs_server *); + +#else /* CONFIG_NFS_V4_1 */ + +static inline void set_pnfs_layoutdriver (struct nfs_server *s,u32 id) +{ +} + +static inline void unset_pnfs_layoutdriver (struct nfs_server *s) +{ +} + +#endif /* CONFIG_NFS_V4_1 */ + +#endif /* FS_NFS_PNFS_H */ include/linux/nfs_fs.h +1
-0@@ -615,6 +615,7 @@ nfs_fileid_to_ino_t(u64 fileid) #define NFSDBG_CLIENT 0x0200 #define NFSDBG_MOUNT 0x0400 #define NFSDBG_FSCACHE 0x0800 +#define NFSDBG_PNFS 0x1000 #define NFSDBG_ALL 0xFFFFQuoted 14 lines
Quoted 10 lines
include/linux/nfs_fs.h +0
-0#ifdef __KERNEL__include/linux/nfs_fs_sb.h +0
-0@@ -144,6 +144,7 @@ struct nfs_server {u32
acl_bitmask; /* V4 bitmask representing the ACEs that are supported on this filesystem */
+ structpnfs_layoutdriver_type*pnfs_curr_ld; /* Active layout driver */This is under #ifdef
CONFIG_NFS_V4
Why not only underCONFIG_NFS_V4_1?
(see 9170cf5pnfs_submit: set and unset pnfs layoutdriver modules)Quoted 2 lines
That was part of the campaign to remove as many ifdef statements as possible. I have no preference one way or the other.
Quoted 7 lines
I'll defer this to Trond then.
In any case, I think that this patch can be merged with the following one - "full mount/umount infrastructure" for completeness.
Benny
This was an attempt to minimize the size of the patch touching the current NFS code patch.
Fred
Quoted 19 lines
Quoted 14 lines
Fred
Quoted 11 lines
Benny
Quoted 3 lines
#endif
void (*destroy)(structnfs_server*);--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 06/12] RFC: nfs: set layout driver
In reply to Fred IsamanQuoted 58 lines · Fred Isaman
Quoted 57 lines · Benny Halevy
Quoted 43 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Put in the infrastructure that uses information returned from the server at mount to select a layout driver module.
In this patch, a stub is used that always returns "no driver found".
Signed-off-by TBD - melding/reorganization of several patches--- fs/nfs/Makefile | 1 + fs/nfs/client.c | 4 ++ fs/nfs/pnfs.c | 78 +++++++++++++++++++++++++++++++++++++++++++++ fs/nfs/pnfs.h | 56 ++++++++++++++++++++++++++++++++ include/linux/nfs_fs.h | 1 + include/linux/nfs_fs_sb.h | 1 + 6 files changed, 141 insertions(+), 0 deletions(-) create mode 100644 fs/nfs/pnfs.c create mode 100644 fs/nfs/pnfs.hfs/nfs/Makefile +1
-0@@ -15,5 +15,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ delegation.o idmap.o \ callback.o callback_xdr.o callback_proc.o \ nfs4namespace.o +nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.ofs/nfs/client.c +1
-0@@ -48,6 +48,7 @@#include "iostat.h" #include "internal.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_CLIENTfs/nfs/client.c static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * · line 903
Quoted 5 lines
fs/nfs/client.c +2
-0@@ -898,6 +899,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo *if (server->wsize >NFS_MAX_FILE_IO_SIZE ) server->wsize =NFS_MAX_FILE_IO_SIZE ; server->wpages = (server->wsize +PAGE_CACHE_SIZE -1 ) >>PAGE_CACHE_SHIFT ; +set_pnfs_layoutdriver (server, fsinfo->layouttype); +Originally (9170cf5
pnfs_submit: set and unset pnfs layoutdriver modules),nfs4_init_pnfscalledset_pnfs_layoutdriverunder the following conditions:if (nfs4_has_session(clp) && (clp->cl_exchange_flags & EXCHGID4_FLAG_USE_PNFS_MDS)) set_pnfs_layoutdriver(server, fsinfo->layouttype);
Quoted 8 lines
Quoted 4 lines
So in the DS only case we don't want to set
server->pnfs_curr_ld. did you test your code with a standalone DS to make there are no ill side effects in this case?This is called from the mount code. It is never called for a DS (unless the DS is also a MDS, in which case it will be called in the MDS role).
So at least let's add a BUG_ON and a comment since we're relying on side effects of code implemented elsewhere.
Benny
P.S. I wonder what happens if you try to explicitly mount a DS (which is not an MDS).
Quoted 201 lines
Quoted 170 lines
Quoted 170 lines
fs/nfs/client.c +1
-0server->wtmult =nfs_block_bits (fsinfo->wtmult,NULL ); server->dtsize =nfs_block_size (fsinfo->dtpref,NULL );@@ -1017,6 +1020,7 @@ void nfs_free_server(struct nfs_server *server) {dprintk ("--> nfs_free_server()\n" ); +unset_pnfs_layoutdriver (server);spin_lock (&nfs_client_lock);list_del (&server->client_link);list_del (&server->master_link);fs/nfs/pnfs.c new file +78
-0@@ -0,0 +1,78 @@ +/* +* pNFS functions to call and manage layout drivers. +* +* Copyright (c) 2002 [year of first publication] +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#include <linux/nfs_fs.h> +#include "pnfs.h" + +#define NFSDBG_FACILITY NFSDBG_PNFS + +/* STUB that returns the equivalent of "no module found" */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver (u32 id) { +return NULL ; +} + +/* Unitialize a mountpoint in a layout driver */ +void +unset_pnfs_layoutdriver (struct nfs_server *nfss) +{ + nfss->pnfs_curr_ld =NULL ; +} + +/* +* Try to set the server's pnfs module to the pnfs layout type specified by id. +* Currently only one pNFS layout driver per filesystem is supported. +* +* @id layout type. Zero (illegal layout type) indicates pNFS not in use. +*/ +void +set_pnfs_layoutdriver (struct nfs_server *server,u32 id) +{ +struct pnfs_layoutdriver_type *ld_type =NULL ; + +if (id ==0 ) + goto out_no_driver; + ld_type =find_pnfs_driver (id); +if (!ld_type) { +request_module ("%s-%u" ,LAYOUT_NFSV4_1_MODULE_PREFIX , id); + ld_type =find_pnfs_driver (id); +if (!ld_type) { +dprintk ("%s: No pNFS module found for %u.\n" , + __func__, id); + goto out_no_driver; + } + } + server->pnfs_curr_ld = ld_type; +dprintk ("%s: pNFS module for %u set\n" , __func__, id); +return ; + +out_no_driver: +dprintk ("%s: Using NFSv4 I/O\n" , __func__); + server->pnfs_curr_ld =NULL ; +}fs/nfs/pnfs.h new file +56
-0@@ -0,0 +1,56 @@ +/* +* pNFS client data structures. +* +* Copyright (c) 2002 +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#ifndef FS_NFS_PNFS_H +#define FS_NFS_PNFS_H + +#ifdef CONFIG_NFS_V4_1 + +#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4" + +/* Per-layout driver specific registration structure */ +struct pnfs_layoutdriver_type { +}; + +void set_pnfs_layoutdriver (struct nfs_server *,u32 id); +void unset_pnfs_layoutdriver (struct nfs_server *); + +#else /* CONFIG_NFS_V4_1 */ + +static inline void set_pnfs_layoutdriver (struct nfs_server *s,u32 id) +{ +} + +static inline void unset_pnfs_layoutdriver (struct nfs_server *s) +{ +} + +#endif /* CONFIG_NFS_V4_1 */ + +#endif /* FS_NFS_PNFS_H */ include/linux/nfs_fs.h +1
-0@@ -615,6 +615,7 @@ nfs_fileid_to_ino_t(u64 fileid)#define NFSDBG_CLIENT 0x0200#define NFSDBG_MOUNT 0x0400#define NFSDBG_FSCACHE 0x0800 +#define NFSDBG_PNFS 0x1000#define NFSDBG_ALL 0xFFFFQuoted 14 lines
include/linux/nfs_fs_sb.h struct nfs_server { · line 147
Quoted 10 lines
include/linux/nfs_fs.h +0
-0#ifdef __KERNEL__include/linux/nfs_fs_sb.h +1
-0@@ -144,6 +144,7 @@ struct nfs_server {u32 acl_bitmask;/* V4 bitmask representing the ACEs that are supported on this filesystem */ +struct pnfs_layoutdriver_type *pnfs_curr_ld;/* Active layout driver */ This is under #ifdef
CONFIG_NFS_V4
Why not only underCONFIG_NFS_V4_1?
(see 9170cf5pnfs_submit: set and unset pnfs layoutdriver modules)That was part of the campaign to remove as many ifdef statements as possible. I have no preference one way or the other.
Fred
Quoted 11 lines
Benny
include/linux/nfs_fs_sb.h struct nfs_server { · line 150
Quoted 3 lines
include/linux/nfs_fs_sb.h +0
-0#endif void (*destroy)(struct nfs_server *);--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 06/12] RFC: nfs: set layout driver
Quoted 64 lines · Benny Halevy
Quoted 63 lines · Fred Isaman
Quoted 61 lines · Benny Halevy
Quoted 43 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Put in the infrastructure that uses information returned from the server at mount to select a layout driver module.
In this patch, a stub is used that always returns "no driver found".
Signed-off-by TBD - melding/reorganization of several patches---
fs/nfs/Makefile | 1 +
fs/nfs/client.c | 4 ++
fs/nfs/pnfs.c | 78 +++++++++++++++++++++++++++++++++++++++++++++fs/nfs/pnfs.h| 56 ++++++++++++++++++++++++++++++++include/linux/nfs_fs.h| 1 +
include/linux/nfs_fs_sb.h | 1 +6 files changed, 141 insertions(+), 0 deletions(-)create mode 100644
fs/nfs/pnfs.c
create mode 100644fs/nfs/pnfs.hfs/nfs/Makefile +0
-0@@ -15,5 +15,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ delegation.o idmap.o \ callback.o callback_xdr.o callback_proc.o \nfs4namespace.o +nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o
fs/nfs/client.c +1
-0@@ -48,6 +48,7 @@ #include "iostat.h" #include "internal.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_CLIENTfs/nfs/client.c static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * · line 903
Quoted 5 lines
fs/nfs/client.c +2
-0@@ -898,6 +899,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * if (server->wsize > NFS_MAX_FILE_IO_SIZE) server->wsize = NFS_MAX_FILE_IO_SIZE; server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; + set_pnfs_layoutdriver(server, fsinfo->layouttype); +Originally (9170cf5
pnfs_submit: set and unset pnfs layoutdriver modules),nfs4_init_pnfscalledset_pnfs_layoutdriverunder the following conditions:if (
nfs4_has_session(clp)&& (clp->cl_exchange_flags&EXCHGID4_FLAG_USE_PNFS_MDS))set_pnfs_layoutdriver(server, fsinfo->layouttype);So in the DS only case we don't want to set
server->pnfs_curr_ld. did you test your code with a standalone DS to make there are no ill side effects in this case?
Quoted 7 lines
Quoted 3 lines
This is called from the mount code. It is never called for a DS (unless the DS is also a MDS, in which case it will be called in the MDS role).
So at least let's add a
BUG_ONand a comment since we're relying on side effects of code implemented elsewhere.
OK.
Quoted 6 lines
Benny
P.S. I wonder what happens if you try to explicitly mount a DS (which is not an MDS).
Note there are two possibilities here, DS only, or DS and normal NFS server (but not a MDS).
In the second case, the mount should work, and it does. In the first, the client should abort when it sees the EXCHID flags (which it doesn't...this needs to be fixed) or when the server sends NFS4ERR_NOTSUPP to the PUTROOTFH, which it MUST do.
Fred
Re: [PATCH 06/12] RFC: nfs: set layout driver
In reply to Fred IsamanQuoted 92 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Put in the infrastructure that uses information returned from the server at mount to select a layout driver module.
In this patch, a stub is used that always returns "no driver found".
Signed-off-by TBD - melding/reorganization of several patches--- fs/nfs/Makefile | 1 + fs/nfs/client.c | 4 ++ fs/nfs/pnfs.c | 78 +++++++++++++++++++++++++++++++++++++++++++++ fs/nfs/pnfs.h | 56 ++++++++++++++++++++++++++++++++ include/linux/nfs_fs.h | 1 + include/linux/nfs_fs_sb.h | 1 + 6 files changed, 141 insertions(+), 0 deletions(-) create mode 100644 fs/nfs/pnfs.c create mode 100644 fs/nfs/pnfs.hfs/nfs/Makefile +1
-0@@ -15,5 +15,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ delegation.o idmap.o \ callback.o callback_xdr.o callback_proc.o \ nfs4namespace.o +nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.ofs/nfs/client.c +4
-0@@ -48,6 +48,7 @@#include "iostat.h" #include "internal.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_CLIENT@@ -898,6 +899,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo *if (server->wsize >NFS_MAX_FILE_IO_SIZE ) server->wsize =NFS_MAX_FILE_IO_SIZE ; server->wpages = (server->wsize +PAGE_CACHE_SIZE -1 ) >>PAGE_CACHE_SHIFT ; +set_pnfs_layoutdriver (server, fsinfo->layouttype); + server->wtmult =nfs_block_bits (fsinfo->wtmult,NULL ); server->dtsize =nfs_block_size (fsinfo->dtpref,NULL );@@ -1017,6 +1020,7 @@ void nfs_free_server(struct nfs_server *server) {dprintk ("--> nfs_free_server()\n" ); +unset_pnfs_layoutdriver (server);spin_lock (&nfs_client_lock);list_del (&server->client_link);list_del (&server->master_link);fs/nfs/pnfs.c new file +27
-0@@ -0,0 +1,78 @@ +/* +* pNFS functions to call and manage layout drivers. +* +* Copyright (c) 2002 [year of first publication] +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages.
fs/nfs/pnfs.c line 37
Quoted 10 lines
fs/nfs/pnfs.c new file +10
-0+*/ + +#include <linux/nfs_fs.h> +#include "pnfs.h" + +#define NFSDBG_FACILITY NFSDBG_PNFS + +/* STUB that returns the equivalent of "no module found" */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver (u32 id) {
nit: the curly brace should be moved down a line.
Benny
Quoted 127 lines
fs/nfs/pnfs.c new file +41
-0+return NULL ; +} + +/* Unitialize a mountpoint in a layout driver */ +void +unset_pnfs_layoutdriver (struct nfs_server *nfss) +{ + nfss->pnfs_curr_ld =NULL ; +} + +/* +* Try to set the server's pnfs module to the pnfs layout type specified by id. +* Currently only one pNFS layout driver per filesystem is supported. +* +* @id layout type. Zero (illegal layout type) indicates pNFS not in use. +*/ +void +set_pnfs_layoutdriver (struct nfs_server *server,u32 id) +{ +struct pnfs_layoutdriver_type *ld_type =NULL ; + +if (id ==0 ) + goto out_no_driver; + ld_type =find_pnfs_driver (id); +if (!ld_type) { +request_module ("%s-%u" ,LAYOUT_NFSV4_1_MODULE_PREFIX , id); + ld_type =find_pnfs_driver (id); +if (!ld_type) { +dprintk ("%s: No pNFS module found for %u.\n" , + __func__, id); + goto out_no_driver; + } + } + server->pnfs_curr_ld = ld_type; +dprintk ("%s: pNFS module for %u set\n" , __func__, id); +return ; + +out_no_driver: +dprintk ("%s: Using NFSv4 I/O\n" , __func__); + server->pnfs_curr_ld =NULL ; +}fs/nfs/pnfs.h new file +56
-0@@ -0,0 +1,56 @@ +/* +* pNFS client data structures. +* +* Copyright (c) 2002 +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#ifndef FS_NFS_PNFS_H +#define FS_NFS_PNFS_H + +#ifdef CONFIG_NFS_V4_1 + +#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4" + +/* Per-layout driver specific registration structure */ +struct pnfs_layoutdriver_type { +}; + +void set_pnfs_layoutdriver (struct nfs_server *,u32 id); +void unset_pnfs_layoutdriver (struct nfs_server *); + +#else /* CONFIG_NFS_V4_1 */ + +static inline void set_pnfs_layoutdriver (struct nfs_server *s,u32 id) +{ +} + +static inline void unset_pnfs_layoutdriver (struct nfs_server *s) +{ +} + +#endif /* CONFIG_NFS_V4_1 */ + +#endif /* FS_NFS_PNFS_H */ include/linux/nfs_fs.h +1
-0@@ -615,6 +615,7 @@ nfs_fileid_to_ino_t(u64 fileid)#define NFSDBG_CLIENT 0x0200#define NFSDBG_MOUNT 0x0400#define NFSDBG_FSCACHE 0x0800 +#define NFSDBG_PNFS 0x1000#define NFSDBG_ALL 0xFFFF#ifdef __KERNEL__include/linux/nfs_fs_sb.h +1
-0@@ -144,6 +144,7 @@ struct nfs_server {u32 acl_bitmask;/* V4 bitmask representing the ACEs that are supported on this filesystem */ +struct pnfs_layoutdriver_type *pnfs_curr_ld;/* Active layout driver */ #endif void (*destroy)(struct nfs_server *);
Re: [PATCH 06/12] RFC: nfs: set layout driver
Quoted 106 lines · Benny Halevy
Quoted 92 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Put in the infrastructure that uses information returned from the server at mount to select a layout driver module.
In this patch, a stub is used that always returns "no driver found".
Signed-off-by TBD - melding/reorganization of several patches---
fs/nfs/Makefile | 1 +
fs/nfs/client.c | 4 ++
fs/nfs/pnfs.c | 78 +++++++++++++++++++++++++++++++++++++++++++++fs/nfs/pnfs.h| 56 ++++++++++++++++++++++++++++++++include/linux/nfs_fs.h| 1 +
include/linux/nfs_fs_sb.h | 1 +6 files changed, 141 insertions(+), 0 deletions(-)create mode 100644
fs/nfs/pnfs.c
create mode 100644fs/nfs/pnfs.hfs/nfs/Makefile +0
-0@@ -15,5 +15,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ delegation.o idmap.o \ callback.o callback_xdr.o callback_proc.o \nfs4namespace.o +nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o
fs/nfs/client.c +4
-0@@ -48,6 +48,7 @@ #include "iostat.h" #include "internal.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_CLIENT@@ -898,6 +899,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo * if (server->wsize > NFS_MAX_FILE_IO_SIZE) server->wsize = NFS_MAX_FILE_IO_SIZE; server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; + set_pnfs_layoutdriver(server, fsinfo->layouttype); + server->wtmult = nfs_block_bits(fsinfo->wtmult, NULL); server->dtsize = nfs_block_size(fsinfo->dtpref, NULL);@@ -1017,6 +1020,7 @@ void nfs_free_server(struct nfs_server *server) { dprintk("--> nfs_free_server()\n"); + unset_pnfs_layoutdriver(server); spin_lock(&nfs_client_lock); list_del(&server->client_link); list_del(&server->master_link);fs/nfs/pnfs.c new file +27
-0@@ -0,0 +1,78 @@ +/* +* pNFS functions to call and manage layout drivers. +* +* Copyright (c) 2002 [year of first publication] +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. fs/nfs/pnfs.c line 37
Quoted 10 lines
fs/nfs/pnfs.c new file +10
-0+*/ + +#include <linux/nfs_fs.h> +#include "pnfs.h" + +#define NFSDBG_FACILITY NFSDBG_PNFS + +/* STUB that returns the equivalent of "no module found" */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver (u32 id) {nit: the curly brace should be moved down a line.
Quoted 2 lines
Benny
OK.
Fred
Quoted 132 lines
Quoted 120 lines
fs/nfs/pnfs.c new file +41
-0+ return NULL; +} + +/* Unitialize a mountpoint in a layout driver */ +void +unset_pnfs_layoutdriver (struct nfs_server *nfss) +{ + nfss->pnfs_curr_ld = NULL; +} + +/* +* Try to set the server's pnfs module to the pnfs layout type specified by id. +* Currently only one pNFS layout driver per filesystem is supported. +* +* @id layout type. Zero (illegal layout type) indicates pNFS not in use. +*/ +void +set_pnfs_layoutdriver (struct nfs_server *server,u32 id) +{ + struct pnfs_layoutdriver_type *ld_type = NULL; + + if (id == 0) + goto out_no_driver; + ld_type = find_pnfs_driver(id); + if (!ld_type) { + request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); + ld_type = find_pnfs_driver(id); + if (!ld_type) { + dprintk("%s: No pNFS module found for %u.\n", + __func__, id); + goto out_no_driver; + } + } + server->pnfs_curr_ld = ld_type; + dprintk("%s: pNFS module for %u set\n", __func__, id); + return; + +out_no_driver: + dprintk("%s: Using NFSv4 I/O\n", __func__); + server->pnfs_curr_ld = NULL; +}fs/nfs/pnfs.h new file +56
-0@@ -0,0 +1,56 @@ +/* +* pNFS client data structures. +* +* Copyright (c) 2002 +* The Regents of the University of Michigan +* All Rights Reserved +* +* Dean Hildebrand <[email protected]> +* +* Permission is granted to use, copy, create derivative works, and +* redistribute this software and such derivative works for any purpose, +* so long as the name of the University of Michigan is not used in +* any advertising or publicity pertaining to the use or distribution +* of this software without specific, written prior authorization. If +* the above copyright notice or any other identification of the +* University of Michigan is included in any copy of any portion of +* this software, then the disclaimer below must also be included. +* +* This software is provided as is, without representation or warranty +* of any kind either express or implied, including without limitation +* the implied warranties of merchantability, fitness for a particular +* purpose, or noninfringement. The Regents of the University of +* Michigan shall not be liable for any damages, including special, +* indirect, incidental, or consequential damages, with respect to any +* claim arising out of or in connection with the use of the software, +* even if it has been or is hereafter advised of the possibility of +* such damages. +*/ + +#ifndef FS_NFS_PNFS_H +#define FS_NFS_PNFS_H + +#ifdef CONFIG_NFS_V4_1 + +#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4" + +/* Per-layout driver specific registration structure */ +struct pnfs_layoutdriver_type { +}; + +void set_pnfs_layoutdriver (struct nfs_server *,u32 id); +void unset_pnfs_layoutdriver (struct nfs_server *); + +#else /* CONFIG_NFS_V4_1 */ + +static inline void set_pnfs_layoutdriver (struct nfs_server *s,u32 id) +{ +} + +static inline void unset_pnfs_layoutdriver (struct nfs_server *s) +{ +} + +#endif /* CONFIG_NFS_V4_1 */ + +#endif /* FS_NFS_PNFS_H */ include/linux/nfs_fs.h +1
-0@@ -615,6 +615,7 @@ nfs_fileid_to_ino_t(u64 fileid) #define NFSDBG_CLIENT 0x0200 #define NFSDBG_MOUNT 0x0400 #define NFSDBG_FSCACHE 0x0800 +#define NFSDBG_PNFS 0x1000 #define NFSDBG_ALL 0xFFFF #ifdef __KERNEL__include/linux/nfs_fs_sb.h +0
-0@@ -144,6 +144,7 @@ struct nfs_server {Quoted 7 lines
u32
acl_bitmask; /* V4 bitmask representing the ACEs that are supported on this filesystem */
+ structpnfs_layoutdriver_type*pnfs_curr_ld; /* Active layout driver */ #endif
void (*destroy)(structnfs_server*);--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
PATCH 7/12 RFC: pnfs: full mount/umount infrastructure
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
Allow a module implementing a layout type to register, and have its mount/umount routines called for filesystems that the server declares support it.
--- Documentation/filesystems/nfs/00-INDEX | 2 + Documentation/filesystems/nfs/pnfs.txt | 48 ++++++++++++++++++ fs/nfs/Kconfig | 2 +- fs/nfs/pnfs.c | 82 +++++++++++++++++++++++++++++++- fs/nfs/pnfs.h | 9 ++++ 5 files changed, 140 insertions(+), 3 deletions(-) create mode 100644 Documentation/filesystems/nfs/pnfs.txt
Documentation/filesystems/nfs/00-INDEX +2 -0
@@ -12,5 +12,7 @@ nfs-rdma.txt - how to install and setup the Linux NFS/RDMA client and server software nfsroot.txt - short guide on setting up a diskless box with NFS root filesystem. +pnfs.txt + - short explanation of some of the internals of the pnfs code rpc-cache.txt - introduction to the caching mechanisms in the sunrpc layer.
Documentation/filesystems/nfs/pnfs.txt new file +48 -0
48 lines not on this page · raw message
fs/nfs/Kconfig +1 -1
@@ -78,7 +78,7 @@ config NFS_V4_1 depends on NFS_V4 && EXPERIMENTAL help This option enables support for minor version 1 of the NFSv4 protocol - (draft-ietf-nfsv4-minorversion1) in the kernel's NFS client. + (RFC 5661) in the kernel's NFS client. If unsure, say N.
fs/nfs/pnfs.c +80 -2
105 lines not on this page · raw message
fs/nfs/pnfs.h +9 -0
@@ -36,8 +36,17 @@/* Per-layout driver specific registration structure */ struct pnfs_layoutdriver_type { +struct list_head pnfs_tblid; +const u32 id; +const char *name; +struct module *owner; +int (*initialize_mountpoint) (struct nfs_server *); +int (*uninitialize_mountpoint) (struct nfs_server *); }; +extern int pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *); +extern void pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *); +void set_pnfs_layoutdriver (struct nfs_server *,u32 id);void unset_pnfs_layoutdriver (struct nfs_server *);
Signature · 1 line
1.7.2.1
Re: [PATCH 07/12] RFC: pnfs: full mount/umount infrastructure
Quoted 21 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Allow a module implementing a layout type to register, and have its mount/umount routines called for filesystems that the server declares support it.
Signed-off-by TBD - melding/reorganization of several patches--- Documentation/filesystems/nfs/00-INDEX | 2 + Documentation/filesystems/nfs/pnfs.txt | 48 ++++++++++++++++++ fs/nfs/Kconfig | 2 +- fs/nfs/pnfs.c | 82 +++++++++++++++++++++++++++++++- fs/nfs/pnfs.h | 9 ++++ 5 files changed, 140 insertions(+), 3 deletions(-) create mode 100644 Documentation/filesystems/nfs/pnfs.txtDocumentation/filesystems/nfs/00-INDEX +0
-0
Documentation/filesystems/nfs/00-INDEX nfs-rdma.txt · line 16
Quoted 5 lines
Documentation/filesystems/nfs/00-INDEX +2
-0@@ -12,5 +12,7 @@ nfs-rdma.txt - how to install and setup the Linux NFS/RDMA client and server software nfsroot.txt - short guide on setting up a diskless box with NFS root filesystem. +pnfs.txt + - short explanation of some of the internals of the pnfs code
that is, pnfs _client_ code...
Quoted 83 lines
Documentation/filesystems/nfs/00-INDEX +0
-0rpc-cache.txt - introduction to the caching mechanisms in the sunrpc layer.Documentation/filesystems/nfs/pnfs.txt new file +48
-0@@ -0,0 +1,48 @@ +Reference counting in pnfs: +========================== + +The are several inter-related caches. We have layouts which can +reference multiple devices, each of which can reference multiple data servers. +Each data server can be referenced by multiple devices. Each device +can be referenced by multiple layouts. To keep all of this straight, +we need to reference count. + + +struct pnfs_layout_hdr +---------------------- +The on-the-wire command LAYOUTGET corresponds to struct +pnfs_layout_segment, usually referred to by the variable name lseg. +Each nfs_inode may hold a pointer to a cache of of these layout +segments in nfsi->layout, of type struct pnfs_layout_hdr. + +We reference the header for the inode pointing to it, across each +outstanding RPC call that references it (LAYOUTGET, fs/nfs/LAYOUTRETURN, +LAYOUTCOMMIT), and for each lseg held within. + +Each header is also (when non-empty) put on a list associated with +struct nfs_client (cl_layouts). Being put on this list does not bump +the reference count, as the layout is kept around by the lseg that +keeps it in the list. + +deviceid_cache +-------------- +lsegs reference device ids, which are resolved per nfs_client and +layout driver type. The device ids are held in a RCU cache (struct +nfs4_deviceid_cache). The cache itself is referenced across each +mount. The entries (struct nfs4_deviceid) themselves are held across +the lifetime of each lseg referencing them. + +RCU is used because the deviceid is basically a write once, read many +data structure. The hlist size of 32 buckets needs better +justification, but seems reasonable given that we can have multiple +deviceid's per filesystem, and multiple filesystems per nfs_client. + +The hash code is copied from the nfsd code base. A discussion of +hashing and variations of this algorithm can be found at: +http://groups.google.com/group/comp.lang.c/browse_thread/thread/9522965e2b8d3809 + +data server cache +----------------- +file driver devices refer to data servers, which are kept in a module +level cache. Its reference is held over the lifetime of the deviceid +pointing to it.fs/nfs/Kconfig +1
-1@@ -78,7 +78,7 @@ config NFS_V4_1 depends on NFS_V4 && EXPERIMENTAL help This option enables support for minor version 1 of the NFSv4 protocol - (draft-ietf-nfsv4-minorversion1) in the kernel's NFS client. + (RFC 5661) in the kernel's NFS client. If unsure, say N.fs/nfs/pnfs.c +5
-1@@ -32,16 +32,53 @@#define NFSDBG_FACILITY NFSDBG_PNFS -/* STUB that returns the equivalent of "no module found" */ +/* Locking: +* +* pnfs_spinlock: +* protects pnfs_modules_tbl. +*/
fs/nfs/pnfs.c line 49
Quoted 10 lines
fs/nfs/pnfs.c +10
-0+static DEFINE_SPINLOCK (pnfs_spinlock); + +/* +* pnfs_modules_tbl holds all pnfs modules +*/ +static LIST_HEAD (pnfs_modules_tbl); + +/* Return the registered pnfs layout driver module matching given id */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver_locked (u32 id) {
nit: the curly brace should be moved down a line
fs/nfs/pnfs.c line 52
Quoted 3 lines
fs/nfs/pnfs.c +3
-0+struct pnfs_layoutdriver_type *local; + +dprintk ("PNFS: %s: Searching for %u\n" , __func__, id);
I'd move this printk down, before returning and print the result as well.
fs/nfs/pnfs.c line 56
Quoted 4 lines
fs/nfs/pnfs.c +4
-0+list_for_each_entry (local, &pnfs_modules_tbl, pnfs_tblid) +if (local->id == id) { +if (!try_module_get (local->owner)) + local =NULL ;
If this happens (e.g. in the case the module exited without calling pnfs_unregister_layoutdriver) another (or a different instance) layout driver might have registered on the same id so we need to keep looking.
fs/nfs/pnfs.c line 62
Quoted 6 lines
fs/nfs/pnfs.c +6
-0+ goto out; + } + local =NULL ; +out: +return local; +}
how about the following?
static struct pnfs_layoutdriver_type *find_pnfs_driver_locked(u32 id)
{
struct pnfs_layoutdriver_type *local, *found = NULL;
list_for_each_entry (local, &pnfs_modules_tbl, pnfs_tblid)
if (local->id == id) {
if (!try_module_get(local->owner))
continue;
found = local;
break;
}
out:
dprintk("%s: layout_type %u found %p\n", __func__, id, found);
return found;
}Quoted 38 lines
fs/nfs/pnfs.c +18
-1+static struct pnfs_layoutdriver_type *find_pnfs_driver (u32 id) { -return NULL ; +struct pnfs_layoutdriver_type *local; + +spin_lock (&pnfs_spinlock); + local =find_pnfs_driver_locked (id); +spin_unlock (&pnfs_spinlock); +return local; }/* Unitialize a mountpoint in a layout driver */ void unset_pnfs_layoutdriver (struct nfs_server *nfss) { +if (nfss->pnfs_curr_ld) { + nfss->pnfs_curr_ld->uninitialize_mountpoint (nfss); +module_put (nfss->pnfs_curr_ld->owner); + } nfss->pnfs_curr_ld =NULL ; }@@ -68,6 +105,13 @@ set_pnfs_layoutdriver(struct nfs_server *server, u32 id) goto out_no_driver; } } +if (ld_type->initialize_mountpoint (server)) { +printk (KERN_ERR +"%s: Error initializing mount point for layout driver %u.\n" , + __func__, id); +module_put (ld_type->owner); + goto out_no_driver; + } server->pnfs_curr_ld = ld_type;dprintk ("%s: pNFS module for %u set\n" , __func__, id);return ;
fs/nfs/pnfs.c out_no_driver: · line 129
Quoted 10 lines
fs/nfs/pnfs.c +7
-0@@ -76,3 +120,37 @@ out_no_driver:dprintk ("%s: Using NFSv4 I/O\n" , __func__); server->pnfs_curr_ld =NULL ; } + +int +pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *ld_type) +{ +int status = -EINVAL ; +struct pnfs_layoutdriver_type *tmp; +
Since we're relying on the fact the ld_type->id != 0 let's add
BUG_ON(ld_type->id == 0);
Benny
Quoted 49 lines
fs/nfs/pnfs.c +27
-0+spin_lock (&pnfs_spinlock); + tmp =find_pnfs_driver_locked (ld_type->id); +if (!tmp) { +list_add (&ld_type->pnfs_tblid, &pnfs_modules_tbl); + status =0 ; +dprintk ("%s Registering id:%u name:%s\n" , __func__, ld_type->id, + ld_type->name); + }else { +module_put (tmp->owner); +printk (KERN_ERR "%s Module with id %d already loaded!\n" , + __func__, ld_type->id); + } +spin_unlock (&pnfs_spinlock); + +return status; +} +EXPORT_SYMBOL_GPL (pnfs_register_layoutdriver); + +void +pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *ld_type) +{ +dprintk ("%s Deregistering id:%u\n" , __func__, ld_type->id); +spin_lock (&pnfs_spinlock); +list_del (&ld_type->pnfs_tblid); +spin_unlock (&pnfs_spinlock); +} +EXPORT_SYMBOL_GPL (pnfs_unregister_layoutdriver);fs/nfs/pnfs.h +9
-0@@ -36,8 +36,17 @@/* Per-layout driver specific registration structure */ struct pnfs_layoutdriver_type { +struct list_head pnfs_tblid; +const u32 id; +const char *name; +struct module *owner; +int (*initialize_mountpoint) (struct nfs_server *); +int (*uninitialize_mountpoint) (struct nfs_server *); }; +extern int pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *); +extern void pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *); +void set_pnfs_layoutdriver (struct nfs_server *,u32 id);void unset_pnfs_layoutdriver (struct nfs_server *);
Re: [PATCH 07/12] RFC: pnfs: full mount/umount infrastructure
Quoted 17 lines · Benny Halevy
Quoted 16 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Allow a module implementing a layout type to register, and have its mount/umount routines called for filesystems that the server declares support it.
Signed-off-by TBD - melding/reorganization of several patches---
Documentation/filesystems/nfs/00-INDEX | 2 +Documentation/filesystems/nfs/pnfs.txt| 48 ++++++++++++++++++fs/nfs/Kconfig| 2 +-
fs/nfs/pnfs.c | 82 +++++++++++++++++++++++++++++++-fs/nfs/pnfs.h| 9 ++++5 files changed, 140 insertions(+), 3 deletions(-)create mode 100644
Documentation/filesystems/nfs/pnfs.txt
Quoted 13 lines
Quoted 10 lines
Documentation/filesystems/nfs/00-INDEX +0
-0@@ -12,5 +12,7 @@ nfs-rdma.txt- how to install and setup the Linux NFS/RDMA client and server software nfsroot.txt
- short guide on setting up a diskless box with NFS root filesystem. +pnfs.txt
+ - short explanation of some of the internals of the pnfs codethat is, pnfs _client_ code...
OK
Quoted 87 lines
Quoted 87 lines
rpc-cache.txt
- introduction to the caching mechanisms in the sunrpc layer.Documentation/filesystems/nfs/pnfs.txt new file +48
-0@@ -0,0 +1,48 @@ +Reference counting in pnfs: +========================== + +The are several inter-related caches. We have layouts which can +reference multiple devices, each of which can reference multiple data servers. +Each data server can be referenced by multiple devices. Each device +can be referenced by multiple layouts. To keep all of this straight, +we need to reference count. + + +struct pnfs_layout_hdr +---------------------- +The on-the-wire command LAYOUTGET corresponds to struct +pnfs_layout_segment, usually referred to by the variable name lseg. +Each nfs_inode may hold a pointer to a cache of of these layout +segments in nfsi->layout, of type struct pnfs_layout_hdr. + +We reference the header for the inode pointing to it, across each +outstanding RPC call that references it (LAYOUTGET, fs/nfs/LAYOUTRETURN, +LAYOUTCOMMIT), and for each lseg held within. + +Each header is also (when non-empty) put on a list associated with +struct nfs_client (cl_layouts). Being put on this list does not bump +the reference count, as the layout is kept around by the lseg that +keeps it in the list. + +deviceid_cache +-------------- +lsegs reference device ids, which are resolved per nfs_client and +layout driver type. The device ids are held in a RCU cache (struct +nfs4_deviceid_cache). The cache itself is referenced across each +mount. The entries (struct nfs4_deviceid) themselves are held across +the lifetime of each lseg referencing them. + +RCU is used because the deviceid is basically a write once, read many +data structure. The hlist size of 32 buckets needs better +justification, but seems reasonable given that we can have multiple +deviceid's per filesystem, and multiple filesystems per nfs_client. + +The hash code is copied from the nfsd code base. A discussion of +hashing and variations of this algorithm can be found at: +http://groups.google.com/group/comp.lang.c/browse_thread/thread/9522965e2b8d3809 + +data server cache +----------------- +file driver devices refer to data servers, which are kept in a module +level cache. Its reference is held over the lifetime of the deviceid +pointing to it.fs/nfs/Kconfig +0
-0@@ -78,7 +78,7 @@ config NFS_V4_1depends on
NFS_V4&& EXPERIMENTAL
help
This option enables support for minor version 1 of the NFSv4 protocol
- (draft-ietf-nfsv4-minorversion1) in the kernel's NFS client.
+ (RFC 5661) in the kernel's NFS client.If unsure, say N.
fs/nfs/pnfs.c +9
-1@@ -32,16 +32,53 @@ #define NFSDBG_FACILITY NFSDBG_PNFS -/* STUB that returns the equivalent of "no module found" */ +/* Locking: +* +* pnfs_spinlock: +* protects pnfs_modules_tbl. +*/ +static DEFINE_SPINLOCK (pnfs_spinlock); + +/* +* pnfs_modules_tbl holds all pnfs modules
Quoted 8 lines
fs/nfs/pnfs.c line 49
Quoted 6 lines
fs/nfs/pnfs.c +6
-0+*/ +static LIST_HEAD (pnfs_modules_tbl); + +/* Return the registered pnfs layout driver module matching given id */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver_locked (u32 id) {nit: the curly brace should be moved down a line
OK
Quoted 8 lines
fs/nfs/pnfs.c line 52
Quoted 4 lines
fs/nfs/pnfs.c +3
-0+ struct pnfs_layoutdriver_type *local; + + dprintk("PNFS: %s: Searching for %u\n", __func__, id);I'd move this printk down, before returning and print the result as well.
OK
Quoted 9 lines
fs/nfs/pnfs.c line 56
Quoted 4 lines
fs/nfs/pnfs.c +4
-0+ list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) + if (local->id == id) { + if (!try_module_get(local->owner)) + local = NULL;If this happens (e.g. in the case the module exited without calling
pnfs_unregister_layoutdriver) another (or a different instance) layout driver might have registered on the same id so we need to keep looking.
There can only be a single driver registered with a particular id. So continuing the search is pointless.
Quoted 80 lines
fs/nfs/pnfs.c line 62
Quoted 7 lines
fs/nfs/pnfs.c +6
-0+ goto out; + } + local = NULL; +out: + return local; +}how about the following?
static struct
pnfs_layoutdriver_type*find_pnfs_driver_locked(u32 id)
{
structpnfs_layoutdriver_type*local, *found = NULL;list_for_each_entry (local,
&pnfs_modules_tbl,pnfs_tblid) if (local->id== id) {
if (!try_module_get(local->owner)) continue;
found = local;
break;
}
out:
dprintk("%s:layout_type%u found %p\n",__func__, id, found); return found;
}Quoted 38 lines
fs/nfs/pnfs.c +1
-0+static struct
pnfs_layoutdriver_type*
find_pnfs_driver(u32 id) {
- return NULL;
+ struct pnfs_layoutdriver_type *local;
+
+spin_lock(&pnfs_spinlock);
+ local =find_pnfs_driver_locked(id);
+spin_unlock(&pnfs_spinlock);
+ return local;
}/* Unitialize a mountpoint in a layout driver */
void
unset_pnfs_layoutdriver(structnfs_server*nfss)
{
+ if (nfss->pnfs_curr_ld) {
+nfss->pnfs_curr_ld->uninitialize_mountpoint(nfss);
+module_put(nfss->pnfs_curr_ld->owner);
+ }
nfss->pnfs_curr_ld= NULL;
}fs/nfs/pnfs.c +7
-0@@ -68,6 +105,13 @@ set_pnfs_layoutdriver(struct nfs_server *server, u32 id) goto out_no_driver; } } + if (ld_type->initialize_mountpoint(server)) { + printk(KERN_ERR + "%s: Error initializing mount point for layout driver %u.\n", + __func__, id); + module_put(ld_type->owner); + goto out_no_driver; + } server->pnfs_curr_ld = ld_type; dprintk("%s: pNFS module for %u set\n", __func__, id); return;fs/nfs/pnfs.c out_no_driver: · line 129
Quoted 10 lines
fs/nfs/pnfs.c +7
-0@@ -76,3 +120,37 @@ out_no_driver: dprintk("%s: Using NFSv4 I/O\n", __func__); server->pnfs_curr_ld = NULL; } + +int +pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *ld_type) +{ + int status = -EINVAL; + struct pnfs_layoutdriver_type *tmp; +Since we're relying on the fact the
ld_type->id!= 0 let's add
BUG_ON(ld_type->id == 0);
Quoted 1 line
Benny
OK.
Fred
Quoted 55 lines
Quoted 40 lines
fs/nfs/pnfs.c +27
-0+ spin_lock(&pnfs_spinlock); + tmp = find_pnfs_driver_locked(ld_type->id); + if (!tmp) { + list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); + status = 0; + dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, + ld_type->name); + } else { + module_put(tmp->owner); + printk(KERN_ERR "%s Module with id %d already loaded!\n", + __func__, ld_type->id); + } + spin_unlock(&pnfs_spinlock); + + return status; +} +EXPORT_SYMBOL_GPL (pnfs_register_layoutdriver); + +void +pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *ld_type) +{ + dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); + spin_lock(&pnfs_spinlock); + list_del(&ld_type->pnfs_tblid); + spin_unlock(&pnfs_spinlock); +} +EXPORT_SYMBOL_GPL (pnfs_unregister_layoutdriver);fs/nfs/pnfs.h +4
-0@@ -36,8 +36,17 @@ /* Per-layout driver specific registration structure */ struct pnfs_layoutdriver_type { + struct list_head pnfs_tblid; + const u32 id; + const char *name; + struct module *owner;fs/nfs/pnfs.h line 52
Quoted 10 lines
fs/nfs/pnfs.h +5
-0+ int (*initialize_mountpoint) (struct nfs_server *); + int (*uninitialize_mountpoint) (struct nfs_server *); }; +extern int pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *); +extern void pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *); + void set_pnfs_layoutdriver(struct nfs_server *, u32 id); void unset_pnfs_layoutdriver(struct nfs_server *);--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 07/12] RFC: pnfs: full mount/umount infrastructure
Quoted 151 lines · Fred Isaman
Quoted 17 lines · Benny Halevy
Quoted 16 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Allow a module implementing a layout type to register, and have its mount/umount routines called for filesystems that the server declares support it.
Signed-off-by TBD - melding/reorganization of several patches--- Documentation/filesystems/nfs/00-INDEX | 2 + Documentation/filesystems/nfs/pnfs.txt | 48 ++++++++++++++++++ fs/nfs/Kconfig | 2 +- fs/nfs/pnfs.c | 82 +++++++++++++++++++++++++++++++- fs/nfs/pnfs.h | 9 ++++ 5 files changed, 140 insertions(+), 3 deletions(-) create mode 100644 Documentation/filesystems/nfs/pnfs.txtQuoted 13 lines
Documentation/filesystems/nfs/00-INDEX nfs-rdma.txt · line 16
Quoted 10 lines
Documentation/filesystems/nfs/00-INDEX +2
-0@@ -12,5 +12,7 @@ nfs-rdma.txt - how to install and setup the Linux NFS/RDMA client and server software nfsroot.txt - short guide on setting up a diskless box with NFS root filesystem. +pnfs.txt + - short explanation of some of the internals of the pnfs codethat is, pnfs _client_ code...
OK
Quoted 87 lines
Quoted 87 lines
Documentation/filesystems/nfs/00-INDEX +0
-0rpc-cache.txt - introduction to the caching mechanisms in the sunrpc layer.Documentation/filesystems/nfs/pnfs.txt new file +48
-0@@ -0,0 +1,48 @@ +Reference counting in pnfs: +========================== + +The are several inter-related caches. We have layouts which can +reference multiple devices, each of which can reference multiple data servers. +Each data server can be referenced by multiple devices. Each device +can be referenced by multiple layouts. To keep all of this straight, +we need to reference count. + + +struct pnfs_layout_hdr +---------------------- +The on-the-wire command LAYOUTGET corresponds to struct +pnfs_layout_segment, usually referred to by the variable name lseg. +Each nfs_inode may hold a pointer to a cache of of these layout +segments in nfsi->layout, of type struct pnfs_layout_hdr. + +We reference the header for the inode pointing to it, across each +outstanding RPC call that references it (LAYOUTGET, fs/nfs/LAYOUTRETURN, +LAYOUTCOMMIT), and for each lseg held within. + +Each header is also (when non-empty) put on a list associated with +struct nfs_client (cl_layouts). Being put on this list does not bump +the reference count, as the layout is kept around by the lseg that +keeps it in the list. + +deviceid_cache +-------------- +lsegs reference device ids, which are resolved per nfs_client and +layout driver type. The device ids are held in a RCU cache (struct +nfs4_deviceid_cache). The cache itself is referenced across each +mount. The entries (struct nfs4_deviceid) themselves are held across +the lifetime of each lseg referencing them. + +RCU is used because the deviceid is basically a write once, read many +data structure. The hlist size of 32 buckets needs better +justification, but seems reasonable given that we can have multiple +deviceid's per filesystem, and multiple filesystems per nfs_client. + +The hash code is copied from the nfsd code base. A discussion of +hashing and variations of this algorithm can be found at: +http://groups.google.com/group/comp.lang.c/browse_thread/thread/9522965e2b8d3809 + +data server cache +----------------- +file driver devices refer to data servers, which are kept in a module +level cache. Its reference is held over the lifetime of the deviceid +pointing to it.fs/nfs/Kconfig +1
-1@@ -78,7 +78,7 @@ config NFS_V4_1 depends on NFS_V4 && EXPERIMENTAL help This option enables support for minor version 1 of the NFSv4 protocol - (draft-ietf-nfsv4-minorversion1) in the kernel's NFS client. + (RFC 5661) in the kernel's NFS client. If unsure, say N.fs/nfs/pnfs.c +9
-1@@ -32,16 +32,53 @@#define NFSDBG_FACILITY NFSDBG_PNFS -/* STUB that returns the equivalent of "no module found" */ +/* Locking: +* +* pnfs_spinlock: +* protects pnfs_modules_tbl. +*/ +static DEFINE_SPINLOCK (pnfs_spinlock); + +/* +* pnfs_modules_tbl holds all pnfs modules Quoted 8 lines
fs/nfs/pnfs.c line 49
Quoted 6 lines
fs/nfs/pnfs.c +6
-0+*/ +static LIST_HEAD (pnfs_modules_tbl); + +/* Return the registered pnfs layout driver module matching given id */ +static struct pnfs_layoutdriver_type * +find_pnfs_driver_locked (u32 id) {nit: the curly brace should be moved down a line
OK
Quoted 8 lines
fs/nfs/pnfs.c line 52
Quoted 4 lines
fs/nfs/pnfs.c +3
-0+struct pnfs_layoutdriver_type *local; + +dprintk ("PNFS: %s: Searching for %u\n" , __func__, id);I'd move this printk down, before returning and print the result as well.
OK
Quoted 5 lines
Quoted 4 lines
fs/nfs/pnfs.c +4
-0+list_for_each_entry (local, &pnfs_modules_tbl, pnfs_tblid) +if (local->id == id) { +if (!try_module_get (local->owner)) + local =NULL ;
Quoted 9 lines
Quoted 4 lines
If this happens (e.g. in the case the module exited without calling
pnfs_unregister_layoutdriver) another (or a different instance) layout driver might have registered on the same id so we need to keep looking.There can only be a single driver registered with a particular id. So continuing the search is pointless.
After this conditions happens, find_pnfs_driver_locked will not find the current one so next time pnfs_register_layoutdriver is called, it may register a new layout driver with the same id
and then there will be two entries with the same id. One, for which try_module_get returns zero and another one which you will never find if you don't keep looking for it.
Benny
Quoted 141 lines
Quoted 78 lines
fs/nfs/pnfs.c line 62
Quoted 7 lines
fs/nfs/pnfs.c +6
-0+ goto out; + } + local =NULL ; +out: +return local; +}how about the following?
static struct
pnfs_layoutdriver_type*find_pnfs_driver_locked(u32 id){ struct pnfs_layoutdriver_type *local, *found = NULL; list_for_each_entry (local, &pnfs_modules_tbl, pnfs_tblid) if (local->id == id) { if (!try_module_get(local->owner)) continue; found = local; break; } out: dprintk("%s: layout_type %u found %p\n", __func__, id, found); return found; }Quoted 38 lines
fs/nfs/pnfs.c +18
-1+static struct pnfs_layoutdriver_type *find_pnfs_driver (u32 id) { -return NULL ; +struct pnfs_layoutdriver_type *local; + +spin_lock (&pnfs_spinlock); + local =find_pnfs_driver_locked (id); +spin_unlock (&pnfs_spinlock); +return local; }/* Unitialize a mountpoint in a layout driver */ void unset_pnfs_layoutdriver (struct nfs_server *nfss) { +if (nfss->pnfs_curr_ld) { + nfss->pnfs_curr_ld->uninitialize_mountpoint (nfss); +module_put (nfss->pnfs_curr_ld->owner); + } nfss->pnfs_curr_ld =NULL ; }@@ -68,6 +105,13 @@ set_pnfs_layoutdriver(struct nfs_server *server, u32 id) goto out_no_driver; } } +if (ld_type->initialize_mountpoint (server)) { +printk (KERN_ERR +"%s: Error initializing mount point for layout driver %u.\n" , + __func__, id); +module_put (ld_type->owner); + goto out_no_driver; + } server->pnfs_curr_ld = ld_type;dprintk ("%s: pNFS module for %u set\n" , __func__, id);return ;fs/nfs/pnfs.c out_no_driver: · line 129
Quoted 10 lines
fs/nfs/pnfs.c +7
-0@@ -76,3 +120,37 @@ out_no_driver:dprintk ("%s: Using NFSv4 I/O\n" , __func__); server->pnfs_curr_ld =NULL ; } + +int +pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *ld_type) +{ +int status = -EINVAL ; +struct pnfs_layoutdriver_type *tmp; +Since we're relying on the fact the
ld_type->id!= 0 let's addQuoted 3 lines
BUG_ON(ld_type->id == 0);Benny
OK.
Fred
Quoted 55 lines
Quoted 40 lines
fs/nfs/pnfs.c +27
-0+spin_lock (&pnfs_spinlock); + tmp =find_pnfs_driver_locked (ld_type->id); +if (!tmp) { +list_add (&ld_type->pnfs_tblid, &pnfs_modules_tbl); + status =0 ; +dprintk ("%s Registering id:%u name:%s\n" , __func__, ld_type->id, + ld_type->name); + }else { +module_put (tmp->owner); +printk (KERN_ERR "%s Module with id %d already loaded!\n" , + __func__, ld_type->id); + } +spin_unlock (&pnfs_spinlock); + +return status; +} +EXPORT_SYMBOL_GPL (pnfs_register_layoutdriver); + +void +pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *ld_type) +{ +dprintk ("%s Deregistering id:%u\n" , __func__, ld_type->id); +spin_lock (&pnfs_spinlock); +list_del (&ld_type->pnfs_tblid); +spin_unlock (&pnfs_spinlock); +} +EXPORT_SYMBOL_GPL (pnfs_unregister_layoutdriver);fs/nfs/pnfs.h +4
-0@@ -36,8 +36,17 @@/* Per-layout driver specific registration structure */ struct pnfs_layoutdriver_type { +struct list_head pnfs_tblid; +const u32 id; +const char *name; +struct module *owner;fs/nfs/pnfs.h line 52
Quoted 10 lines
fs/nfs/pnfs.h +5
-0+int (*initialize_mountpoint) (struct nfs_server *); +int (*uninitialize_mountpoint) (struct nfs_server *); }; +extern int pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *); +extern void pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *); +void set_pnfs_layoutdriver (struct nfs_server *,u32 id);void unset_pnfs_layoutdriver (struct nfs_server *);--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
PATCH 8/12 RFC: pnfs: filelayout: introduce minimal file layout driver
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
This driver just registers itself and supplies trivial mount/umount functions.
--- fs/nfs/Makefile | 3 ++ fs/nfs/nfs4filelayout.c | 78 +++++++++++++++++++++++++++++++++++++++++++++++ include/linux/nfs_fs.h | 1 + 3 files changed, 82 insertions(+), 0 deletions(-) create mode 100644 fs/nfs/nfs4filelayout.c
fs/nfs/Makefile +3 -0
@@ -18,3 +18,6 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \ nfs-$(CONFIG_NFS_V4_1) += pnfs.o nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o + +obj-$(CONFIG_NFS_V4_1) += nfs_layout_nfsv41_files.o +nfs_layout_nfsv41_files-y := nfs4filelayout.o
fs/nfs/nfs4filelayout.c new file +78 -0
78 lines not on this page · raw message
include/linux/nfs_fs.h +1 -0
@@ -616,6 +616,7 @@ nfs_fileid_to_ino_t(u64 fileid)#define NFSDBG_MOUNT 0x0400#define NFSDBG_FSCACHE 0x0800#define NFSDBG_PNFS 0x1000 +#define NFSDBG_PNFS_LD 0x2000#define NFSDBG_ALL 0xFFFF#ifdef __KERNEL__
Signature · 1 line
1.7.2.1
PATCH 9/12 RFC: nfs: create and destroy inode's layout cache
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
At the start of the io paths, try to grab the relevant layout information. This will initiate the inode's layout cache, but stubs ensure the cache stays empty.
--- fs/nfs/file.c | 5 ++ fs/nfs/inode.c | 3 + fs/nfs/pnfs.c | 140 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/nfs/pnfs.h | 39 +++++++++++++ fs/nfs/read.c | 3 + include/linux/nfs_fs.h | 3 + 6 files changed, 193 insertions(+), 0 deletions(-)
fs/nfs/file.c +5 -0
@@ -36,6 +36,7 @@#include "internal.h" #include "iostat.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_FILE
@@ -386,6 +387,10 @@ static int nfs_write_begin(struct file *file, struct address_space *mapping, file->f_path.dentry->d_name.name, mapping->host->i_ino, len, (long long) pos); +pnfs_update_layout (mapping->host, +nfs_file_open_context (file), +IOMODE_RW ); + start:/* * Prevent starvation issues if someone is doing a consistency
fs/nfs/inode.c +3 -0
@@ -48,6 +48,7 @@#include "internal.h" #include "fscache.h" #include "dns_resolve.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_VFS
@@ -1409,6 +1410,7 @@ void nfs4_evict_inode(struct inode *inode) {truncate_inode_pages (&inode->i_data,0 );end_writeback (inode); +pnfs_destroy_layout (NFS_I (inode));/* If we are holding a delegation, return it! */ nfs_inode_return_delegation_noreclaim (inode);/* First call standard NFS clear_inode() code */
@@ -1446,6 +1448,7 @@ static inline void nfs4_init_once(struct nfs_inode *nfsi) nfsi->delegation =NULL ; nfsi->delegation_state =0 ;init_rwsem (&nfsi->rwsem); + nfsi->layout =NULL ;#endif }
fs/nfs/pnfs.c +140 -0
143 lines not on this page · raw message
fs/nfs/pnfs.h +39 -0
@@ -34,6 +34,11 @@#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4" +enum { +NFS_LAYOUT_RO_FAILED =0 ,/* get ro layout failed stop trying */ +NFS_LAYOUT_RW_FAILED ,/* get rw layout failed stop trying */ +}; +/* Per-layout driver specific registration structure */ struct pnfs_layoutdriver_type {struct list_head pnfs_tblid;
@@ -44,14 +49,48 @@ struct pnfs_layoutdriver_type {int (*uninitialize_mountpoint) (struct nfs_server *); }; +struct pnfs_layout_hdr { +unsigned long refcount; +unsigned long state; +struct inode *inode; +}; +extern int pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *);extern void pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *); +struct pnfs_layout_segment * +pnfs_update_layout (struct inode *ino,struct nfs_open_context *ctx, +enum pnfs_iomode access_type);void set_pnfs_layoutdriver (struct nfs_server *,u32 id);void unset_pnfs_layoutdriver (struct nfs_server *); +void pnfs_destroy_layout (struct nfs_inode *); + + +static inline int lo_fail_bit (u32 iomode) +{ +return iomode ==IOMODE_RW ? +NFS_LAYOUT_RW_FAILED :NFS_LAYOUT_RO_FAILED ; +} + +/* Return true if a layout driver is being used for this mountpoint */ +static inline int pnfs_enabled_sb (struct nfs_server *nfss) +{ +return nfss->pnfs_curr_ld !=NULL ; +}#else /* CONFIG_NFS_V4_1 */ +static inline void pnfs_destroy_layout (struct nfs_inode *nfsi) +{ +} + +static inline struct pnfs_layout_segment * +pnfs_update_layout (struct inode *ino,struct nfs_open_context *ctx, +enum pnfs_iomode access_type) +{ +return NULL ; +} +static inline void set_pnfs_layoutdriver (struct nfs_server *s,u32 id) { }
fs/nfs/read.c +3 -0
@@ -25,6 +25,7 @@#include "internal.h" #include "iostat.h" #include "fscache.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_PAGECACHE
@@ -121,6 +122,7 @@ int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, len =nfs_page_length (page);if (len ==0 )return nfs_return_empty_page (page); +pnfs_update_layout (inode, ctx,IOMODE_READ ); new =nfs_create_request (ctx, inode, page,0 , len);if (IS_ERR (new)) {unlock_page (page);
@@ -625,6 +627,7 @@ int nfs_readpages(struct file *filp, struct address_space *mapping,if (ret ==0 ) goto read_complete;/* all pages were read */ +pnfs_update_layout (inode, desc.ctx,IOMODE_READ );if (rsize <PAGE_CACHE_SIZE )nfs_pageio_init (&pgio, inode, nfs_pagein_multi, rsize,0 );else
include/linux/nfs_fs.h +3 -0
@@ -188,6 +188,9 @@ struct nfs_inode {struct nfs_delegation *delegation;fmode_t delegation_state;struct rw_semaphore rwsem; + +/* pNFS layout information */ +struct pnfs_layout_hdr *layout;#endif /* CONFIG_NFS_V4*/ #ifdef CONFIG_NFS_FSCACHE struct fscache_cookie *fscache;
Signature · 1 line
1.7.2.1
PATCH 10/12 RFC: nfs: client needs to maintain list of inodes with active layouts
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
In particular, server reboot will invalidate all layouts.
Note that in order to have an active layout, we must get a successful response from the server. To avoid adding that machinery, this patch just includes a stub that fakes up a successful return. Since the layout is never referenced for io, this is not a problem.
--- fs/nfs/client.c | 4 +- fs/nfs/nfs4state.c | 2 + fs/nfs/pnfs.c | 131 +++++++++++++++++++++++++++++++++++++++++++- fs/nfs/pnfs.h | 14 +++++ include/linux/nfs_fs_sb.h | 1 + 5 files changed, 148 insertions(+), 4 deletions(-)
fs/nfs/client.c +3 -1
@@ -156,7 +156,9 @@ static struct nfs_client *nfs_alloc_client(const struct nfs_client_initdata *cl_ cred =rpc_lookup_machine_cred ();if (!IS_ERR (cred)) clp->cl_machine_cred = cred; - +#if defined(CONFIG_NFS_V4_1 ) +INIT_LIST_HEAD (&clp->cl_layouts); +#endif nfs_fscache_get_client_cookie (clp);return clp;
fs/nfs/nfs4state.c +2 -0
@@ -53,6 +53,7 @@#include "callback.h" #include "delegation.h" #include "internal.h" +#include "pnfs.h" #define OPENOWNER_POOL_SIZE 8
@@ -1447,6 +1448,7 @@ static void nfs4_state_manager(struct nfs_client *clp) }clear_bit (NFS4CLNT_CHECK_LEASE , &clp->cl_state);set_bit (NFS4CLNT_RECLAIM_REBOOT , &clp->cl_state); +pnfs_destroy_all_layouts (clp); }if (test_and_clear_bit (NFS4CLNT_CHECK_LEASE , &clp->cl_state)) {
fs/nfs/pnfs.c +128 -3
176 lines not on this page · raw message
fs/nfs/pnfs.h +14 -0
@@ -30,6 +30,13 @@#ifndef FS_NFS_PNFS_H #define FS_NFS_PNFS_H +struct pnfs_layout_segment { +struct list_head fi_list; +u32 iomode; +struct kref kref; +struct pnfs_layout_hdr *layout; +}; +#ifdef CONFIG_NFS_V4_1 #define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4"
@@ -51,6 +58,8 @@ struct pnfs_layoutdriver_type {struct pnfs_layout_hdr {unsigned long refcount; +struct list_head layouts;/* other client layouts */ +struct list_head segs;/* layout segments list */ unsigned long state;struct inode *inode; };
@@ -64,6 +73,7 @@ pnfs_update_layout(struct inode *ino, struct nfs_open_context *ctx,void set_pnfs_layoutdriver (struct nfs_server *,u32 id);void unset_pnfs_layoutdriver (struct nfs_server *);void pnfs_destroy_layout (struct nfs_inode *); +void pnfs_destroy_all_layouts (struct nfs_client *);static inline int lo_fail_bit (u32 iomode)
@@ -80,6 +90,10 @@ static inline int pnfs_enabled_sb(struct nfs_server *nfss)#else /* CONFIG_NFS_V4_1 */ +static inline void pnfs_destroy_all_layouts (struct nfs_client *clp) +{ +} +static inline void pnfs_destroy_layout (struct nfs_inode *nfsi) { }
include/linux/nfs_fs_sb.h +1 -0
@@ -82,6 +82,7 @@ struct nfs_client {/* The flags used for obtaining the clientid during EXCHANGE_ID */ u32 cl_exchange_flags;struct nfs4_session *cl_session;/* sharred session */ +struct list_head cl_layouts;#endif /* CONFIG_NFS_V4_1 */ #ifdef CONFIG_NFS_FSCACHE
Signature · 1 line
1.7.2.1
PATCH 11/12 RFC: pnfs: add LAYOUTGET and GETDEVICEINFO infrastructure
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
Add the ability to actually send LAYOUTGET and GETDEVICEINFO. This also adds in the machinery to handle layout state and the deviceid cache. Note that GETDEVICEINFO is not called directly by the generic layer. Instead it is called by the drivers while parsing the LAYOUTGET opaque data in response to an unknown device id embedded therein. Annoyingly, RFC 5661 only encodes device ids within the driver-specific opaque data.
--- fs/nfs/nfs4proc.c | 150 ++++++++++++++++++ fs/nfs/nfs4xdr.c | 302 ++++++++++++++++++++++++++++++++++++ fs/nfs/pnfs.c | 376 +++++++++++++++++++++++++++++++++++++++++---- fs/nfs/pnfs.h | 81 ++++++++++- include/linux/nfs4.h | 2 + include/linux/nfs_fs_sb.h | 1 + include/linux/nfs_xdr.h | 49 ++++++ 7 files changed, 933 insertions(+), 28 deletions(-)
fs/nfs/nfs4proc.c +150 -0
@@ -55,6 +55,7 @@#include "internal.h" #include "iostat.h" #include "callback.h" +#include "pnfs.h" #define NFSDBG_FACILITY NFSDBG_PROC
@@ -5256,6 +5257,155 @@ out:dprintk ("<-- %s status=%d\n" , __func__, status);return status; } + +static void +nfs4_layoutget_prepare (struct rpc_task *task,void *calldata) +{ +struct nfs4_layoutget *lgp = calldata; +struct inode *ino = lgp->args.inode; +struct nfs_server *server =NFS_SERVER (ino); + +dprintk ("--> %s\n" , __func__); +if (nfs4_setup_sequence (server, &lgp->args.seq_args, + &lgp->res.seq_res,0 , task)) +return ; +rpc_call_start (task); +} + +static void nfs4_layoutget_done (struct rpc_task *task,void *calldata) +{ +struct nfs4_layoutget *lgp = calldata; + +dprintk ("--> %s\n" , __func__); + +if (!nfs4_sequence_done (task, &lgp->res.seq_res)) +return ; + +/* Error handling done later using nfs4_handle_exception to get +* exponential backoff. +*/ + lgp->status = task->tk_status; +dprintk ("<-- %s\n" , __func__); +} + +static void nfs4_layoutget_release (void *calldata) +{ +struct nfs4_layoutget *lgp = calldata; + +dprintk ("--> %s\n" , __func__); +put_layout_hdr (lgp->args.inode); +if (lgp->res.layout.buf !=NULL ) +free_page ((unsigned long ) lgp->res.layout.buf); +put_nfs_open_context (lgp->args.ctx); +kfree (calldata); +dprintk ("<-- %s\n" , __func__); +} + +static const struct rpc_call_ops nfs4_layoutget_call_ops = { + .rpc_call_prepare = nfs4_layoutget_prepare, + .rpc_call_done = nfs4_layoutget_done, + .rpc_release = nfs4_layoutget_release, +}; + +static int _nfs4_proc_layoutget (struct nfs4_layoutget *lgp) +{ +struct nfs_server *server =NFS_SERVER (lgp->args.inode); +struct rpc_task *task; +struct rpc_message msg = { + .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LAYOUTGET ], + .rpc_argp = &lgp->args, + .rpc_resp = &lgp->res, + }; +struct rpc_task_setup task_setup_data = { + .rpc_client = server->client, + .rpc_message = &msg, + .callback_ops = &nfs4_layoutget_call_ops, + .callback_data = lgp, + .flags =RPC_TASK_ASYNC , + }; +int status =0 ; + +dprintk ("--> %s\n" , __func__); + + lgp->res.layout.buf = (void *)__get_free_page (GFP_NOFS ); +if (lgp->res.layout.buf ==NULL ) { +nfs4_layoutget_release (lgp); +return -ENOMEM ; + } + + lgp->res.seq_res.sr_slotid =NFS4_MAX_SLOT_TABLE ; + task =rpc_run_task (&task_setup_data); +if (IS_ERR (task)) +return PTR_ERR (task); + status =nfs4_wait_for_completion_rpc_task (task); +if (status !=0 ) + goto out; + status = lgp->status; +if (status !=0 ) + goto out; + status =pnfs_layout_process (lgp); +out: +rpc_put_task (task); +dprintk ("<-- %s status=%d\n" , __func__, status); +return status; +} + +int nfs4_proc_layoutget (struct nfs4_layoutget *lgp) +{ +struct nfs_server *server =NFS_SERVER (lgp->args.inode); +struct nfs4_exception exception = { }; +int err; +do { + err =_nfs4_proc_layoutget (lgp); +switch (err) { +case -NFS4ERR_LAYOUTTRYLATER : +case -NFS4ERR_RECALLCONFLICT : + err = -NFS4ERR_DELAY ; +/* Fall through */ +default : + err =nfs4_handle_exception (server, err, &exception); + } + }while (exception.retry); +return err; +} + +static int +_nfs4_proc_getdeviceinfo (struct nfs_server *server,struct pnfs_device *pdev) +{ +struct nfs4_getdeviceinfo_args args = { + .pdev = pdev, + }; +struct nfs4_getdeviceinfo_res res = { + .pdev = pdev, + }; +struct rpc_message msg = { + .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_GETDEVICEINFO ], + .rpc_argp = &args, + .rpc_resp = &res, + }; +int status; + +dprintk ("--> %s\n" , __func__); + status =nfs4_call_sync (server, &msg, &args, &res,0 ); +dprintk ("<-- %s status=%d\n" , __func__, status); + +return status; +} + +int nfs4_proc_getdeviceinfo (struct nfs_server *server,struct pnfs_device *pdev) +{ +struct nfs4_exception exception = { }; +int err; + +do { + err =nfs4_handle_exception (server, +_nfs4_proc_getdeviceinfo (server, pdev), + &exception); + }while (exception.retry); +return err; +} +EXPORT_SYMBOL_GPL (nfs4_proc_getdeviceinfo); +#endif /* CONFIG_NFS_V4_1 */ struct nfs4_state_recovery_ops nfs40_reboot_recovery_ops = {
fs/nfs/nfs4xdr.c +302 -0
350 lines not on this page · raw message
fs/nfs/pnfs.c +349 -27
461 lines not on this page · raw message
fs/nfs/pnfs.h +80 -1
123 lines not on this page · raw message
include/linux/nfs4.h +2 -0
@@ -545,6 +545,8 @@ enum {NFSPROC4_CLNT_SEQUENCE ,NFSPROC4_CLNT_GET_LEASE_TIME ,NFSPROC4_CLNT_RECLAIM_COMPLETE , +NFSPROC4_CLNT_LAYOUTGET , + NFSPROC4_CLNT_GETDEVICEINFO, };/* nfs41 types */
include/linux/nfs_fs_sb.h +1 -0
@@ -83,6 +83,7 @@ struct nfs_client {u32 cl_exchange_flags;struct nfs4_session *cl_session;/* sharred session */ struct list_head cl_layouts; +struct nfs4_deviceid_cache *cl_devid_cache;/* pNFS deviceid cache */ #endif /* CONFIG_NFS_V4_1 */ #ifdef CONFIG_NFS_FSCACHE
include/linux/nfs_xdr.h +49 -0
55 lines not on this page · raw message
Signature · 1 line
1.7.2.1
Re: [PATCH 11/12] RFC: pnfs: add LAYOUTGET and GETDEVICEINFO infrastructure
Quoted 9 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Add the ability to actually send LAYOUTGET and GETDEVICEINFO. This also adds in the machinery to handle layout state and the deviceid cache. Note that GETDEVICEINFO is not called directly by the generic layer. Instead it is called by the drivers while parsing the LAYOUTGET opaque data in response to an unknown device id embedded therein. Annoyingly, RFC 5661 only encodes device ids within the driver-specific opaque data.
Fred, In regard to device cache at least, this is a complete new code then what I knew before. I wish you could send a SQUASHME diff first of what has changed.
Quoted 1 line
Signed-off-by TBD - melding/reorganization of several patches
<snip>
Quoted 52 lines
fs/nfs/pnfs.c +52
-0+/* +* Device ID cache. Currently supports one layout type per struct nfs_client. +* Add layout type to the lookup key to expand to support multiple types. +*/ +int +nfs4_alloc_init_deviceid_cache (struct nfs_client *clp, +void (*free_callback)(struct nfs4_deviceid *)) +{ +struct nfs4_deviceid_cache *c; + + c =kzalloc (sizeof (struct nfs4_deviceid_cache ),GFP_KERNEL ); +if (!c) +return -ENOMEM ; +spin_lock (&clp->cl_lock); +if (clp->cl_devid_cache !=NULL ) { +atomic_inc (&clp->cl_devid_cache->dc_ref); +dprintk ("%s [kref [%d]]\n" , __func__, +atomic_read (&clp->cl_devid_cache->dc_ref)); +kfree (c); + }else { +/* kzalloc initializes hlists */ +spin_lock_init (&c->dc_lock); +atomic_set (&c->dc_ref,1 ); + c->dc_free_callback = free_callback; + clp->cl_devid_cache = c; +dprintk ("%s [new]\n" , __func__); + } +spin_unlock (&clp->cl_lock); +return 0 ; +} +EXPORT_SYMBOL_GPL (nfs4_alloc_init_deviceid_cache); + +void +nfs4_init_deviceid_node (struct nfs4_deviceid *d) +{ +INIT_HLIST_NODE (&d->de_node); +atomic_set (&d->de_ref,1 ); +} +EXPORT_SYMBOL_GPL (nfs4_init_deviceid_node); + +/* +* Called from pnfs_layoutdriver_type->free_lseg +* last layout segment reference frees deviceid +*/ +void +nfs4_put_deviceid (struct nfs4_deviceid_cache *c,struct nfs4_deviceid *devid) +{ +struct pnfs_deviceid *id = &devid->de_id; +struct nfs4_deviceid *d; +struct hlist_node *n; +long h =nfs4_deviceid_hash (id); +
fs/nfs/pnfs.c out_unlock: · line 670
Quoted 10 lines
fs/nfs/pnfs.c +10
-0+dprintk ("%s [%d]\n" , __func__,atomic_read (&devid->de_ref)); +if (!atomic_dec_and_lock (&devid->de_ref, &c->dc_lock)) +return ; + +hlist_for_each_entry_rcu (d, n, &c->dc_deviceids[h], de_node) +if (!memcmp (&d->de_id, id,sizeof (*id))) { +hlist_del_rcu (&d->de_node); +spin_unlock (&c->dc_lock); +synchronize_rcu (); + c->dc_free_callback (devid);
What?!? free the devid on every put? surly something went hey-wire
This used to be kref(ed)
Quoted 41 lines
fs/nfs/pnfs.c +41
-0+return ; + } +spin_unlock (&c->dc_lock); +} +EXPORT_SYMBOL_GPL (nfs4_put_deviceid); + +/* Find and reference a deviceid */ +struct nfs4_deviceid * +nfs4_find_get_deviceid (struct nfs4_deviceid_cache *c,struct pnfs_deviceid *id) +{ +struct nfs4_deviceid *d; +struct hlist_node *n; +long hash =nfs4_deviceid_hash (id); + +dprintk ("--> %s hash %ld\n" , __func__, hash); +rcu_read_lock (); +hlist_for_each_entry_rcu (d, n, &c->dc_deviceids[hash], de_node) { +if (!memcmp (&d->de_id, id,sizeof (*id))) { +if (!atomic_inc_not_zero (&d->de_ref)) { + goto fail; + }else { +rcu_read_unlock (); +return d; + } + } + } +fail: +rcu_read_unlock (); +return NULL ; +} +EXPORT_SYMBOL_GPL (nfs4_find_get_deviceid); + +/* +* Add a deviceid to the cache. +* GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new +*/ +struct nfs4_deviceid * +nfs4_add_deviceid (struct nfs4_deviceid_cache *c,struct nfs4_deviceid *new) +{ +struct nfs4_deviceid *d; +struct hlist_node *n;
fs/nfs/pnfs.c out_unlock: · line 721
Quoted 10 lines
fs/nfs/pnfs.c +10
-0+long hash =nfs4_deviceid_hash (&new->de_id); + +dprintk ("--> %s hash %ld\n" , __func__, hash); +spin_lock (&c->dc_lock); +hlist_for_each_entry_rcu (d, n, &c->dc_deviceids[hash], de_node) { +if (!memcmp (&d->de_id, &new->de_id,sizeof (new->de_id))) { +spin_unlock (&c->dc_lock); +dprintk ("%s [discard]\n" , __func__); + c->dc_free_callback (new); +return d;
I am totally confused and should be smacked on the head strong. As said above one alloc_lseg adds the first, second comes in, but same deviceid, finds the first and returns it (no refcounting). Then a first free_lseg nfs4_put_deviceid is called, so the second lseg holds free memory?
Quoted 6 lines
fs/nfs/pnfs.c +6
-0+ } + } +hlist_add_head_rcu (&new->de_node, &c->dc_deviceids[hash]); +spin_unlock (&c->dc_lock); +dprintk ("%s [new]\n" , __func__); +return new;
fs/nfs/pnfs.c out_unlock: · line 737
Quoted 10 lines
fs/nfs/pnfs.c +10
-0+} +EXPORT_SYMBOL_GPL (nfs4_add_deviceid); + +void +nfs4_put_deviceid_cache (struct nfs_client *clp) +{ +struct nfs4_deviceid_cache *local = clp->cl_devid_cache; + +dprintk ("--> %s cl_devid_cache %p\n" , __func__, clp->cl_devid_cache); +if (atomic_dec_and_lock (&local->dc_ref, &clp->cl_lock)) {
Don't you want to make sure there are no hanging devices on this cache. Please note that an add or find does not take the cache reference only server_init does, right?
fs/nfs/pnfs.c out_unlock: · line 743
Quoted 6 lines
fs/nfs/pnfs.c +6
-0+ clp->cl_devid_cache =NULL ; +spin_unlock (&clp->cl_lock); +kfree (local); + } +} +EXPORT_SYMBOL_GPL (nfs4_put_deviceid_cache);
Note to self. Test code alot.
Fred:
I don't understand. A cache property I'm looking for and which I do not See in this code, perhaps because I'm clueless.
(And to be honest I have not understood it before as well, since I never
actually ran with that code I posted.)
Is:
- An _add_ to queue should magically hold two references to the item, so
- A matching _put_ does not deallocate the item but keeps it referenced once.
- A find/put pair does not remove the item from cache as well (+=1/-=1)
- [optional] At some future out-of-memory condition or when cache is to big, according to some specified max cache. The most unused items are dereferenced and if are not used (ref==0) are then removed.
- At driver shut-down all cached devices are dereferenced (and if not used which they should) are dealocated.
So in effect if [optional] code above is not yet implemented then a getdeviceinfo is never preformed twice for the same deviceid (Assuming no device recall). This is what I had in osd, and is cardinal for it's performance. Is that what we are aiming at?
Quoted 36 lines
fs/nfs/pnfs.h +6
-1@@ -32,11 +32,13 @@struct pnfs_layout_segment {struct list_head fi_list; -u32 iomode; +struct pnfs_layout_range range;struct kref kref;struct pnfs_layout_hdr *layout; }; +#define NFS4_PNFS_DEVICEID4_SIZE 16 +#ifdef CONFIG_NFS_V4_1 #define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4"@@ -44,6 +46,7 @@ struct pnfs_layout_segment {enum {NFS_LAYOUT_RO_FAILED =0 ,/* get ro layout failed stop trying */ NFS_LAYOUT_RW_FAILED ,/* get rw layout failed stop trying */ +NFS_LAYOUT_STATEID_SET ,/* have a valid layout stateid */ };/* Per-layout driver specific registration structure */ @@ -54,26 +57,102 @@ struct pnfs_layoutdriver_type {struct module *owner;int (*initialize_mountpoint) (struct nfs_server *);int (*uninitialize_mountpoint) (struct nfs_server *); +struct pnfs_layout_segment * (*alloc_lseg) (struct pnfs_layout_hdr *layoutid,struct nfs4_layoutget_res *lgr); +void (*free_lseg) (struct pnfs_layout_segment *lseg); };struct pnfs_layout_hdr {
fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 74
Quoted 10 lines
fs/nfs/pnfs.h +3
-0unsigned long refcount;struct list_head layouts;/* other client layouts */ struct list_head segs;/* layout segments list */ +seqlock_t seqlock;/* Protects the stateid */ +nfs4_stateid stateid;unsigned long state;struct inode *inode; }; +struct pnfs_deviceid {
Please for my own sanity, could we change those names. I'll post a patch once we settle on final code. I want this to be nfs4_deviceid (And perhaps in that other header)
Quoted 26 lines
fs/nfs/pnfs.h +26
-0+char data[NFS4_PNFS_DEVICEID4_SIZE ]; +}; + +struct pnfs_device { +struct pnfs_deviceid dev_id; +unsigned int layout_type; +unsigned int mincount; +struct page **pages; +void *area; +unsigned int pgbase; +unsigned int pglen; +}; + +/* +* Device ID RCU cache. A device ID is unique per client ID and layout type. +*/ +#define NFS4_DEVICE_ID_HASH_BITS 5 +#define NFS4_DEVICE_ID_HASH_SIZE (1 << NFS4_DEVICE_ID_HASH_BITS) +#define NFS4_DEVICE_ID_HASH_MASK (NFS4_DEVICE_ID_HASH_SIZE - 1) + +static inline u32 +nfs4_deviceid_hash (struct pnfs_deviceid *id) +{ +unsigned char *cptr = (unsigned char *)id->data; +unsigned int nbytes =NFS4_PNFS_DEVICEID4_SIZE ; +u32 x =0 ;
fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 110
Quoted 10 lines
fs/nfs/pnfs.h +10
-0+ +while (nbytes--) { + x *=37 ; + x += *cptr++; + } +return x &NFS4_DEVICE_ID_HASH_MASK ; +} + +/* Device ID cache node */ +struct nfs4_deviceid {
And this one rename to pnfs_deviceid_node
fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 123
Quoted 13 lines
fs/nfs/pnfs.h +13
-0+struct hlist_node de_node; +struct pnfs_deviceid de_id; +atomic_t de_ref; +}; + +struct nfs4_deviceid_cache { +spinlock_t dc_lock; +atomic_t dc_ref; +void (*dc_free_callback)(struct nfs4_deviceid *); +struct hlist_head dc_deviceids[NFS4_DEVICE_ID_HASH_SIZE ]; +struct hlist_head dc_to_free; +}; +
All these cache APIs below, the naming convention in this header is to call them pnfs_xxx. They don't have any use in general nfs4. (and are implemented in pnfs.c)
fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 127
Quoted 4 lines
fs/nfs/pnfs.h +4
-0+extern int nfs4_alloc_init_deviceid_cache (struct nfs_client *, +void (*free_callback)(struct nfs4_deviceid *)); +extern void nfs4_put_deviceid_cache (struct nfs_client *); +extern void nfs4_init_deviceid_node (struct nfs4_deviceid *);
Why is above exported? Can't we just call it from _add_
Boaz
Quoted 116 lines
fs/nfs/pnfs.h +18
-0+extern struct nfs4_deviceid *nfs4_find_get_deviceid ( +struct nfs4_deviceid_cache *, +struct pnfs_deviceid *); +extern struct nfs4_deviceid *nfs4_add_deviceid (struct nfs4_deviceid_cache *, +struct nfs4_deviceid *); +extern void nfs4_put_deviceid (struct nfs4_deviceid_cache *c, +struct nfs4_deviceid *devid); +extern int pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *);extern void pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *); +/* nfs4proc.c */ +extern int nfs4_proc_getdeviceinfo (struct nfs_server *server, +struct pnfs_device *dev); +extern int nfs4_proc_layoutget (struct nfs4_layoutget *lgp); + +/* pnfs.c */ struct pnfs_layout_segment *pnfs_update_layout (struct inode *ino,struct nfs_open_context *ctx,enum pnfs_iomode access_type);void set_pnfs_layoutdriver (struct nfs_server *,u32 id);void unset_pnfs_layoutdriver (struct nfs_server *); +int pnfs_layout_process (struct nfs4_layoutget *lgp);void pnfs_destroy_layout (struct nfs_inode *);void pnfs_destroy_all_layouts (struct nfs_client *); +void put_layout_hdr (struct inode *inode); +void pnfs_get_layout_stateid (nfs4_stateid *dst,struct pnfs_layout_hdr *lo, +struct nfs4_state *open_state);static inline int lo_fail_bit (u32 iomode)include/linux/nfs4.h +2
-0@@ -545,6 +545,8 @@ enum {NFSPROC4_CLNT_SEQUENCE ,NFSPROC4_CLNT_GET_LEASE_TIME ,NFSPROC4_CLNT_RECLAIM_COMPLETE , +NFSPROC4_CLNT_LAYOUTGET , + NFSPROC4_CLNT_GETDEVICEINFO, };/* nfs41 types */ include/linux/nfs_fs_sb.h +1
-0@@ -83,6 +83,7 @@ struct nfs_client {u32 cl_exchange_flags;struct nfs4_session *cl_session;/* sharred session */ struct list_head cl_layouts; +struct nfs4_deviceid_cache *cl_devid_cache;/* pNFS deviceid cache */ #endif /* CONFIG_NFS_V4_1 */ #ifdef CONFIG_NFS_FSCACHE include/linux/nfs_xdr.h +49
-0@@ -186,6 +186,55 @@ struct nfs4_get_lease_time_res {struct nfs4_sequence_res lr_seq_res; }; +#define PNFS_LAYOUT_MAXSIZE 4096 + +struct nfs4_layoutdriver_data { +__u32 len; +void *buf; +}; + +struct pnfs_layout_range { +u32 iomode; +u64 offset; +u64 length; +}; + +struct nfs4_layoutget_args { +__u32 type; +struct pnfs_layout_range range; +__u64 minlength; +__u32 maxcount; +struct inode *inode; +struct nfs_open_context *ctx; +struct nfs4_sequence_args seq_args; +}; + +struct nfs4_layoutget_res { +__u32 return_on_close; +struct pnfs_layout_range range; +__u32 type; +nfs4_stateid stateid; +struct nfs4_layoutdriver_data layout; +struct nfs4_sequence_res seq_res; +}; + +struct nfs4_layoutget { +struct nfs4_layoutget_args args; +struct nfs4_layoutget_res res; +struct pnfs_layout_segment **lsegpp; +int status; +}; + +struct nfs4_getdeviceinfo_args { +struct pnfs_device *pdev; +struct nfs4_sequence_args seq_args; +}; + +struct nfs4_getdeviceinfo_res { +struct pnfs_device *pdev; +struct nfs4_sequence_res seq_res; +}; +/* * Arguments to the open call. */
Re: [PATCH 11/12] RFC: pnfs: add LAYOUTGET and GETDEVICEINFO infrastructure
Quoted 11 lines · Boaz Harrosh
Quoted 9 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Add the ability to actually send LAYOUTGET and GETDEVICEINFO. This also adds in the machinery to handle layout state and the deviceid cache. Note that GETDEVICEINFO is not called directly by the generic layer. Instead it is called by the drivers while parsing the LAYOUTGET opaque data in response to an unknown device id embedded therein. Annoyingly, RFC 5661 only encodes device ids within the driver-specific opaque data.
Quoted 3 lines
Fred, In regard to device cache at least, this is a complete new code then what I knew before. I wish you could send a SQUASHME diff first of what has changed.
Many of these changes were sent off-list to you in the initial format of original-code plus squashmes. In particular,
there was a patch labeled "SQUASHME pnfs_submit_1 fix pnfs_submit:
generic pnfs deviceid cache"
Quoted 60 lines
Quoted 1 line
Signed-off-by TBD - melding/reorganization of several patches<snip>
Quoted 56 lines
fs/nfs/pnfs.c +56
-0+/* +* Device ID cache. Currently supports one layout type per struct nfs_client. +* Add layout type to the lookup key to expand to support multiple types. +*/ +int +nfs4_alloc_init_deviceid_cache (struct nfs_client *clp, + void (*free_callback)(struct nfs4_deviceid *)) +{ + struct nfs4_deviceid_cache *c; + + c = kzalloc(sizeof(struct nfs4_deviceid_cache), GFP_KERNEL); + if (!c) + return -ENOMEM; + spin_lock(&clp->cl_lock); + if (clp->cl_devid_cache != NULL) { + atomic_inc(&clp->cl_devid_cache->dc_ref); + dprintk("%s [kref [%d]]\n", __func__, + atomic_read(&clp->cl_devid_cache->dc_ref)); + kfree(c); + } else { + /* kzalloc initializes hlists */ + spin_lock_init(&c->dc_lock); + atomic_set(&c->dc_ref, 1); + c->dc_free_callback = free_callback; + clp->cl_devid_cache = c; + dprintk("%s [new]\n", __func__); + } + spin_unlock(&clp->cl_lock); + return 0; +} +EXPORT_SYMBOL_GPL (nfs4_alloc_init_deviceid_cache); + +void +nfs4_init_deviceid_node (struct nfs4_deviceid *d) +{ + INIT_HLIST_NODE(&d->de_node); + atomic_set(&d->de_ref, 1); +} +EXPORT_SYMBOL_GPL (nfs4_init_deviceid_node); + +/* +* Called from pnfs_layoutdriver_type->free_lseg +* last layout segment reference frees deviceid +*/ +void +nfs4_put_deviceid (struct nfs4_deviceid_cache *c,struct nfs4_deviceid *devid) +{ + struct pnfs_deviceid *id = &devid->de_id; + struct nfs4_deviceid *d; + struct hlist_node *n; + long h = nfs4_deviceid_hash(id); + + dprintk("%s [%d]\n", __func__, atomic_read(&devid->de_ref)); + if (!atomic_dec_and_lock(&devid->de_ref, &c->dc_lock)) + return; +
Quoted 9 lines
Quoted 6 lines
+
hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[h], de_node)
+ if (!memcmp(&d->de_id, id,sizeof(*id))) {
+hlist_del_rcu(&d->de_node);
+spin_unlock(&c->dc_lock);
+synchronize_rcu();
+c->dc_free_callback(devid);What?!? free the devid on every put? surly something went hey-wire
Did you miss the atomic_dec_and_lock above?
Quoted 2 lines
This used to be kref(ed)
The original code required access to the internals of the kref, which was frowned upon.
Quoted 48 lines
Quoted 48 lines
+ return;
+ }
+spin_unlock(&c->dc_lock);fs/nfs/pnfs.c +45
-0+} +EXPORT_SYMBOL_GPL (nfs4_put_deviceid); + +/* Find and reference a deviceid */ +struct nfs4_deviceid * +nfs4_find_get_deviceid (struct nfs4_deviceid_cache *c,struct pnfs_deviceid *id) +{ + struct nfs4_deviceid *d; + struct hlist_node *n; + long hash = nfs4_deviceid_hash(id); + + dprintk("--> %s hash %ld\n", __func__, hash); + rcu_read_lock(); + hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) { + if (!memcmp(&d->de_id, id, sizeof(*id))) { + if (!atomic_inc_not_zero(&d->de_ref)) { + goto fail; + } else { + rcu_read_unlock(); + return d; + } + } + } +fail: + rcu_read_unlock(); + return NULL; +} +EXPORT_SYMBOL_GPL (nfs4_find_get_deviceid); + +/* +* Add a deviceid to the cache. +* GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new +*/ +struct nfs4_deviceid * +nfs4_add_deviceid (struct nfs4_deviceid_cache *c,struct nfs4_deviceid *new) +{ + struct nfs4_deviceid *d; + struct hlist_node *n; + long hash = nfs4_deviceid_hash(&new->de_id); + + dprintk("--> %s hash %ld\n", __func__, hash); + spin_lock(&c->dc_lock); + hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) { + if (!memcmp(&d->de_id, &new->de_id, sizeof(new->de_id))) { + spin_unlock(&c->dc_lock);
Quoted 9 lines
Quoted 3 lines
+
dprintk("%s [discard]\n", __func__);
+c->dc_free_callback(new);
+ return d;I am totally confused and should be smacked on the head strong. As said above one
alloc_lsegadds the first, second comes in, but same deviceid, finds the first and returns it (no refcounting). Then a firstfree_lsegnfs4_put_deviceidis called, so the second lseg holds free memory?
You are right, there is a missing increment of the refcount here.
Part of the problem here is that the flow is obscurd because it is being called form the driver layer.
I think it might be better if nfs4_find_get_deviceid did the GETDEVICEINFO on a cache miss, instead of relying on the driver to do it.
Quoted 12 lines
Quoted 12 lines
+ }
+ }
+hlist_add_head_rcu(&new->de_node, &c->dc_deviceids[hash]);
+spin_unlock(&c->dc_lock);
+dprintk("%s [new]\n", __func__);
+ return new;fs/nfs/pnfs.c +6
-0+} +EXPORT_SYMBOL_GPL (nfs4_add_deviceid); + +void +nfs4_put_deviceid_cache (struct nfs_client *clp) +{
Quoted 9 lines
Quoted 4 lines
+ struct
nfs4_deviceid_cache*local =clp->cl_devid_cache; +
+dprintk("--> %s cl_devid_cache %p\n", __func__, clp->cl_devid_cache);
+ if (atomic_dec_and_lock(&local->dc_ref, &clp->cl_lock)) {Don't you want to make sure there are no hanging devices on this cache. Please note that an add or find does not take the cache reference only
server_initdoes, right?
If there are any hanging devices at this point, it is a pretty serious bug.
Quoted 9 lines
fs/nfs/pnfs.c out_unlock: · line 743
Quoted 6 lines
+
clp->cl_devid_cache= NULL;
+spin_unlock(&clp->cl_lock);
+ kfree(local);
+ }fs/nfs/pnfs.c +2
-0+} +EXPORT_SYMBOL_GPL (nfs4_put_deviceid_cache);Note to self. Test code alot.
Quoted 15 lines
Fred:
I don't understand. A cache property I'm looking for and which I do not See in this code, perhaps because I'm clueless.
(And to be honest I have not understood it before as well, since I never actually ran with that code I posted.)
Is:
- An _add_ to queue should magically hold two references to the item, so
- A matching _put_ does not deallocate the item but keeps it referenced once.
- A find/put pair does not remove the item from cache as well (+=1/-=1)
- [optional] At some future out-of-memory condition or when cache is to big, according to some specified max cache. The most unused items are dereferenced and if are not used (ref==0) are then removed.
- At driver shut-down all cached devices are dereferenced (and if not used which they should) are dealocated.
I agree. The original deviceid code had an issue where nothing was ever deallocated until shutdown. The code as given here ties the deviceid lifetime to existing references. This is not ideal, because a brief loss of reference to a deviceid will cause an unnecessary GETDEVICEINFO. However, for the current submission, it has the advantage that it is simple and correct. Adding the machinery to do as you suggest above is indeed a (fairly high) priority goal, but was intentionally deferred.
Quoted 5 lines
So in effect if [optional] code above is not yet implemented then a getdeviceinfo is never preformed twice for the same deviceid (Assuming no device recall). This is what I had in osd, and is cardinal for it's performance. Is that what we are aiming at?
Yes, that is the goal. Right now, if a layout is still around that references a deviceid, a second GETDEVICEINFO will not be sent. However, if all layouts referencing a specific deviceid have been returned, then yes, a second GETDEVICEINFO will be called.
Quoted 100 lines
Quoted 46 lines
fs/nfs/pnfs.h +3
-1@@ -32,11 +32,13 @@ struct pnfs_layout_segment { struct list_head fi_list; - u32 iomode; + struct pnfs_layout_range range; struct kref kref; struct pnfs_layout_hdr *layout; }; +#define NFS4_PNFS_DEVICEID4_SIZE 16 + #ifdef CONFIG_NFS_V4_1 #define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4"@@ -44,6 +46,7 @@ struct pnfs_layout_segment { enum {
NFS_LAYOUT_RO_FAILED= 0, /* get ro layout failed stop trying */NFS_LAYOUT_RW_FAILED, /* get rw layout failed stop trying */
+NFS_LAYOUT_STATEID_SET, /* have a valid layout stateid */ };/* Per-layout driver specific registration structure */
fs/nfs/pnfs.h +2
-0@@ -54,26 +57,102 @@ struct pnfs_layoutdriver_type { struct module *owner; int (*initialize_mountpoint) (struct nfs_server *); int (*uninitialize_mountpoint) (struct nfs_server *); + struct pnfs_layout_segment * (*alloc_lseg) (struct pnfs_layout_hdr *layoutid, struct nfs4_layoutget_res *lgr); + void (*free_lseg) (struct pnfs_layout_segment *lseg); }; struct pnfs_layout_hdr { unsigned long refcount;struct
list_headlayouts; /* other client layouts */ structlist_headsegs; /* layout segments list */
+seqlock_tseqlock; /* Protects the stateid */
+nfs4_stateidstateid;
unsigned long state;
struct inode *inode;
};Quoted 1 line
+struct
pnfs_deviceid{Please for my own sanity, could we change those names. I'll post a patch once we settle on final code. I want this to be
nfs4_deviceid(And perhaps in that other header)Quoted 26 lines
+ char data[NFS4_PNFS_DEVICEID4_SIZE];
fs/nfs/pnfs.h +25
-0+}; + +struct pnfs_device { + struct pnfs_deviceid dev_id; + unsigned int layout_type; + unsigned int mincount; + struct page **pages; + void *area; + unsigned int pgbase; + unsigned int pglen; +}; + +/* +* Device ID RCU cache. A device ID is unique per client ID and layout type. +*/ +#define NFS4_DEVICE_ID_HASH_BITS 5 +#define NFS4_DEVICE_ID_HASH_SIZE (1 << NFS4_DEVICE_ID_HASH_BITS) +#define NFS4_DEVICE_ID_HASH_MASK (NFS4_DEVICE_ID_HASH_SIZE - 1) + +static inline u32 +nfs4_deviceid_hash (struct pnfs_deviceid *id) +{ + unsigned char *cptr = (unsigned char *)id->data; + unsigned int nbytes = NFS4_PNFS_DEVICEID4_SIZE; + u32 x = 0;fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 110
Quoted 10 lines
fs/nfs/pnfs.h +10
-0+ + while (nbytes--) { + x *= 37; + x += *cptr++; + } + return x & NFS4_DEVICE_ID_HASH_MASK; +} + +/* Device ID cache node */ +struct nfs4_deviceid {And this one rename to
pnfs_deviceid_nodeQuoted 9 lines
+ struct
hlist_nodede_node;
+ structpnfs_deviceidde_id;
+atomic_tde_ref;fs/nfs/pnfs.h +6
-0+}; + +struct nfs4_deviceid_cache { + spinlock_t dc_lock; + atomic_t dc_ref; + void (*dc_free_callback)(struct nfs4_deviceid *);
Quoted 8 lines
fs/nfs/pnfs.h
Quoted 4 lines
+ struct
hlist_headdc_deviceids[NFS4_DEVICE_ID_HASH_SIZE];
+ structhlist_headdc_to_free;fs/nfs/pnfs.h +2
-0+}; +All these cache APIs below, the naming convention in this header is to call them
pnfs_xxx. They don't have any use in general nfs4. (and are implemented inpnfs.c)
I'll defer to Andy regarding naming.
Quoted 8 lines
fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 127
Quoted 5 lines
fs/nfs/pnfs.h +4
-0+extern int nfs4_alloc_init_deviceid_cache (struct nfs_client *, + void (*free_callback)(struct nfs4_deviceid *)); +extern void nfs4_put_deviceid_cache (struct nfs_client *); +extern void nfs4_init_deviceid_node (struct nfs4_deviceid *);Why is above exported? Can't we just call it from _add_
You are right. It is a remnant of older code organization.
Fred
Quoted 124 lines
Boaz
Quoted 106 lines
fs/nfs/pnfs.h +8
-0+extern struct nfs4_deviceid *nfs4_find_get_deviceid ( + struct nfs4_deviceid_cache *, + struct pnfs_deviceid *); +extern struct nfs4_deviceid *nfs4_add_deviceid (struct nfs4_deviceid_cache *, + struct nfs4_deviceid *); +extern void nfs4_put_deviceid (struct nfs4_deviceid_cache *c, + struct nfs4_deviceid *devid); +extern int
pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *); extern voidpnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *);fs/nfs/pnfs.h +6
-0+/* nfs4proc.c */ +extern int nfs4_proc_getdeviceinfo (struct nfs_server *server, + struct pnfs_device *dev); +extern int nfs4_proc_layoutget (struct nfs4_layoutget *lgp); + +/* pnfs.c */ struct
pnfs_layout_segment*
pnfs_update_layout(struct inode *ino, structnfs_open_context*ctx, enumpnfs_iomodeaccess_type);
voidset_pnfs_layoutdriver(struct nfs_server *, u32 id); voidunset_pnfs_layoutdriver(struct nfs_server *);
+intpnfs_layout_process(struct nfs4_layoutget *lgp); voidpnfs_destroy_layout(struct nfs_inode *);
voidpnfs_destroy_all_layouts(struct nfs_client *);fs/nfs/pnfs.h +3
-0+void put_layout_hdr (struct inode *inode); +void pnfs_get_layout_stateid (nfs4_stateid *dst,struct pnfs_layout_hdr *lo, + struct nfs4_state *open_state);static inline int
lo_fail_bit(u32 iomode)include/linux/nfs4.h +2
-0@@ -545,6 +545,8 @@ enum { NFSPROC4_CLNT_SEQUENCE, NFSPROC4_CLNT_GET_LEASE_TIME, NFSPROC4_CLNT_RECLAIM_COMPLETE, + NFSPROC4_CLNT_LAYOUTGET, + NFSPROC4_CLNT_GETDEVICEINFO, }; /* nfs41 types */include/linux/nfs_fs_sb.h +0
-0@@ -83,6 +83,7 @@ struct nfs_client { u32 cl_exchange_flags;struct
nfs4_session*cl_session; /* sharred session */ structlist_headcl_layouts;
+ structnfs4_deviceid_cache*cl_devid_cache; /* pNFS deviceid cache */ #endif /*CONFIG_NFS_V4_1*/#ifdef
CONFIG_NFS_FSCACHEinclude/linux/nfs_xdr.h +42
-0@@ -186,6 +186,55 @@ struct nfs4_get_lease_time_res { struct nfs4_sequence_res lr_seq_res; }; +#define PNFS_LAYOUT_MAXSIZE 4096 + +struct nfs4_layoutdriver_data { + __u32 len; + void *buf; +}; + +struct pnfs_layout_range { + u32 iomode; + u64 offset; + u64 length; +}; + +struct nfs4_layoutget_args { + __u32 type; + struct pnfs_layout_range range; + __u64 minlength; + __u32 maxcount; + struct inode *inode; + struct nfs_open_context *ctx; + struct nfs4_sequence_args seq_args; +}; + +struct nfs4_layoutget_res { + __u32 return_on_close; + struct pnfs_layout_range range; + __u32 type; + nfs4_stateid stateid; + struct nfs4_layoutdriver_data layout; + struct nfs4_sequence_res seq_res; +}; + +struct nfs4_layoutget { + struct nfs4_layoutget_args args; + struct nfs4_layoutget_res res; + struct pnfs_layout_segment **lsegpp; + int status; +}; + +struct nfs4_getdeviceinfo_args { + struct pnfs_device *pdev; + struct nfs4_sequence_args seq_args;include/linux/nfs_xdr.h struct nfs4_get_lease_time_res { · line 240
Quoted 10 lines
include/linux/nfs_xdr.h +7
-0+}; + +struct nfs4_getdeviceinfo_res { + struct pnfs_device *pdev; + struct nfs4_sequence_res seq_res; +}; + /** Arguments to the open call. */ --
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 11/12] RFC: pnfs: add LAYOUTGET and GETDEVICEINFO infrastructure
Quoted 12 lines · Fred Isaman
Quoted 11 lines · Boaz Harrosh
Quoted 9 lines · Fred Isaman
From: The pNFS Team <[email protected]>
Add the ability to actually send LAYOUTGET and GETDEVICEINFO. This also adds in the machinery to handle layout state and the deviceid cache. Note that GETDEVICEINFO is not called directly by the generic layer. Instead it is called by the drivers while parsing the LAYOUTGET opaque data in response to an unknown device id embedded therein. Annoyingly, RFC 5661 only encodes device ids within the driver-specific opaque data.
Quoted 9 lines
Quoted 3 lines
Fred, In regard to device cache at least, this is a complete new code then what I knew before. I wish you could send a SQUASHME diff first of what has changed.
Many of these changes were sent off-list to you in the initial format of original-code plus squashmes. In particular,
there was a patch labeled "SQUASHMEpnfs_submit_1fixpnfs_submit:
generic pnfs deviceid cache"
Sorry for that then, That was confusing and me travelling to LSF got everything mixed up.
But you should have smacked me on the head for hacking the wrong code.
Quoted 71 lines
Quoted 61 lines
Quoted 2 lines
Signed-off-by TBD - melding/reorganization of several patches<snip>
Quoted 56 lines
fs/nfs/pnfs.c +56
-0+/* +* Device ID cache. Currently supports one layout type per struct nfs_client. +* Add layout type to the lookup key to expand to support multiple types. +*/ +int +nfs4_alloc_init_deviceid_cache (struct nfs_client *clp, +void (*free_callback)(struct nfs4_deviceid *)) +{ +struct nfs4_deviceid_cache *c; + + c =kzalloc (sizeof (struct nfs4_deviceid_cache ),GFP_KERNEL ); +if (!c) +return -ENOMEM ; +spin_lock (&clp->cl_lock); +if (clp->cl_devid_cache !=NULL ) { +atomic_inc (&clp->cl_devid_cache->dc_ref); +dprintk ("%s [kref [%d]]\n" , __func__, +atomic_read (&clp->cl_devid_cache->dc_ref)); +kfree (c); + }else { +/* kzalloc initializes hlists */ +spin_lock_init (&c->dc_lock); +atomic_set (&c->dc_ref,1 ); + c->dc_free_callback = free_callback; + clp->cl_devid_cache = c; +dprintk ("%s [new]\n" , __func__); + } +spin_unlock (&clp->cl_lock); +return 0 ; +} +EXPORT_SYMBOL_GPL (nfs4_alloc_init_deviceid_cache); + +void +nfs4_init_deviceid_node (struct nfs4_deviceid *d) +{ +INIT_HLIST_NODE (&d->de_node); +atomic_set (&d->de_ref,1 ); +} +EXPORT_SYMBOL_GPL (nfs4_init_deviceid_node); + +/* +* Called from pnfs_layoutdriver_type->free_lseg +* last layout segment reference frees deviceid +*/ +void +nfs4_put_deviceid (struct nfs4_deviceid_cache *c,struct nfs4_deviceid *devid) +{ +struct pnfs_deviceid *id = &devid->de_id; +struct nfs4_deviceid *d; +struct hlist_node *n; +long h =nfs4_deviceid_hash (id); + +dprintk ("%s [%d]\n" , __func__,atomic_read (&devid->de_ref)); +if (!atomic_dec_and_lock (&devid->de_ref, &c->dc_lock)) +return ; +Quoted 9 lines
fs/nfs/pnfs.c out_unlock: · line 670
Quoted 6 lines
fs/nfs/pnfs.c +6
-0+hlist_for_each_entry_rcu (d, n, &c->dc_deviceids[h], de_node) +if (!memcmp (&d->de_id, id,sizeof (*id))) { +hlist_del_rcu (&d->de_node); +spin_unlock (&c->dc_lock); +synchronize_rcu (); + c->dc_free_callback (devid);What?!? free the devid on every put? surly something went hey-wire
Quoted 2 lines
Did you miss the
atomic_dec_and_lockabove?
Yes thanks that's what I missed, the early return.
Quoted 10 lines
Quoted 3 lines
This used to be kref(ed)
The original code required access to the internals of the kref, which was frowned upon.
Quoted 3 lines
Quoted 3 lines
fs/nfs/pnfs.c +3
-0+return ; + } +spin_unlock (&c->dc_lock);
BUG() not found then, right?
Quoted 37 lines
Quoted 37 lines
Quoted 37 lines
fs/nfs/pnfs.c +37
-0+} +EXPORT_SYMBOL_GPL (nfs4_put_deviceid); + +/* Find and reference a deviceid */ +struct nfs4_deviceid * +nfs4_find_get_deviceid (struct nfs4_deviceid_cache *c,struct pnfs_deviceid *id) +{ +struct nfs4_deviceid *d; +struct hlist_node *n; +long hash =nfs4_deviceid_hash (id); + +dprintk ("--> %s hash %ld\n" , __func__, hash); +rcu_read_lock (); +hlist_for_each_entry_rcu (d, n, &c->dc_deviceids[hash], de_node) { +if (!memcmp (&d->de_id, id,sizeof (*id))) { +if (!atomic_inc_not_zero (&d->de_ref)) { + goto fail; + }else { +rcu_read_unlock (); +return d; + } + } + } +fail: +rcu_read_unlock (); +return NULL ; +} +EXPORT_SYMBOL_GPL (nfs4_find_get_deviceid); + +/* +* Add a deviceid to the cache. +* GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new +*/ +struct nfs4_deviceid * +nfs4_add_deviceid (struct nfs4_deviceid_cache *c,struct nfs4_deviceid *new) +{ +struct nfs4_deviceid *d;
Quoted 8 lines
Quoted 8 lines
fs/nfs/pnfs.c out_unlock: · line 718
Quoted 8 lines
fs/nfs/pnfs.c +8
-0+struct hlist_node *n; +long hash =nfs4_deviceid_hash (&new->de_id); + +dprintk ("--> %s hash %ld\n" , __func__, hash); +spin_lock (&c->dc_lock); +hlist_for_each_entry_rcu (d, n, &c->dc_deviceids[hash], de_node) { +if (!memcmp (&d->de_id, &new->de_id,sizeof (new->de_id))) { +spin_unlock (&c->dc_lock);
did_get(d); Right?
Quoted 10 lines
Quoted 9 lines
fs/nfs/pnfs.c out_unlock: · line 721
Quoted 3 lines
fs/nfs/pnfs.c +3
-0+dprintk ("%s [discard]\n" , __func__); + c->dc_free_callback (new); +return d;I am totally confused and should be smacked on the head strong. As said above one
alloc_lsegadds the first, second comes in, but same deviceid, finds the first and returns it (no refcounting). Then a firstfree_lsegnfs4_put_deviceidis called, so the second lseg holds free memory?
Quoted 8 lines
You are right, there is a missing increment of the refcount here.
Part of the problem here is that the flow is obscurd because it is being called form the driver layer.
I think it might be better ifnfs4_find_get_deviceiddid the GETDEVICEINFO on a cache miss, instead of relying on the driver to do it.
That's fine. It means nothing the failing-find and _add_ are not locked so it changes nothing. The _add_ should be self-sustained regardless.
Quoted 3 lines
Quoted 3 lines
fs/nfs/pnfs.c out_unlock: · line 724
Quoted 3 lines
fs/nfs/pnfs.c +3
-0+ } + } +hlist_add_head_rcu (&new->de_node, &c->dc_deviceids[hash]);
See below I would like to add here;
did_get(new);
Quoted 19 lines
Quoted 9 lines
Quoted 9 lines
fs/nfs/pnfs.c +9
-0+spin_unlock (&c->dc_lock); +dprintk ("%s [new]\n" , __func__); +return new; +} +EXPORT_SYMBOL_GPL (nfs4_add_deviceid); + +void +nfs4_put_deviceid_cache (struct nfs_client *clp) +{Quoted 9 lines
fs/nfs/pnfs.c out_unlock: · line 737
Quoted 4 lines
fs/nfs/pnfs.c +4
-0+struct nfs4_deviceid_cache *local = clp->cl_devid_cache; + +dprintk ("--> %s cl_devid_cache %p\n" , __func__, clp->cl_devid_cache); +if (atomic_dec_and_lock (&local->dc_ref, &clp->cl_lock)) {Don't you want to make sure there are no hanging devices on this cache. Please note that an add or find does not take the cache reference only
server_initdoes, right?
Quoted 2 lines
If there are any hanging devices at this point, it is a pretty serious bug.
Yes therefore a BUG_ON() is do.
But no! I would like for the code that decrements all did refs back in here. See below.
Quoted 26 lines
Quoted 10 lines
fs/nfs/pnfs.c out_unlock: · line 743
Quoted 7 lines
fs/nfs/pnfs.c +6
-0+ clp->cl_devid_cache =NULL ; +spin_unlock (&clp->cl_lock); +kfree (local); + } +} +EXPORT_SYMBOL_GPL (nfs4_put_deviceid_cache);Note to self. Test code alot.
Quoted 15 lines
Fred:
I don't understand. A cache property I'm looking for and which I do not See in this code, perhaps because I'm clueless.
(And to be honest I have not understood it before as well, since I never actually ran with that code I posted.)
Is:
- An _add_ to queue should magically hold two references to the item, so
- A matching _put_ does not deallocate the item but keeps it referenced once.
- A find/put pair does not remove the item from cache as well (+=1/-=1)
- [optional] At some future out-of-memory condition or when cache is to big, according to some specified max cache. The most unused items are dereferenced and if are not used (ref==0) are then removed.
- At driver shut-down all cached devices are dereferenced (and if not used which they should) are dealocated.
Quoted 9 lines
I agree. The original deviceid code had an issue where nothing was ever deallocated until shutdown. The code as given here ties the deviceid lifetime to existing references. This is not ideal, because a brief loss of reference to a deviceid will cause an unnecessary GETDEVICEINFO. However, for the current submission, it has the advantage that it is simple and correct. Adding the machinery to do as you suggest above is indeed a (fairly high) priority goal, but was intentionally deferred.
I would like to please return it. That is deallocate it only at shut down. Otherwise we are for a lot of trouble. We are not yet in a situation of 1000nd of devices where it might matter. Even up to 100rd it is still fine to never de-allocate. (Currently on the planet Panasas is the only one with 1000nd of devices in a single installation)
Look at it this way. Before we mounted all the NFS servers in our domain prior to usage, and/or as part of auto-mount which never got unmounted until shotdown/unmount. So we do the same with pNFS only we have the privilege of doing a single mount and have everything dynamically logged in for us. So we are just as before.
For all pNFS filesystems today the first IOs will GETDEVICEINFO for the 10s of devices deploid and keep them allocated. If we free them then what will happen if we need to GETDEVICEINFO when all memory is dirty and we need it for cleaning that dirty memory. We don't have any forward progress provision for that yet.
The simple thing to do is keep them allocated for ever (until unmount) by adding a single did_get() call in the did_add() function. (And did_put() in cache deallocation). That's more simple then hardening the code.
And If we don't the pNFS performance will suck big time specially for that humongous files-layout get_device_info. Because it will be done for every get_layout. A regular bash script opens a file then closes it. Most of the time we are not parallel as we think.
Quoted 11 lines
Quoted 5 lines
So in effect if [optional] code above is not yet implemented then a getdeviceinfo is never preformed twice for the same deviceid (Assuming no device recall). This is what I had in osd, and is cardinal for it's performance. Is that what we are aiming at?
Yes, that is the goal. Right now, if a layout is still around that references a deviceid, a second GETDEVICEINFO will not be sent. However, if all layouts referencing a specific deviceid have been returned, then yes, a second GETDEVICEINFO will be called.
Then sure when we come to the situation that we need proper support for more then 100 machines in a cluster, then we can add the clean on add stage where if we are higher then x devices we start to replace entries.
What you guys think?
Boaz
Quoted 254 lines
Quoted 100 lines
Quoted 37 lines
fs/nfs/pnfs.h +6
-1@@ -32,11 +32,13 @@struct pnfs_layout_segment {struct list_head fi_list; -u32 iomode; +struct pnfs_layout_range range;struct kref kref;struct pnfs_layout_hdr *layout; }; +#define NFS4_PNFS_DEVICEID4_SIZE 16 +#ifdef CONFIG_NFS_V4_1 #define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4"@@ -44,6 +46,7 @@ struct pnfs_layout_segment {enum {NFS_LAYOUT_RO_FAILED =0 ,/* get ro layout failed stop trying */ NFS_LAYOUT_RW_FAILED ,/* get rw layout failed stop trying */ +NFS_LAYOUT_STATEID_SET ,/* have a valid layout stateid */ };/* Per-layout driver specific registration structure */ @@ -54,26 +57,102 @@ struct pnfs_layoutdriver_type {struct module *owner;int (*initialize_mountpoint) (struct nfs_server *);int (*uninitialize_mountpoint) (struct nfs_server *); +struct pnfs_layout_segment * (*alloc_lseg) (struct pnfs_layout_hdr *layoutid,struct nfs4_layoutget_res *lgr); +void (*free_lseg) (struct pnfs_layout_segment *lseg); };struct pnfs_layout_hdr {fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 74
Quoted 10 lines
fs/nfs/pnfs.h +3
-0unsigned long refcount;struct list_head layouts;/* other client layouts */ struct list_head segs;/* layout segments list */ +seqlock_t seqlock;/* Protects the stateid */ +nfs4_stateid stateid;unsigned long state;struct inode *inode; }; +struct pnfs_deviceid {Please for my own sanity, could we change those names. I'll post a patch once we settle on final code. I want this to be
nfs4_deviceid(And perhaps in that other header)Quoted 26 lines
fs/nfs/pnfs.h +26
-0+char data[NFS4_PNFS_DEVICEID4_SIZE ]; +}; + +struct pnfs_device { +struct pnfs_deviceid dev_id; +unsigned int layout_type; +unsigned int mincount; +struct page **pages; +void *area; +unsigned int pgbase; +unsigned int pglen; +}; + +/* +* Device ID RCU cache. A device ID is unique per client ID and layout type. +*/ +#define NFS4_DEVICE_ID_HASH_BITS 5 +#define NFS4_DEVICE_ID_HASH_SIZE (1 << NFS4_DEVICE_ID_HASH_BITS) +#define NFS4_DEVICE_ID_HASH_MASK (NFS4_DEVICE_ID_HASH_SIZE - 1) + +static inline u32 +nfs4_deviceid_hash (struct pnfs_deviceid *id) +{ +unsigned char *cptr = (unsigned char *)id->data; +unsigned int nbytes =NFS4_PNFS_DEVICEID4_SIZE ; +u32 x =0 ;fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 110
Quoted 10 lines
fs/nfs/pnfs.h +10
-0+ +while (nbytes--) { + x *=37 ; + x += *cptr++; + } +return x &NFS4_DEVICE_ID_HASH_MASK ; +} + +/* Device ID cache node */ +struct nfs4_deviceid {And this one rename to
pnfs_deviceid_nodeQuoted 9 lines
fs/nfs/pnfs.h +9
-0+struct hlist_node de_node; +struct pnfs_deviceid de_id; +atomic_t de_ref; +}; + +struct nfs4_deviceid_cache { +spinlock_t dc_lock; +atomic_t dc_ref; +void (*dc_free_callback)(struct nfs4_deviceid *);Quoted 8 lines
fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 123
Quoted 4 lines
fs/nfs/pnfs.h +4
-0+struct hlist_head dc_deviceids[NFS4_DEVICE_ID_HASH_SIZE ]; +struct hlist_head dc_to_free; +}; +All these cache APIs below, the naming convention in this header is to call them
pnfs_xxx. They don't have any use in general nfs4. (and are implemented inpnfs.c)I'll defer to Andy regarding naming.
Quoted 8 lines
fs/nfs/pnfs.h struct pnfs_layoutdriver_type { · line 127
Quoted 5 lines
fs/nfs/pnfs.h +4
-0+extern int nfs4_alloc_init_deviceid_cache (struct nfs_client *, +void (*free_callback)(struct nfs4_deviceid *)); +extern void nfs4_put_deviceid_cache (struct nfs_client *); +extern void nfs4_init_deviceid_node (struct nfs4_deviceid *);Why is above exported? Can't we just call it from _add_
You are right. It is a remnant of older code organization.
Fred
Quoted 115 lines
Boaz
Quoted 113 lines
fs/nfs/pnfs.h +18
-0+extern struct nfs4_deviceid *nfs4_find_get_deviceid ( +struct nfs4_deviceid_cache *, +struct pnfs_deviceid *); +extern struct nfs4_deviceid *nfs4_add_deviceid (struct nfs4_deviceid_cache *, +struct nfs4_deviceid *); +extern void nfs4_put_deviceid (struct nfs4_deviceid_cache *c, +struct nfs4_deviceid *devid); +extern int pnfs_register_layoutdriver (struct pnfs_layoutdriver_type *);extern void pnfs_unregister_layoutdriver (struct pnfs_layoutdriver_type *); +/* nfs4proc.c */ +extern int nfs4_proc_getdeviceinfo (struct nfs_server *server, +struct pnfs_device *dev); +extern int nfs4_proc_layoutget (struct nfs4_layoutget *lgp); + +/* pnfs.c */ struct pnfs_layout_segment *pnfs_update_layout (struct inode *ino,struct nfs_open_context *ctx,enum pnfs_iomode access_type);void set_pnfs_layoutdriver (struct nfs_server *,u32 id);void unset_pnfs_layoutdriver (struct nfs_server *); +int pnfs_layout_process (struct nfs4_layoutget *lgp);void pnfs_destroy_layout (struct nfs_inode *);void pnfs_destroy_all_layouts (struct nfs_client *); +void put_layout_hdr (struct inode *inode); +void pnfs_get_layout_stateid (nfs4_stateid *dst,struct pnfs_layout_hdr *lo, +struct nfs4_state *open_state);static inline int lo_fail_bit (u32 iomode)include/linux/nfs4.h +2
-0@@ -545,6 +545,8 @@ enum {NFSPROC4_CLNT_SEQUENCE ,NFSPROC4_CLNT_GET_LEASE_TIME ,NFSPROC4_CLNT_RECLAIM_COMPLETE , +NFSPROC4_CLNT_LAYOUTGET , + NFSPROC4_CLNT_GETDEVICEINFO, };/* nfs41 types */ include/linux/nfs_fs_sb.h +1
-0@@ -83,6 +83,7 @@ struct nfs_client {u32 cl_exchange_flags;struct nfs4_session *cl_session;/* sharred session */ struct list_head cl_layouts; +struct nfs4_deviceid_cache *cl_devid_cache;/* pNFS deviceid cache */ #endif /* CONFIG_NFS_V4_1 */ #ifdef CONFIG_NFS_FSCACHE include/linux/nfs_xdr.h +49
-0@@ -186,6 +186,55 @@ struct nfs4_get_lease_time_res {struct nfs4_sequence_res lr_seq_res; }; +#define PNFS_LAYOUT_MAXSIZE 4096 + +struct nfs4_layoutdriver_data { +__u32 len; +void *buf; +}; + +struct pnfs_layout_range { +u32 iomode; +u64 offset; +u64 length; +}; + +struct nfs4_layoutget_args { +__u32 type; +struct pnfs_layout_range range; +__u64 minlength; +__u32 maxcount; +struct inode *inode; +struct nfs_open_context *ctx; +struct nfs4_sequence_args seq_args; +}; + +struct nfs4_layoutget_res { +__u32 return_on_close; +struct pnfs_layout_range range; +__u32 type; +nfs4_stateid stateid; +struct nfs4_layoutdriver_data layout; +struct nfs4_sequence_res seq_res; +}; + +struct nfs4_layoutget { +struct nfs4_layoutget_args args; +struct nfs4_layoutget_res res; +struct pnfs_layout_segment **lsegpp; +int status; +}; + +struct nfs4_getdeviceinfo_args { +struct pnfs_device *pdev; +struct nfs4_sequence_args seq_args; +}; + +struct nfs4_getdeviceinfo_res { +struct pnfs_device *pdev; +struct nfs4_sequence_res seq_res; +}; +Quoted 9 lines
include/linux/nfs_xdr.h struct nfs4_get_lease_time_res { · line 240
Quoted 3 lines
include/linux/nfs_xdr.h +0
-0/* * Arguments to the open call. */ --
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 11/12] RFC: pnfs: add LAYOUTGET and GETDEVICEINFO infrastructure
Quoted 26 lines · Boaz Harrosh
Quoted 12 lines · Fred Isaman
Quoted 1 line · Boaz Harrosh
On 09/18/2010 05:17 AM, Fred Isaman wrote:
I agree. The original deviceid code had an issue where nothing was ever deallocated until shutdown. The code as given here ties the deviceid lifetime to existing references. This is not ideal, because a brief loss of reference to a deviceid will cause an unnecessary GETDEVICEINFO. However, for the current submission, it has the advantage that it is simple and correct. Adding the machinery to do as you suggest above is indeed a (fairly high) priority goal, but was intentionally deferred.
I would like to please return it. That is deallocate it only at shut down. Otherwise we are for a lot of trouble. We are not yet in a situation of 1000nd of devices where it might matter. Even up to 100rd it is still fine to never de-allocate. (Currently on the planet Panasas is the only one with 1000nd of devices in a single installation)
Look at it this way. Before we mounted all the NFS servers in our domain prior to usage, and/or as part of auto-mount which never got unmounted until shotdown/unmount. So we do the same with pNFS only we have the privilege of doing a single mount and have everything dynamically logged in for us. So we are just as before.
Quoted 5 lines
For all pNFS filesystems today the first IOs will GETDEVICEINFO for the 10s of devices deploid and keep them allocated. If we free them then what will happen if we need to GETDEVICEINFO when all memory is dirty and we need it for cleaning that dirty memory. We don't have any forward progress provision for that yet.
Devices are no different than layouts for this matter.
To flush your cache under low memory conditions using pnfs you'll need both a layout and the corresponding devices. So why do you want to keep the devices forever but not the layouts?
Quoted 11 lines
The simple thing to do is keep them allocated for ever (until unmount) by adding a single
did_get()call in thedid_add()function. (Anddid_put()in cache deallocation). That's more simple then hardening the code.And If we don't the pNFS performance will suck big time specially for that humongous files-layout
get_device_info. Because it will be done for everyget_layout. A regular bash script opens a file then closes it. Most of the time we are not parallel as we think.
I'd like to see the actual numbers for a given benchmark. Keep in mind that typically for the files layout the server won't ask for return_on_close so the layout will actually be kept around (and the associated devices) while the inode is resident, right?
Quoted 12 lines
Quoted 11 lines
Quoted 5 lines
So in effect if [optional] code above is not yet implemented then a getdeviceinfo is never preformed twice for the same deviceid (Assuming no device recall). This is what I had in osd, and is cardinal for it's performance. Is that what we are aiming at?
Yes, that is the goal. Right now, if a layout is still around that references a deviceid, a second GETDEVICEINFO will not be sent. However, if all layouts referencing a specific deviceid have been returned, then yes, a second GETDEVICEINFO will be called.
Quoted 6 lines
Then sure when we come to the situation that we need proper support for more then 100 machines in a cluster, then we can add the clean on add stage where if we are higher then x devices we start to replace entries.
What you guys think?
It's hard to tell without the actual numbers.
Keeping a reference count on the device from the layout makes sure you'll have the devices to use the layout and that you don't issue multiple GETDEVICEINFOs in case a device is shared between more than one layout.
Ideally, a laundromat service could clean up unused client state once in a while so to keep it around for a short while so it can be reused if possible.
Benny
Quoted 3 lines
Boaz
Re: [PATCH 11/12] RFC: pnfs: add LAYOUTGET and GETDEVICEINFO infrastructure
Quoted 50 lines · Benny Halevy
Quoted 26 lines · Boaz Harrosh
Quoted 12 lines · Fred Isaman
Quoted 1 line · Boaz Harrosh
On 09/18/2010 05:17 AM, Fred Isaman wrote:
I agree. The original deviceid code had an issue where nothing was ever deallocated until shutdown. The code as given here ties the deviceid lifetime to existing references. This is not ideal, because a brief loss of reference to a deviceid will cause an unnecessary GETDEVICEINFO. However, for the current submission, it has the advantage that it is simple and correct. Adding the machinery to do as you suggest above is indeed a (fairly high) priority goal, but was intentionally deferred.
I would like to please return it. That is deallocate it only at shut down. Otherwise we are for a lot of trouble. We are not yet in a situation of 1000nd of devices where it might matter. Even up to 100rd it is still fine to never de-allocate. (Currently on the planet Panasas is the only one with 1000nd of devices in a single installation)
Look at it this way. Before we mounted all the NFS servers in our domain prior to usage, and/or as part of auto-mount which never got unmounted until shotdown/unmount. So we do the same with pNFS only we have the privilege of doing a single mount and have everything dynamically logged in for us. So we are just as before.
Quoted 5 lines
For all pNFS filesystems today the first IOs will GETDEVICEINFO for the 10s of devices deploid and keep them allocated. If we free them then what will happen if we need to GETDEVICEINFO when all memory is dirty and we need it for cleaning that dirty memory. We don't have any forward progress provision for that yet.
Devices are no different than layouts for this matter.
To flush your cache under low memory conditions using pnfs you'll need both a layout and the corresponding devices. So why do you want to keep the devices forever but not the layouts?Quoted 11 lines
The simple thing to do is keep them allocated for ever (until unmount) by adding a single
did_get()call in thedid_add()function. (Anddid_put()in cache deallocation). That's more simple then hardening the code.And If we don't the pNFS performance will suck big time specially for that humongous files-layout
get_device_info. Because it will be done for everyget_layout. A regular bash script opens a file then closes it. Most of the time we are not parallel as we think.
Quoted 5 lines
I'd like to see the actual numbers for a given benchmark. Keep in mind that typically for the files layout the server won't ask for
return_on_closeso the layout will actually be kept around (and the associated devices) while the inode is resident, right?
True. The current code (once bugfixed) is simple, spec-compliant, and works reasonably well for filelayout. Why do improvements like maintaining cache lifetime past reference need to be in the initial submission? I agree they are important and will need to be done, but why now?
Fred
Quoted 40 lines
Quoted 12 lines
Quoted 11 lines
Quoted 5 lines
So in effect if [optional] code above is not yet implemented then a getdeviceinfo is never preformed twice for the same deviceid (Assuming no device recall). This is what I had in osd, and is cardinal for it's performance. Is that what we are aiming at?
Yes, that is the goal. Right now, if a layout is still around that references a deviceid, a second GETDEVICEINFO will not be sent. However, if all layouts referencing a specific deviceid have been returned, then yes, a second GETDEVICEINFO will be called.
Quoted 6 lines
Then sure when we come to the situation that we need proper support for more then 100 machines in a cluster, then we can add the clean on add stage where if we are higher then x devices we start to replace entries.
What you guys think?
It's hard to tell without the actual numbers.
Keeping a reference count on the device from the layout makes sure you'll have the devices to use the layout and that you don't issue multiple GETDEVICEINFOs in case a device is shared between more than one layout.
Ideally, a laundromat service could clean up unused client state once in a while so to keep it around for a short while so it can be reused if possible.
Benny
Quoted 3 lines
Boaz
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
PATCH 12/12 RFC: pnfs: filelayout: add driver's LAYOUTGET and GETDEVICEINFO infrastructure
In reply to Fred IsamanFrom: The pNFS Team <[email protected]>
Implement the driver's io_ops->alloc_lseg and free_lseg functions, which integrate into the deviceid cache and calls out to nfs4_proc_getdeviceinfo when necessary.
--- fs/nfs/Makefile | 2 +- fs/nfs/client.c | 1 + fs/nfs/nfs4filelayout.c | 204 ++++++++++++++++++++- fs/nfs/nfs4filelayout.h | 94 +++++++++ fs/nfs/nfs4filelayoutdev.c | 451 ++++++++++++++++++++++++++++++++++++++++++++ 5 files changed, 750 insertions(+), 2 deletions(-) create mode 100644 fs/nfs/nfs4filelayout.h create mode 100644 fs/nfs/nfs4filelayoutdev.c
fs/nfs/Makefile +1 -1
@@ -20,4 +20,4 @@ nfs-$(CONFIG_SYSCTL) += sysctl.o nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o obj-$(CONFIG_NFS_V4_1) += nfs_layout_nfsv41_files.o -nfs_layout_nfsv41_files-y := nfs4filelayout.o +nfs_layout_nfsv41_files-y := nfs4filelayout.o nfs4filelayoutdev.o
fs/nfs/client.c +1 -0
@@ -255,6 +255,7 @@ void nfs_put_client(struct nfs_client *clp)nfs_free_client (clp); } } +EXPORT_SYMBOL_GPL (nfs_put_client);#if defined(CONFIG_IPV6 ) || defined(CONFIG_IPV6_MODULE )/*
fs/nfs/nfs4filelayout.c +203 -1
233 lines not on this page · raw message
fs/nfs/nfs4filelayout.h new file +94 -0
94 lines not on this page · raw message
fs/nfs/nfs4filelayoutdev.c new file +451 -0
451 lines not on this page · raw message
Signature · 1 line
1.7.2.1