Skip to contentSkip to conversation

staging/lustre: lloop depends on BLOCK

PATCH In review Xiong Zhou 5 participants 11 messages Jul 30, 2013 – Aug 8, 2013 Staging

lore
Share & tools
lore

Xiong Zhou

author
#0 lore raw

PATCH v2 1/1 staging/lustre: lloop depends on BLOCK

From: Xiong Zhou <[email protected]>

In the lustre client driver, lloop depends on BLOCK. Add an option for this dependence. Last version of this patch makes LUSTRE_FS depends on BLOCK.
Remove unnecessary jdb head files which depends on BLOCK.

---
 drivers/staging/lustre/lustre/Kconfig              |    7 ++++++-
 drivers/staging/lustre/lustre/fld/fld_cache.c      |    1 -
 drivers/staging/lustre/lustre/fld/fld_request.c    |    1 -
 .../lustre/lustre/include/linux/lustre_compat25.h  |    2 ++
 drivers/staging/lustre/lustre/llite/Makefile       |    2 +-
 drivers/staging/lustre/lustre/lvfs/fsfilt.c        |    1 -
 6 files changed, 9 insertions(+), 5 deletions(-)
drivers/staging/lustre/lustre/Kconfig +6 -1
index 0b45de0..4e898e4 100644
@@ -1,6 +1,6 @@
 config LUSTRE_FS
 	tristate "Lustre file system client support"
-	depends on STAGING && INET && BLOCK && m
+	depends on INET && m
 	select LNET
 	select CRYPTO
 	select CRYPTO_CRC32
@@ -53,3 +53,8 @@ config LUSTRE_TRANSLATE_ERRNOS
 	bool
 	depends on LUSTRE_FS && !X86
 	default true
+
+config LUSTRE_LLITE_LLOOP
+	bool "Lustre virtual block device"
+	depends on LUSTRE_FS && BLOCK
+	default m
drivers/staging/lustre/lustre/fld/fld_cache.c +0 -1
index 347f2ae..8410107 100644
@@ -45,7 +45,6 @@
 
 # include <linux/libcfs/libcfs.h>
 # include <linux/module.h>
-# include <linux/jbd.h>
 # include <asm/div64.h>
 
 #include <obd.h>
drivers/staging/lustre/lustre/fld/fld_request.c +0 -1
index c99b945..049322a 100644
@@ -44,7 +44,6 @@
 
 # include <linux/libcfs/libcfs.h>
 # include <linux/module.h>
-# include <linux/jbd.h>
 # include <asm/div64.h>
 
 #include <obd.h>
drivers/staging/lustre/lustre/include/linux/lustre_compat25.h +2 -0
index 426533b..c0156e3 100644
@@ -111,12 +111,14 @@ static inline void ll_set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
 #define TREE_READ_LOCK_IRQ(mapping)	spin_lock_irq(&(mapping)->tree_lock)
 #define TREE_READ_UNLOCK_IRQ(mapping)	spin_unlock_irq(&(mapping)->tree_lock)
 
+#ifdef CONFIG_BLOCK
 static inline
 int ll_unregister_blkdev(unsigned int dev, const char *name)
 {
 	unregister_blkdev(dev, name);
 	return 0;
 }
+#endif
 
 #define ll_invalidate_bdev(a,b)	 invalidate_bdev((a))
 
drivers/staging/lustre/lustre/llite/Makefile +1 -1
index dff0c04..f493e07 100644
@@ -1,5 +1,5 @@
 obj-$(CONFIG_LUSTRE_FS) += lustre.o
-obj-$(CONFIG_LUSTRE_FS) += llite_lloop.o
+obj-$(CONFIG_LUSTRE_LLITE_LLOOP) += llite_lloop.o
 lustre-y := dcache.o dir.o file.o llite_close.o llite_lib.o llite_nfs.o \
 	    rw.o lproc_llite.o namei.o symlink.o llite_mmap.o \
 	    xattr.o remote_perm.o llite_rmtacl.o llite_capa.o \
drivers/staging/lustre/lustre/lvfs/fsfilt.c +0 -1
index 064445c..ce71f80 100644
@@ -35,7 +35,6 @@
 #define DEBUG_SUBSYSTEM S_FILTER
 
 #include <linux/fs.h>
-#include <linux/jbd.h>
 #include <linux/module.h>
 #include <linux/kmod.h>
 #include <linux/slab.h>
Reviewed-by on v2 1/1
#1 lore raw
Quoted 7 lines · Xiong Zhou

From: Xiong Zhou <[email protected]>

In the lustre client driver, lloop depends on BLOCK. Add an option for this dependence. Last version of this patch makes LUSTRE_FS depends on BLOCK.
Remove unnecessary jdb head files which depends on BLOCK.

Thanks for the patch. One minor comment below. Other than that, you can add

Quoted 60 lines
---
 drivers/staging/lustre/lustre/Kconfig              |    7 ++++++-
 drivers/staging/lustre/lustre/fld/fld_cache.c      |    1 -
 drivers/staging/lustre/lustre/fld/fld_request.c    |    1 -
 .../lustre/lustre/include/linux/lustre_compat25.h  |    2 ++
 drivers/staging/lustre/lustre/llite/Makefile       |    2 +-
 drivers/staging/lustre/lustre/lvfs/fsfilt.c        |    1 -
 6 files changed, 9 insertions(+), 5 deletions(-)
drivers/staging/lustre/lustre/Kconfig +6 -1
index 0b45de0..4e898e4 100644
@@ -1,6 +1,6 @@
 config LUSTRE_FS
        tristate "Lustre file system client support"
-       depends on STAGING && INET && BLOCK && m
+       depends on INET && m
        select LNET
        select CRYPTO
        select CRYPTO_CRC32
@@ -53,3 +53,8 @@ config LUSTRE_TRANSLATE_ERRNOS
        bool
        depends on LUSTRE_FS && !X86
        default true
+
+config LUSTRE_LLITE_LLOOP
+       bool "Lustre virtual block device"
+       depends on LUSTRE_FS && BLOCK
+       default m
drivers/staging/lustre/lustre/fld/fld_cache.c +0 -1
index 347f2ae..8410107 100644
@@ -45,7 +45,6 @@
 
 # include <linux/libcfs/libcfs.h>
 # include <linux/module.h>
-# include <linux/jbd.h>
 # include <asm/div64.h>
 
 #include <obd.h>
drivers/staging/lustre/lustre/fld/fld_request.c +0 -1
index c99b945..049322a 100644
@@ -44,7 +44,6 @@
 
 # include <linux/libcfs/libcfs.h>
 # include <linux/module.h>
-# include <linux/jbd.h>
 # include <asm/div64.h>
 
 #include <obd.h>
drivers/staging/lustre/lustre/include/linux/lustre_compat25.h +0 -0
index 426533b..c0156e3 100644

drivers/staging/lustre/lustre/include/linux/lustre_compat25.h static inline void ll_set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, · line 116

Quoted 6 lines
drivers/staging/lustre/lustre/include/linux/lustre_compat25.h +1 -0
@@ -111,12 +111,14 @@ static inline void ll_set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
 #define TREE_READ_LOCK_IRQ(mapping)    spin_lock_irq(&(mapping)->tree_lock)
 #define TREE_READ_UNLOCK_IRQ(mapping)  spin_unlock_irq(&(mapping)->tree_lock)
 
+#ifdef CONFIG_BLOCK
 static inline
 int ll_unregister_blkdev(unsigned int dev, const char *name)

It is better to remove the wrapper and just call unregister_blkdev in lloop.c. The wrapper exists to support old kernels in Lustre master but doesn't make sense in upstream kernel.

Thanks,
Tao

Quoted 31 lines
drivers/staging/lustre/lustre/include/linux/lustre_compat25.h +1 -0
 {
        unregister_blkdev(dev, name);
        return 0;
 }
+#endif
 
 #define ll_invalidate_bdev(a,b)         invalidate_bdev((a))
 
drivers/staging/lustre/lustre/llite/Makefile +1 -1
index dff0c04..f493e07 100644
@@ -1,5 +1,5 @@
 obj-$(CONFIG_LUSTRE_FS) += lustre.o
-obj-$(CONFIG_LUSTRE_FS) += llite_lloop.o
+obj-$(CONFIG_LUSTRE_LLITE_LLOOP) += llite_lloop.o
 lustre-y := dcache.o dir.o file.o llite_close.o llite_lib.o llite_nfs.o \
            rw.o lproc_llite.o namei.o symlink.o llite_mmap.o \
            xattr.o remote_perm.o llite_rmtacl.o llite_capa.o \
drivers/staging/lustre/lustre/lvfs/fsfilt.c +0 -1
index 064445c..ce71f80 100644
@@ -35,7 +35,6 @@
 #define DEBUG_SUBSYSTEM S_FILTER
 
 #include <linux/fs.h>
-#include <linux/jbd.h>
 #include <linux/module.h>
 #include <linux/kmod.h>
 #include <linux/slab.h>
author Reviewed-by Peng Tao
#2 lore raw
Quoted 26 lines · Peng Tao
Quoted 7 lines · Xiong Zhou

From: Xiong Zhou <[email protected]>

In the lustre client driver, lloop depends on BLOCK. Add an option for this dependence. Last version of this patch makes LUSTRE_FS depends on BLOCK.
Remove unnecessary jdb head files which depends on BLOCK.

Thanks for the patch. One minor comment below. Other than that, you can add

Quoted 10 lines
---
 drivers/staging/lustre/lustre/Kconfig              |    7 ++++++-
 drivers/staging/lustre/lustre/fld/fld_cache.c      |    1 -
 drivers/staging/lustre/lustre/fld/fld_request.c    |    1 -
 .../lustre/lustre/include/linux/lustre_compat25.h  |    2 ++
 drivers/staging/lustre/lustre/llite/Makefile       |    2 +-
 drivers/staging/lustre/lustre/lvfs/fsfilt.c        |    1 -
 6 files changed, 9 insertions(+), 5 deletions(-)
Quoted 5 lines

#define TREE_READ_UNLOCK_IRQ(mapping) spin_unlock_irq(&(mapping)->tree_lock)

+#ifdef CONFIG_BLOCK
static inline
int ll_unregister_blkdev(unsigned int dev, const char *name)

Quoted 7 lines

It is better to remove the wrapper and just call unregister_blkdev in lloop.c. The wrapper exists to support old kernels in Lustre master but doesn't make sense in upstream kernel.

Thanks,
Tao

Good comment, Thanks for the review.

From: Xiong Zhou <[email protected]>

First version of this patch makes LUSTRE_FS depends on BLOCK. Second version makes only lloop depends on BLOCK with a config option for this dependence, and remove unnecessary jdb header files which depends on BLOCK.
This version removes the wrapper ll_unregister_blkdev which depends on BLOCK in header and just call unregister_blkdev in lloop.c based on Peng Tao's comment.

---
 drivers/staging/lustre/lustre/Kconfig              |    7 ++++++-
 drivers/staging/lustre/lustre/fld/fld_cache.c      |    1 -
 drivers/staging/lustre/lustre/fld/fld_request.c    |    1 -
 .../lustre/lustre/include/linux/lustre_compat25.h  |    7 -------
 drivers/staging/lustre/lustre/llite/Makefile       |    2 +-
 drivers/staging/lustre/lustre/llite/lloop.c        |    6 ++----
 drivers/staging/lustre/lustre/lvfs/fsfilt.c        |    1 -
 7 files changed, 9 insertions(+), 16 deletions(-)
drivers/staging/lustre/lustre/Kconfig +6 -1
index 0b45de0..4e898e4 100644
@@ -1,6 +1,6 @@
 config LUSTRE_FS
 	tristate "Lustre file system client support"
-	depends on STAGING && INET && BLOCK && m
+	depends on INET && m
 	select LNET
 	select CRYPTO
 	select CRYPTO_CRC32
@@ -53,3 +53,8 @@ config LUSTRE_TRANSLATE_ERRNOS
 	bool
 	depends on LUSTRE_FS && !X86
 	default true
+
+config LUSTRE_LLITE_LLOOP
+	bool "Lustre virtual block device"
+	depends on LUSTRE_FS && BLOCK
+	default m
drivers/staging/lustre/lustre/fld/fld_cache.c +0 -1
index 347f2ae..8410107 100644
@@ -45,7 +45,6 @@
 
 # include <linux/libcfs/libcfs.h>
 # include <linux/module.h>
-# include <linux/jbd.h>
 # include <asm/div64.h>
 
 #include <obd.h>
drivers/staging/lustre/lustre/fld/fld_request.c +0 -1
index c99b945..049322a 100644
@@ -44,7 +44,6 @@
 
 # include <linux/libcfs/libcfs.h>
 # include <linux/module.h>
-# include <linux/jbd.h>
 # include <asm/div64.h>
 
 #include <obd.h>
drivers/staging/lustre/lustre/include/linux/lustre_compat25.h +0 -7
index 426533b..018e604 100644
@@ -111,13 +111,6 @@ static inline void ll_set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
 #define TREE_READ_LOCK_IRQ(mapping)	spin_lock_irq(&(mapping)->tree_lock)
 #define TREE_READ_UNLOCK_IRQ(mapping)	spin_unlock_irq(&(mapping)->tree_lock)
 
-static inline
-int ll_unregister_blkdev(unsigned int dev, const char *name)
-{
-	unregister_blkdev(dev, name);
-	return 0;
-}
-
 #define ll_invalidate_bdev(a,b)	 invalidate_bdev((a))
 
 #ifndef FS_HAS_FIEMAP
drivers/staging/lustre/lustre/llite/Makefile +1 -1
index dff0c04..f493e07 100644
@@ -1,5 +1,5 @@
 obj-$(CONFIG_LUSTRE_FS) += lustre.o
-obj-$(CONFIG_LUSTRE_FS) += llite_lloop.o
+obj-$(CONFIG_LUSTRE_LLITE_LLOOP) += llite_lloop.o
 lustre-y := dcache.o dir.o file.o llite_close.o llite_lib.o llite_nfs.o \
 	    rw.o lproc_llite.o namei.o symlink.o llite_mmap.o \
 	    xattr.o remote_perm.o llite_rmtacl.o llite_capa.o \
drivers/staging/lustre/lustre/llite/lloop.c +2 -4
index e063efc..ae3dc9d 100644
@@ -848,10 +848,8 @@ static void lloop_exit(void)
 		blk_cleanup_queue(loop_dev[i].lo_queue);
 		put_disk(disks[i]);
 	}
-	if (ll_unregister_blkdev(lloop_major, "lloop"))
-		CWARN("lloop: cannot unregister blkdev\n");
-	else
-		CDEBUG(D_CONFIG, "unregistered lloop major %d\n", lloop_major);
+
+	unregister_blkdev(lloop_major, "lloop");
 
 	OBD_FREE(disks, max_loop * sizeof(*disks));
 	OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
drivers/staging/lustre/lustre/lvfs/fsfilt.c +0 -1
index 064445c..ce71f80 100644
@@ -35,7 +35,6 @@
 #define DEBUG_SUBSYSTEM S_FILTER
 
 #include <linux/fs.h>
-#include <linux/jbd.h>
#3 lore raw
Quoted 10 lines · Xiong Zhou

From: Xiong Zhou <[email protected]>

First version of this patch makes LUSTRE_FS depends on BLOCK. Second version makes only lloop depends on BLOCK with a config option for this dependence, and remove unnecessary jdb header files which depends on BLOCK.

This version removes the wrapper ll_unregister_blkdev which depends on BLOCK in header and just call unregister_blkdev in lloop.c based on Peng Tao's comment.

This isn't all needed in the patch changelog info, just say what it does.

Also, it doesn't apply for some reason, care to refresh this against my latest tree and resend?

thanks,

greg k-h

author Reviewed-by Peng Tao
#4 lore raw
Quoted 15 lines · Greg Kroah-Hartman
Quoted 10 lines · Xiong Zhou

From: Xiong Zhou <[email protected]>

First version of this patch makes LUSTRE_FS depends on BLOCK. Second version makes only lloop depends on BLOCK with a config option for this dependence, and remove unnecessary jdb header files which depends on BLOCK.

This version removes the wrapper ll_unregister_blkdev which depends on BLOCK in header and just call unregister_blkdev in lloop.c based on Peng Tao's comment.

This isn't all needed in the patch changelog info, just say what it does.

Quoted 7 lines

Also, it doesn't apply for some reason, care to refresh this against my latest tree and resend?

thanks,

greg k-h

OK, this patch is based on the staging-next branch of your staging tree.

From: Xiong Zhou <[email protected]>

Add a config option for llite/lloop in lustre driver, making it depends on BLOCK to fix this better:
drivers/staging/lustre/lustre/fid/../include/linux/lustre_compat25.h:117:2:
error: implicit declaration of function ‘unregister_blkdev’

Also, remove the wrapper ll_unregister_blkdev which depends on BLOCK in the header and just call unregister_blkdev in lloop.c based on Peng Tao's comment. Drop the redundant dependency on STAGING for LUSTRE_FS, remove some unnecessary jdb header files which depends on BLOCK btw.

---
 drivers/staging/lustre/lustre/Kconfig              |    7 ++++++-
 drivers/staging/lustre/lustre/fld/fld_cache.c      |    1 -
 drivers/staging/lustre/lustre/fld/fld_request.c    |    1 -
 .../lustre/lustre/include/linux/lustre_compat25.h  |    7 -------
 drivers/staging/lustre/lustre/llite/Makefile       |    2 +-
 drivers/staging/lustre/lustre/llite/lloop.c        |    6 ++----
 drivers/staging/lustre/lustre/lvfs/fsfilt.c        |    1 -
 7 files changed, 9 insertions(+), 16 deletions(-)
drivers/staging/lustre/lustre/Kconfig +6 -1
index 0b45de0..4e898e4 100644
@@ -1,6 +1,6 @@
 config LUSTRE_FS
 	tristate "Lustre file system client support"
-	depends on STAGING && INET && BLOCK && m
+	depends on INET && m
 	select LNET
 	select CRYPTO
 	select CRYPTO_CRC32
@@ -53,3 +53,8 @@ config LUSTRE_TRANSLATE_ERRNOS
 	bool
 	depends on LUSTRE_FS && !X86
 	default true
+
+config LUSTRE_LLITE_LLOOP
+	bool "Lustre virtual block device"
+	depends on LUSTRE_FS && BLOCK
+	default m
drivers/staging/lustre/lustre/fld/fld_cache.c +0 -1
index 347f2ae..8410107 100644
@@ -45,7 +45,6 @@
 
 # include <linux/libcfs/libcfs.h>
 # include <linux/module.h>
-# include <linux/jbd.h>
 # include <asm/div64.h>
 
 #include <obd.h>
drivers/staging/lustre/lustre/fld/fld_request.c +0 -1
index c99b945..049322a 100644
@@ -44,7 +44,6 @@
 
 # include <linux/libcfs/libcfs.h>
 # include <linux/module.h>
-# include <linux/jbd.h>
 # include <asm/div64.h>
 
 #include <obd.h>
drivers/staging/lustre/lustre/include/linux/lustre_compat25.h +0 -7
index 426533b..018e604 100644
@@ -111,13 +111,6 @@ static inline void ll_set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
 #define TREE_READ_LOCK_IRQ(mapping)	spin_lock_irq(&(mapping)->tree_lock)
 #define TREE_READ_UNLOCK_IRQ(mapping)	spin_unlock_irq(&(mapping)->tree_lock)
 
-static inline
-int ll_unregister_blkdev(unsigned int dev, const char *name)
-{
-	unregister_blkdev(dev, name);
-	return 0;
-}
-
 #define ll_invalidate_bdev(a,b)	 invalidate_bdev((a))
 
 #ifndef FS_HAS_FIEMAP
drivers/staging/lustre/lustre/llite/Makefile +1 -1
index dff0c04..f493e07 100644
@@ -1,5 +1,5 @@
 obj-$(CONFIG_LUSTRE_FS) += lustre.o
-obj-$(CONFIG_LUSTRE_FS) += llite_lloop.o
+obj-$(CONFIG_LUSTRE_LLITE_LLOOP) += llite_lloop.o
 lustre-y := dcache.o dir.o file.o llite_close.o llite_lib.o llite_nfs.o \
 	    rw.o lproc_llite.o namei.o symlink.o llite_mmap.o \
 	    xattr.o remote_perm.o llite_rmtacl.o llite_capa.o \
drivers/staging/lustre/lustre/llite/lloop.c +2 -4
index e063efc..ae3dc9d 100644
@@ -848,10 +848,8 @@ static void lloop_exit(void)
 		blk_cleanup_queue(loop_dev[i].lo_queue);
 		put_disk(disks[i]);
 	}
-	if (ll_unregister_blkdev(lloop_major, "lloop"))
-		CWARN("lloop: cannot unregister blkdev\n");
-	else
-		CDEBUG(D_CONFIG, "unregistered lloop major %d\n", lloop_major);
+
+	unregister_blkdev(lloop_major, "lloop");
 
 	OBD_FREE(disks, max_loop * sizeof(*disks));
 	OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
drivers/staging/lustre/lustre/lvfs/fsfilt.c +0 -1
index 064445c..ce71f80 100644
@@ -35,7 +35,6 @@
 #define DEBUG_SUBSYSTEM S_FILTER
 
 #include <linux/fs.h>
-#include <linux/jbd.h>
 #include <linux/module.h>
 #include <linux/kmod.h>
 #include <linux/slab.h>
#5 lore raw
Quoted 37 lines · Xiong Zhou
Quoted 15 lines · Greg Kroah-Hartman
Quoted 10 lines · Xiong Zhou

From: Xiong Zhou <[email protected]>

First version of this patch makes LUSTRE_FS depends on BLOCK. Second version makes only lloop depends on BLOCK with a config option for this dependence, and remove unnecessary jdb header files which depends on BLOCK.

This version removes the wrapper ll_unregister_blkdev which depends on BLOCK in header and just call unregister_blkdev in lloop.c based on Peng Tao's comment.

This isn't all needed in the patch changelog info, just say what it does.

Quoted 7 lines

Also, it doesn't apply for some reason, care to refresh this against my latest tree and resend?

thanks,

greg k-h

OK, this patch is based on the staging-next branch of your staging tree.

From: Xiong Zhou <[email protected]>

Add a config option for llite/lloop in lustre driver, making it depends on BLOCK to fix this better:
drivers/staging/lustre/lustre/fid/../include/linux/lustre_compat25.h:117:2:
error: implicit declaration of function ‘unregister_blkdev’

Quoted 7 lines

Also, remove the wrapper ll_unregister_blkdev which depends on BLOCK in the header and just call unregister_blkdev in lloop.c based on Peng Tao's comment. Drop the redundant dependency on STAGING for LUSTRE_FS, remove some unnecessary jdb header files which depends on BLOCK btw.

Please resend this so I don't have to edit the body of the email, and the Subject: line as well.

Also, you have a bunch of trailing whitespace in your changelog comments, are you sure you want/need that?

thanks,

greg k-h

#6 lore raw
In reply to Xiong Zhou
Quoted 6 lines · Xiong Zhou

From: Xiong Zhou <[email protected]>

In the lustre client driver, lloop depends on BLOCK. Add an option for this dependence. Last version of this patch makes LUSTRE_FS depends on BLOCK.
Remove unnecessary jdb head files which depends on BLOCK.

The driver should be removed, a filesystem has no business bringing its own loop driver.

#7 lore raw
Quoted 10 lines · Christoph Hellwig
Quoted 6 lines · Xiong Zhou

From: Xiong Zhou <[email protected]>

In the lustre client driver, lloop depends on BLOCK. Add an option for this dependence. Last version of this patch makes LUSTRE_FS depends on BLOCK.
Remove unnecessary jdb head files which depends on BLOCK.

The driver should be removed, a filesystem has no business bringing its own loop driver.

It provides significant performance improvement for network IO on Lustre. It bypasses DLM locking in Lustre and the VFS layer on the client, copying in the loop driver, and page-by-page IO submission in the normal IO path.

Cheers, Andreas

Signature · 4 lines
Andreas Dilger

Lustre Software Architect
Intel High Performance Data Division
#8 lore raw
Quoted 3 lines · Andreas

It provides significant performance improvement for network IO on Lustre. It bypasses DLM locking in Lustre and the VFS layer on the client, copying in the loop driver, and page-by-page IO submission in the normal IO path.

Part of being upstream is improving existing drivers instead of copy and pasting them. Please take a Look at Shaggys in-kernel direct I/O and loop improvements and submit any incremental improvements ontop of that one.

#9 lore raw
Quoted 8 lines · Christoph Hellwig
Quoted 6 lines · Andreas

It provides significant performance improvement for network IO on Lustre.
It bypasses DLM locking in Lustre and the VFS layer on the client, copying
in the loop driver, and page-by-page IO submission in the normal IO path.

Quoted 4 lines

Part of being upstream is improving existing drivers instead of copy and pasting them. Please take a Look at Shaggys in-kernel direct I/O and loop improvements and submit any incremental improvements ontop of that one.

The problem still remains that the kernel loop driver eventually depends on a local block device for the pages/bios to be written. The Lustre lloop driver bypasses the VFS and block layer to generate RPCs from the submitted pages to RDMA over the network without a data copy.

I wouldn't think that anyone would want a Lustre-specific RPC engine in the standard loop.c file, nor would it be practical due to symbol dependencies.

I could imagine being able to register do_bio_lustrebacked() as a BIO submission
routine instead of do_bio_filebacked(), along with some private data to link
the loop file to the underlying storage (in Lustre's case an object layout and
a preallocated I for generating the RPC).

How to register this from userspace compared to a normal file-backed loop might be a bit tricky. Lustre uses its own ioctls to register/deregister the
device, though I could imagine some kind of per-file(system) method table for
specifying the underlying bio submission routine.

In any case, rewriting the lloop/loop driver to merge this functionality is not
high on the priority list compared to other Lustre code that needs to be cleaned
up before it can be merged into the main kernel tree. Can we leave this code
as-is for now, and decide whether to rewrite or remove it when we are closer to
having the rest of the code cleaned up?

Cheers, Andreas

Signature · 4 lines
Andreas Dilger

Lustre Software Architect
Intel High Performance Data Division
#10 lore raw
Quoted 4 lines · Andreas

The problem still remains that the kernel loop driver eventually depends on a local block device for the pages/bios to be written. The Lustre lloop driver bypasses the VFS and block layer to generate RPCs from the submitted pages to RDMA over the network without a data copy.

No, it doesn't. It still consumes bios just like the regular loop driver.

Besides missing all kinds of fixes from years of kernel development the only difference is that it takes a lustre-specific shortcut into the direct I/O code instead of going through the pagecache.

The patch series I've pointed you to does exactly that in a generic way and thus superceeds the lloop driver fully.

In case my previous reference was a bit to vague the series starts at:

[PATCH V8 00/33] loop: Issue O_DIRECT aio using bio_vec

please take a look and make sure to review it in case you see any shortcomings.