Skip to contentSkip to conversation

sys_sync_file_range()

PATCH In review [email protected] 8 participants 17 messages Mar 30, 2006 – Apr 4, 2006

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sys_sync_file_range()

From: Andrew Morton <[email protected]>

Remove the recently-added LINUX_FADV_ASYNC_WRITE and LINUX_FADV_WRITE_WAIT fadvise() additions, do it in a new sys_sync_file_range() syscall instead. Reasons:

- It's more flexible. Things which would require two or three syscalls with fadvise() can be done in a single syscall.

- Using fadvise() in this manner is something not covered by POSIX.

The patch wires up the syscall for x86.

The sycall is implemented in the new fs/sync.c. The intention is that we can move sys_fsync(), sys_fdatasync() and perhaps sys_sync() into there later.

Documentation for the syscall is in fs/sync.c.

A test app (sync_file_range.c) is in
http://www.zip.com.au/~akpm/linux/patches/stuff/ext3-tools.tar.gz.

Note: the `async' writeout mode SYNC_FILE_RANGE_WRITE will turn synchronous if the queue is congested. This is trivial to fix: add a new flag bit, set wbc->nonblocking. But I'm not sure that we want to expose implementation details down to that level.

Note: it's notable that we can sync an fd which wasn't opened for writing. Same with fsync() and fdatasync()).

Note: the code takes some care to handle attempts to sync file contents outside the 16TB offset on 32-bit machines. It makes such attempts appears to succeed, for best 32-bit/64-bit compatibility. Perhaps it should make such requests fail...

Cc Nick Piggin <[email protected]>
Cc Michael Kerrisk <[email protected]>
Cc Ulrich Drepper <[email protected]>

---

 arch/i386/kernel/syscall_table.S |    1 
 fs/Makefile                      |    2 
 fs/sync.c                        |  147 +++++++++++++++++++++++++++++
 include/asm-i386/unistd.h        |    3 
 include/linux/fadvise.h          |    6 -
 include/linux/fs.h               |    5 
 include/linux/syscalls.h         |    2 
 mm/fadvise.c                     |   20 ---
 8 files changed, 158 insertions(+), 28 deletions(-)
mm/fadvise.c renamed from mm/fadvise.c~sys_sync_file_range +0 -20
diff -puN mm/fadvise.c~sys_sync_file_range mm/fadvise.c
@@ -35,17 +35,6 @@
  *
  * LINUX_FADV_ASYNC_WRITE: push some or all of the dirty pages at the disk.
  *
- * LINUX_FADV_WRITE_WAIT, LINUX_FADV_ASYNC_WRITE: push all of the currently
- * dirty pages at the disk.
- *
- * LINUX_FADV_WRITE_WAIT, LINUX_FADV_ASYNC_WRITE, LINUX_FADV_WRITE_WAIT: push
- * all of the currently dirty pages at the disk, wait until they have been
- * written.
- *
- * It should be noted that none of these operations write out the file's
- * metadata.  So unless the application is strictly performing overwrites of
- * already-instantiated disk blocks, there are no guarantees here that the data
- * will be available after a crash.
  */
 asmlinkage long sys_fadvise64_64(int fd, loff_t offset, loff_t len, int advice)
 {
@@ -129,15 +118,6 @@ asmlinkage long sys_fadvise64_64(int fd,
 			invalidate_mapping_pages(mapping, start_index,
 						end_index);
 		break;
-	case LINUX_FADV_ASYNC_WRITE:
-		ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
-						WB_SYNC_NONE);
-		break;
-	case LINUX_FADV_WRITE_WAIT:
-		ret = wait_on_page_writeback_range(mapping,
-					offset >> PAGE_CACHE_SHIFT,
-					endbyte >> PAGE_CACHE_SHIFT);
-		break;
 	default:
 		ret = -EINVAL;
 	}
include/linux/fadvise.h renamed from include/linux/fadvise.h~sys_sync_file_range +0 -6
diff -puN include/linux/fadvise.h~sys_sync_file_range include/linux/fadvise.h
@@ -18,10 +18,4 @@
 #define POSIX_FADV_NOREUSE	5 /* Data will be accessed once.  */
 #endif
 
-/*
- * Linux-specific fadvise() extensions:
- */
-#define LINUX_FADV_ASYNC_WRITE	32	/* Start writeout on range */
-#define LINUX_FADV_WRITE_WAIT	33	/* Wait upon writeout to range */
-
 #endif	/* FADVISE_H_INCLUDED */
fs/sync.c new file +147 -0

147 lines not on this page · raw message

include/linux/syscalls.h renamed from include/linux/syscalls.h~sys_sync_file_range +2 -0
diff -puN include/linux/syscalls.h~sys_sync_file_range include/linux/syscalls.h
@@ -569,5 +569,7 @@ asmlinkage long compat_sys_newfstatat(un
 asmlinkage long compat_sys_openat(unsigned int dfd, const char __user *filename,
 				   int flags, int mode);
 asmlinkage long sys_unshare(unsigned long unshare_flags);
+asmlinkage long sys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
+					int flags);
 
 #endif
include/linux/fs.h renamed from include/linux/fs.h~sys_sync_file_range +5 -0
diff -puN include/linux/fs.h~sys_sync_file_range include/linux/fs.h
@@ -757,6 +757,11 @@ extern void send_sigio(struct fown_struc
 extern int fcntl_setlease(unsigned int fd, struct file *filp, long arg);
 extern int fcntl_getlease(struct file *filp);
 
+/* fs/sync.c */
+#define SYNC_FILE_RANGE_WAIT_BEFORE	1
+#define SYNC_FILE_RANGE_WRITE		2
+#define SYNC_FILE_RANGE_WAIT_AFTER	4
+
 /* fs/locks.c */
 extern void locks_init_lock(struct file_lock *);
 extern void locks_copy_lock(struct file_lock *, struct file_lock *);
fs/Makefile renamed from fs/Makefile~sys_sync_file_range +1 -1
diff -puN fs/Makefile~sys_sync_file_range fs/Makefile
@@ -10,7 +10,7 @@ obj-y :=	open.o read_write.o file_table.
 		ioctl.o readdir.o select.o fifo.o locks.o dcache.o inode.o \
 		attr.o bad_inode.o file.o filesystems.o namespace.o aio.o \
 		seq_file.o xattr.o libfs.o fs-writeback.o mpage.o direct-io.o \
-		ioprio.o pnode.o drop_caches.o
+		ioprio.o pnode.o drop_caches.o sync.o
 
 obj-$(CONFIG_INOTIFY)		+= inotify.o
 obj-$(CONFIG_EPOLL)		+= eventpoll.o
include/asm-i386/unistd.h renamed from include/asm-i386/unistd.h~sys_sync_file_range +2 -1
diff -puN include/asm-i386/unistd.h~sys_sync_file_range include/asm-i386/unistd.h
@@ -318,8 +318,9 @@
 #define __NR_unshare		310
 #define __NR_set_robust_list	311
 #define __NR_get_robust_list	312
+#define __NR_sys_sync_file_range 313
 
-#define NR_syscalls 313
+#define NR_syscalls 314
 
 /*
  * user-visible error numbers are in the range -1 - -128: see
arch/i386/kernel/syscall_table.S renamed from arch/i386/kernel/syscall_table.S~sys_sync_file_range +1 -0
diff -puN arch/i386/kernel/syscall_table.S~sys_sync_file_range arch/i386/kernel/syscall_table.S
@@ -312,3 +312,4 @@ ENTRY(sys_call_table)
 	.long sys_unshare		/* 310 */
 	.long sys_set_robust_list
 	.long sys_get_robust_list
+	.long sys_sync_file_range

_

#1 lore raw
Quoted 6 lines · On Wednesday March 29, [email protected] wrote:

From: Andrew Morton <[email protected]>

Remove the recently-added LINUX_FADV_ASYNC_WRITE and LINUX_FADV_WRITE_WAIT fadvise() additions, do it in a new sys_sync_file_range() syscall instead.

Hmmm... any chance this could be split into a sys_sync_file_range and a vfs_sync_file_range which takes a 'struct file*' and does less (or no) sanity checking, so I can call it from nfsd?

Currently I implement COMMIT (which has a range) with a by messing around with filemap_fdatawrite and filemap_fdatawait (ignoring the range) and I'd rather than a vfs helper.

And in nfsd I call filp->f_op->fsync between the two. Doesn't sys_sync_file_range need to call into the filesystem at all?

NeilBrown

#2 lore raw
Quoted 5 lines · Neil Brown
Quoted 3 lines · On Wednesday March 29, [email protected] wrote:

From: Andrew Morton <[email protected]>

Quoted 7 lines
Quoted 3 lines

Remove the recently-added LINUX_FADV_ASYNC_WRITE and LINUX_FADV_WRITE_WAIT fadvise() additions, do it in a new sys_sync_file_range() syscall instead.

Hmmm... any chance this could be split into a sys_sync_file_range and a vfs_sync_file_range which takes a 'struct file*' and does less (or no) sanity checking, so I can call it from nfsd?

Coming right up. (Will switch it to fget_light() too)

Quoted 6 lines

Currently I implement COMMIT (which has a range) with a by messing around with filemap_fdatawrite and filemap_fdatawait (ignoring the range) and I'd rather than a vfs helper.

And in nfsd I call filp->f_op->fsync between the two. Doesn't sys_sync_file_range need to call into the filesystem at all?

Interesting question. sync_file_range() is purely a pagecache (ie: file contents) operation. It doesn't touch metadata at all.

So if it's being used for data-integrity purposes then it really only makes sense when it's doing file overwrites.

It does call into the filesystem of course - a_ops.writepages() and perhaps a_ops.writepage().

#3 lore raw
In reply to Neil Brown
Quoted 4 lines · Neil Brown

Hmmm... any chance this could be split into a sys_sync_file_range and
a vfs_sync_file_range which takes a 'struct file*' and does less (or no) sanity checking, so I can call it from nfsd?

Problem is, we don't appear to have a way of syncing the file's metadata without also syncing all of its pagecache.

For example, ext3_sync_file() will run a commit, which will sync all data and metadata.

ext2_sync_file() will also sync all pagecache as well as metadata. Even though do_fsync() already synced the file data (!).

Is the below still useful?

mm/fadvise.c renamed from mm/fadvise.c~sys_sync_file_range +0 -20
diff -puN mm/fadvise.c~sys_sync_file_range mm/fadvise.c
@@ -35,17 +35,6 @@
  *
  * LINUX_FADV_ASYNC_WRITE: push some or all of the dirty pages at the disk.
  *
- * LINUX_FADV_WRITE_WAIT, LINUX_FADV_ASYNC_WRITE: push all of the currently
- * dirty pages at the disk.
- *
- * LINUX_FADV_WRITE_WAIT, LINUX_FADV_ASYNC_WRITE, LINUX_FADV_WRITE_WAIT: push
- * all of the currently dirty pages at the disk, wait until they have been
- * written.
- *
- * It should be noted that none of these operations write out the file's
- * metadata.  So unless the application is strictly performing overwrites of
- * already-instantiated disk blocks, there are no guarantees here that the data
- * will be available after a crash.
  */
 asmlinkage long sys_fadvise64_64(int fd, loff_t offset, loff_t len, int advice)
 {
@@ -129,15 +118,6 @@ asmlinkage long sys_fadvise64_64(int fd,
 			invalidate_mapping_pages(mapping, start_index,
 						end_index);
 		break;
-	case LINUX_FADV_ASYNC_WRITE:
-		ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
-						WB_SYNC_NONE);
-		break;
-	case LINUX_FADV_WRITE_WAIT:
-		ret = wait_on_page_writeback_range(mapping,
-					offset >> PAGE_CACHE_SHIFT,
-					endbyte >> PAGE_CACHE_SHIFT);
-		break;
 	default:
 		ret = -EINVAL;
 	}
include/linux/fadvise.h renamed from include/linux/fadvise.h~sys_sync_file_range +0 -6
diff -puN include/linux/fadvise.h~sys_sync_file_range include/linux/fadvise.h
@@ -18,10 +18,4 @@
 #define POSIX_FADV_NOREUSE	5 /* Data will be accessed once.  */
 #endif
 
-/*
- * Linux-specific fadvise() extensions:
- */
-#define LINUX_FADV_ASYNC_WRITE	32	/* Start writeout on range */
-#define LINUX_FADV_WRITE_WAIT	33	/* Wait upon writeout to range */
-
 #endif	/* FADVISE_H_INCLUDED */
fs/sync.c new file +163 -0

163 lines not on this page · raw message

include/linux/syscalls.h renamed from include/linux/syscalls.h~sys_sync_file_range +2 -0
diff -puN include/linux/syscalls.h~sys_sync_file_range include/linux/syscalls.h
@@ -569,5 +569,7 @@ asmlinkage long compat_sys_newfstatat(un
 asmlinkage long compat_sys_openat(unsigned int dfd, const char __user *filename,
 				   int flags, int mode);
 asmlinkage long sys_unshare(unsigned long unshare_flags);
+asmlinkage long sys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
+					int flags);
 
 #endif
include/linux/fs.h renamed from include/linux/fs.h~sys_sync_file_range +7 -0
diff -puN include/linux/fs.h~sys_sync_file_range include/linux/fs.h
@@ -757,6 +757,13 @@ extern void send_sigio(struct fown_struc
 extern int fcntl_setlease(unsigned int fd, struct file *filp, long arg);
 extern int fcntl_getlease(struct file *filp);
 
+/* fs/sync.c */
+#define SYNC_FILE_RANGE_WAIT_BEFORE	1
+#define SYNC_FILE_RANGE_WRITE		2
+#define SYNC_FILE_RANGE_WAIT_AFTER	4
+extern int do_sync_file_range(struct file *file, loff_t offset, loff_t endbyte,
+			int flags);
+
 /* fs/locks.c */
 extern void locks_init_lock(struct file_lock *);
 extern void locks_copy_lock(struct file_lock *, struct file_lock *);
fs/Makefile renamed from fs/Makefile~sys_sync_file_range +1 -1
diff -puN fs/Makefile~sys_sync_file_range fs/Makefile
@@ -10,7 +10,7 @@ obj-y :=	open.o read_write.o file_table.
 		ioctl.o readdir.o select.o fifo.o locks.o dcache.o inode.o \
 		attr.o bad_inode.o file.o filesystems.o namespace.o aio.o \
 		seq_file.o xattr.o libfs.o fs-writeback.o mpage.o direct-io.o \
-		ioprio.o pnode.o drop_caches.o
+		ioprio.o pnode.o drop_caches.o sync.o
 
 obj-$(CONFIG_INOTIFY)		+= inotify.o
 obj-$(CONFIG_EPOLL)		+= eventpoll.o
include/asm-i386/unistd.h renamed from include/asm-i386/unistd.h~sys_sync_file_range +2 -1
diff -puN include/asm-i386/unistd.h~sys_sync_file_range include/asm-i386/unistd.h
@@ -318,8 +318,9 @@
 #define __NR_unshare		310
 #define __NR_set_robust_list	311
 #define __NR_get_robust_list	312
+#define __NR_sys_sync_file_range 313
 
-#define NR_syscalls 313
+#define NR_syscalls 314
 
 /*
  * user-visible error numbers are in the range -1 - -128: see
arch/i386/kernel/syscall_table.S renamed from arch/i386/kernel/syscall_table.S~sys_sync_file_range +1 -0
diff -puN arch/i386/kernel/syscall_table.S~sys_sync_file_range arch/i386/kernel/syscall_table.S
@@ -312,3 +312,4 @@ ENTRY(sys_call_table)
 	.long sys_unshare		/* 310 */
 	.long sys_set_robust_list
 	.long sys_get_robust_list
+	.long sys_sync_file_range

_

#4 lore raw
Quoted 9 lines · On Thursday March 30, [email protected] wrote:
Quoted 4 lines · Neil Brown

Hmmm... any chance this could be split into a sys_sync_file_range and
a vfs_sync_file_range which takes a 'struct file*' and does less (or no) sanity checking, so I can call it from nfsd?

Problem is, we don't appear to have a way of syncing the file's metadata without also syncing all of its pagecache.

Quoted 8 lines

For example, ext3_sync_file() will run a commit, which will sync all data and metadata.

ext2_sync_file() will also sync all pagecache as well as metadata. Even though do_fsync() already synced the file data (!).

Is the below still useful?

Yes. A COMMIT can say NFS_DATA_SYNC or NFS_FILE_SYNC. I can skip the ->fsync call for NFS_DATA_SYNC which is hopefully the more common. I didn't before because when I was righting that (2.4 days) there were a lot fewer options on how to sync things.

So yes, that will be perfect, thanks.

NeilBrown

#5 lore raw
In reply to Andrew Morton

fs/sync.c line 80

Quoted 4 lines · Andrew Morton
fs/sync.c new file +4 -0
+	if ((s64)offset < 0)
+		goto out;
+	if ((s64)endbyte < 0)
+		goto out;

loff_t is long long on all arch. This is not need?

fs/sync.c line 115

Quoted 4 lines
fs/sync.c new file +4 -0
+	if (S_ISFIFO(file->f_dentry->d_inode->i_mode)) {
+		ret = -ESPIPE;
+		goto out_put;
+	}

How about to check "if (!file->f_op || !file->f_op->fsync)" or something?

For chrdev is also strange, and in the case of fsync(), it returns -EINVAL. IMHO it seems there is consistency.

Thanks.

Signature · 1 line
OGAWA Hirofumi <[email protected]>
#6 lore raw
Quoted 9 lines · OGAWA Hirofumi

fs/sync.c line 80

Quoted 4 lines · Andrew Morton
fs/sync.c new file +4 -0
+	if ((s64)offset < 0)
+		goto out;
+	if ((s64)endbyte < 0)
+		goto out;

loff_t is long long on all arch. This is not need?

True, the casts happen to be unneeded. But they do set the reader's mind at ease.

Quoted 6 lines

fs/sync.c line 115

Quoted 4 lines
fs/sync.c new file +4 -0
+	if (S_ISFIFO(file->f_dentry->d_inode->i_mode)) {
+		ret = -ESPIPE;
+		goto out_put;
+	}

How about to check "if (!file->f_op || !file->f_op->fsync)" or something?

This syscall won't call ->fsync.

Quoted 3 lines

For chrdev is also strange, and in the case of fsync(), it returns -EINVAL. IMHO it seems there is consistency.

I guess so. Perhaps it should be S_ISREG|S_ISBLK|S_ISDIR|S_ISLNK.

#7 lore raw
In reply to Neil Brown
Quoted 5 lines · Neil Brown
Quoted 3 lines · On Wednesday March 29, [email protected] wrote:

Remove the recently-added LINUX_FADV_ASYNC_WRITE and LINUX_FADV_WRITE_WAIT fadvise() additions, do it in a new sys_sync_file_range() syscall instead.

Quoted 7 lines

Hmmm... any chance this could be split into a sys_sync_file_range and a vfs_sync_file_range which takes a 'struct file*' and does less (or no) sanity checking, so I can call it from nfsd?

Currently I implement COMMIT (which has a range) with a by messing around with filemap_fdatawrite and filemap_fdatawait (ignoring the range) and I'd rather than a vfs helper.

I'm not 100% sure, but it looks like the PF_SYNCWRITE process flag should be set on the nfsd's while they're doing that, which doesn't seem to be happening atm. Looks like a couple of the IO schedulers will make use of that knowledge now. All the more reason for a VFS helper here I guess. ;)

cheers.

Signature · 21 lines
Nathan


Index: 2.6.x-xfs/fs/nfsd/vfs.c
===================================================================
--- 2.6.x-xfs.orig/fs/nfsd/vfs.c
+++ 2.6.x-xfs/fs/nfsd/vfs.c
@@ -712,11 +712,13 @@ static inline int nfsd_dosync(struct fil
 	int (*fsync) (struct file *, struct dentry *, int);
 	int err;
 
+	current->flags |= PF_SYNCWRITE;
 	err = filemap_fdatawrite(inode->i_mapping);
 	if (err == 0 && fop && (fsync = fop->fsync))
 		err = fsync(filp, dp, 0);
 	if (err == 0)
 		err = filemap_fdatawait(inode->i_mapping);
+	current->flags &= ~PF_SYNCWRITE;
 
 	return err;
 }
#8 lore raw
Quoted 14 lines · On Friday March 31, [email protected] wrote:
Quoted 5 lines · Neil Brown
Quoted 3 lines · On Wednesday March 29, [email protected] wrote:

Remove the recently-added LINUX_FADV_ASYNC_WRITE and LINUX_FADV_WRITE_WAIT fadvise() additions, do it in a new sys_sync_file_range() syscall instead.

Quoted 7 lines

Hmmm... any chance this could be split into a sys_sync_file_range and a vfs_sync_file_range which takes a 'struct file*' and does less (or no) sanity checking, so I can call it from nfsd?

Currently I implement COMMIT (which has a range) with a by messing around with filemap_fdatawrite and filemap_fdatawait (ignoring the range) and I'd rather than a vfs helper.

Quoted 5 lines

I'm not 100% sure, but it looks like the PF_SYNCWRITE process flag should be set on the nfsd's while they're doing that, which doesn't seem to be happening atm. Looks like a couple of the IO schedulers will make use of that knowledge now. All the more reason for a VFS helper here I guess. ;)

PF_SYNCWRITE? What's that???

(find | xargs grep ...)
Oh. The block device schedulers like to know if a request is sync or async (and all reads are assumed to be sync) - which is reasonable - and so have a per-task flag to tell them - which isn't (IMO).

md/raid (particularly raid5) often does the write from a different process than generated the original request, so that will break completely.

What is wrong with a bio flag I wonder....

Hopefully this will get wrapped up in a do_X helper so nfsd won't need to know about it.

Thanks for the heads-up.

NeilBrown

#9 lore raw
Quoted 23 lines · Neil Brown
Quoted 14 lines · On Friday March 31, [email protected] wrote:
Quoted 5 lines · Neil Brown
Quoted 3 lines · On Wednesday March 29, [email protected] wrote:

Remove the recently-added LINUX_FADV_ASYNC_WRITE and LINUX_FADV_WRITE_WAIT fadvise() additions, do it in a new sys_sync_file_range() syscall instead.

Quoted 7 lines

Hmmm... any chance this could be split into a sys_sync_file_range and a vfs_sync_file_range which takes a 'struct file*' and does less (or no) sanity checking, so I can call it from nfsd?

Currently I implement COMMIT (which has a range) with a by messing around with filemap_fdatawrite and filemap_fdatawait (ignoring the range) and I'd rather than a vfs helper.

Quoted 5 lines

I'm not 100% sure, but it looks like the PF_SYNCWRITE process flag should be set on the nfsd's while they're doing that, which doesn't seem to be happening atm. Looks like a couple of the IO schedulers will make use of that knowledge now. All the more reason for a VFS helper here I guess. ;)

PF_SYNCWRITE? What's that???

Quoted 8 lines

(find | xargs grep ...)
Oh. The block device schedulers like to know if a request is sync or async (and all reads are assumed to be sync) - which is reasonable - and so have a per-task flag to tell them - which isn't (IMO).

md/raid (particularly raid5) often does the write from a different process than generated the original request, so that will break completely.

I don't think any disagrees with you, the sync-write process flag is indeed an atrocious beast...

Quoted 1 line

What is wrong with a bio flag I wonder....

Nothing, in fact I would love for it to be changed. I'm sure such a patch would be accepted with open arms! :-)

Signature · 1 line
Jens Axboe
#10 lore raw
Quoted 30 lines · Jens Axboe
Quoted 28 lines · Neil Brown
Quoted 14 lines · On Friday March 31, [email protected] wrote:
Quoted 5 lines · Neil Brown
Quoted 3 lines · On Wednesday March 29, [email protected] wrote:

Remove the recently-added LINUX_FADV_ASYNC_WRITE and LINUX_FADV_WRITE_WAIT fadvise() additions, do it in a new sys_sync_file_range() syscall instead.

Quoted 7 lines

Hmmm... any chance this could be split into a sys_sync_file_range and a vfs_sync_file_range which takes a 'struct file*' and does less (or no) sanity checking, so I can call it from nfsd?

Currently I implement COMMIT (which has a range) with a by messing around with filemap_fdatawrite and filemap_fdatawait (ignoring the range) and I'd rather than a vfs helper.

Quoted 5 lines

I'm not 100% sure, but it looks like the PF_SYNCWRITE process flag should be set on the nfsd's while they're doing that, which doesn't seem to be happening atm. Looks like a couple of the IO schedulers will make use of that knowledge now. All the more reason for a VFS helper here I guess. ;)

PF_SYNCWRITE? What's that???

(find | xargs grep ...)
Oh. The block device schedulers like to know if a request is sync or async (and all reads are assumed to be sync) - which is reasonable - and so have a per-task flag to tell them - which isn't (IMO).

Quoted 6 lines
Quoted 3 lines

md/raid (particularly raid5) often does the write from a different process than generated the original request, so that will break completely.

I don't think any disagrees with you, the sync-write process flag is indeed an atrocious beast...

Yeah. PF_SYNCWRITE was a performance tweak for the anticipatory scheduler. As cfq is using it as well now (hopefully to good effect) I guess it could be formalised more.

Quoted 4 lines
Quoted 1 line

What is wrong with a bio flag I wonder....

Nothing, in fact I would love for it to be changed. I'm sure such a patch would be accepted with open arms! :-)

I think once someone starts coding it, they'll become a big fan of PF_SYNCWRITE...

For the page writeback functions it's probably possible to use writeback_control.sync_mode=WB_SYNC_ALL as a trigger, propagate that into the IO layer. Maybe that'll always be sufficient - it's hard to tell. The writeback paths are twisty and deep...

Then again, using WB_SYNC_ALL as a hint that this process will be waiting for this writeout to complete is a bit hacky too - it doesn't _really_ mean that - it just means that I/O should be _started_ against all relevant dirty data.

Good luck ;)

#11 lore raw
Quoted 35 lines · Andrew Morton
Quoted 34 lines · Jens Axboe
Quoted 31 lines · Neil Brown
Quoted 14 lines · On Friday March 31, [email protected] wrote:
Quoted 5 lines · Neil Brown
Quoted 3 lines · On Wednesday March 29, [email protected] wrote:

Remove the recently-added LINUX_FADV_ASYNC_WRITE and LINUX_FADV_WRITE_WAIT fadvise() additions, do it in a new sys_sync_file_range() syscall instead.

Quoted 7 lines

Hmmm... any chance this could be split into a sys_sync_file_range and a vfs_sync_file_range which takes a 'struct file*' and does less (or no) sanity checking, so I can call it from nfsd?

Currently I implement COMMIT (which has a range) with a by messing around with filemap_fdatawrite and filemap_fdatawait (ignoring the range) and I'd rather than a vfs helper.

Quoted 5 lines

I'm not 100% sure, but it looks like the PF_SYNCWRITE process flag should be set on the nfsd's while they're doing that, which doesn't seem to be happening atm. Looks like a couple of the IO schedulers will make use of that knowledge now. All the more reason for a VFS helper here I guess. ;)

PF_SYNCWRITE? What's that???

(find | xargs grep ...)
Oh. The block device schedulers like to know if a request is sync or async (and all reads are assumed to be sync) - which is reasonable - and so have a per-task flag to tell them - which isn't (IMO).

md/raid (particularly raid5) often does the write from a different process than generated the original request, so that will break completely.

Quoted 6 lines
Quoted 2 lines

I don't think any disagrees with you, the sync-write process flag is indeed an atrocious beast...

Yeah. PF_SYNCWRITE was a performance tweak for the anticipatory scheduler. As cfq is using it as well now (hopefully to good effect) I guess it could be formalised more.

Yup, both 'as' and 'cfq' would prefer to just look at a SYNC bio flag instead. But the logic itself is definitely needed.

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

What is wrong with a bio flag I wonder....

Nothing, in fact I would love for it to be changed. I'm sure such a patch would be accepted with open arms! :-)

I think once someone starts coding it, they'll become a big fan of PF_SYNCWRITE...

They might not become a big fan, but they'll surely appreciate the simplicity of it :-)

Quoted 11 lines

For the page writeback functions it's probably possible to use writeback_control.sync_mode=WB_SYNC_ALL as a trigger, propagate that into the IO layer. Maybe that'll always be sufficient - it's hard to tell. The writeback paths are twisty and deep...

Then again, using WB_SYNC_ALL as a hint that this process will be waiting for this writeout to complete is a bit hacky too - it doesn't _really_ mean that - it just means that I/O should be _started_ against all relevant dirty data.

Good luck ;)

It's not a hard problem, but it will definitely cost a little sweat to go through. I'm sure Neil could pull it off, the question is more if he wants to :-)

Signature · 1 line
Jens Axboe
#12 lore raw
Quoted 10 lines · Jens Axboe
Quoted 7 lines
Quoted 3 lines

I don't think any disagrees with you, the sync-write process flag is indeed an atrocious beast...

Yeah. PF_SYNCWRITE was a performance tweak for the anticipatory scheduler. As cfq is using it as well now (hopefully to good effect) I guess it could be formalised more.

Yup, both 'as' and 'cfq' would prefer to just look at a SYNC bio flag instead. But the logic itself is definitely needed.

hm. I actually thought we were already doing that. We should at least tranfer PF_SYNCWRITE into bi_flags at the point where we start to construct the BIO.

That might well fix RAID, too. If it's handing work off to another thread via BIOs.

#13 lore raw
In reply to [email protected]

include/asm-i386/unistd.h line 321

Quoted 5 lines · On Wed, Mar 29, 2006 at 11:41:11PM -0800, [email protected] wrote:
include/asm-i386/unistd.h renamed from include/asm-i386/unistd.h~sys_sync_file_range +1 -0
@@ -318,8 +318,9 @@
 #define __NR_unshare		310
 #define __NR_set_robust_list	311
 #define __NR_get_robust_list	312
+#define __NR_sys_sync_file_range 313

What's up with the __NR_sys_yada_yada? Except for recent entries (kexec, splice, and now sync_file_range) all of the other names in here have dropped the "sys_".

Is it too late to fix __NR_sys_kexec_load (since it is out in the wild now?)

-Tony

#14 lore raw
Quoted 11 lines · Tony

Review line 321

Quoted 5 lines · On Wed, Mar 29, 2006 at 11:41:11PM -0800, [email protected] wrote:
(quoted diff) +1 -0
@@ -318,8 +318,9 @@
 #define __NR_unshare		310
 #define __NR_set_robust_list	311
 #define __NR_get_robust_list	312
+#define __NR_sys_sync_file_range 313

What's up with the __NR_sys_yada_yada? Except for recent entries (kexec, splice, and now sync_file_range) all of the other names in here have dropped the "sys_".

Quoted 2 lines

Is it too late to fix __NR_sys_kexec_load (since it is out in the wild now?)

already been done: (copy-n-paste warning here:)
~~~~~

On i386, we don't use sys_ prefix for __NR_*. This patch removes it. [FWIW, _syscall*() macros will generate foo() instead of sys_foo().]

---

 include/asm-i386/unistd.h |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
include/asm-i386/unistd.h renamed from include/asm-i386/unistd.h~remove-sys_-prefix +2 -2
diff -puN include/asm-i386/unistd.h~remove-sys_-prefix include/asm-i386/unistd.h
@@ -318,8 +318,8 @@
 #define __NR_unshare		310
 #define __NR_set_robust_list	311
 #define __NR_get_robust_list	312
-#define __NR_sys_splice		313
-#define __NR_sys_sync_file_range 314
+#define __NR_splice		313
+#define __NR_sync_file_range	314
 
 #define NR_syscalls 315
 

---
~Randy

#15 lore raw
In reply to Luck, Tony
Quoted 9 lines · Tony

include/asm-i386/unistd.h line 321

Quoted 5 lines · On Wed, Mar 29, 2006 at 11:41:11PM -0800, [email protected] wrote:
include/asm-i386/unistd.h renamed from include/asm-i386/unistd.h~sys_sync_file_range +1 -0
@@ -318,8 +318,9 @@
 #define __NR_unshare		310
 #define __NR_set_robust_list	311
 #define __NR_get_robust_list	312
+#define __NR_sys_sync_file_range 313

What's up with the __NR_sys_yada_yada?

brainfart. sync_file_range() and splice() will get fixed once Linus resurfaces.

Quoted 6 lines

Except for recent entries (kexec,
splice, and now sync_file_range) all of the other names in here have dropped the "sys_".

Is it too late to fix __NR_sys_kexec_load (since it is out in the wild now?)

It only affects users of the _syscallN macros in unistd.h. I don't expect that fixing kexec will break much.

#16 lore raw
In reply to Luck, Tony
Quoted 2 lines · Tony

Is it too late to fix __NR_sys_kexec_load (since it is out in the wild now?)

It's not too late IMO. The only real users are the libcs. And in fact, I'm already checking for __NR_sync_file_range and not the sys_ variant.
Andrew, please change it.

Signature · 1 line
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖