loop.c to use write ops for fs requiring special locking
Subsystems (2)
PATCH 1/1 loop.c to use write ops for fs requiring special locking
Hi All,
Below is a small patch to loop.c I'd like to see in the kernel.
This is an extension of Anton Altaparmakov's previous fix which allows loop.c to use the aop->write rather than prepare_write/commit_write if prepare_write/commit_write aren't available.
Right now, the current loop.c uses aop->prepare_write/commit_write unless there is no other option. However, due to special locking requirements, some backing filesystems may prefer the use of aop->write rather than prepare_write/commit_write. Since loop.c does not have advisory locking, the backing fs should have a choice of which to use.
In the case of GFS, for example, loop.c's use of aop->prepare_write circumvents proper cluster locking and transaction building, so using aop->write is the right thing for loop.c to do.
How the patch works:
If the backing filesystem has special locking requirements (new flag in fs_flags) loop.c uses aop->write rather than prepare_write/commit_write.
Feedback?
Regards,
Bob Peterson
[email protected]
--- linux-2.6.15.4.orig/drivers/block/loop.c 2006-02-10
01:22:48.000000000 -0600
+++ linux-2.6.15.4.patched/drivers/block/loop.c 2006-03-01 09:38:48.000000000 -0600
@@ -44,6 +44,11 @@
* backing filesystem.
* Anton Altaparmakov, 16 Feb 2005
*
+ * Extension of Anton's idea: Use normal write file operations rather
+ * than prepare_write and commit_write when the backing filesystem
+ * requires special locking.
+ * Robert Peterson <[email protected]>, 01 Mar 2006
+ *
* Still To Fix:
* - Advisory locking is ignored here.
* - Should use an own CAP_* category instead of CAP_SYS_ADMIN
@@ -74,6 +79,7 @@
#include <linux/completion.h>
#include <linux/highmem.h>
#include <linux/gfp.h>
+#include <linux/mount.h>
#include <asm/uaccess.h>
@@ -781,7 +787,8 @@ static int loop_set_fd(struct loop_devic
*/
if (!file->f_op->sendfile)
goto out_putf;- if (aops->prepare_write && aops->commit_write)
+ if (!(file->f_vfsmnt->mnt_sb->s_type->fs_flags &FS_REQUIRES_LOCKING) &&
+ aops->prepare_write && aops->commit_write)
lo_flags |= LO_FLAGS_USE_AOPS;
if (!(lo_flags & LO_FLAGS_USE_AOPS) && !
file->f_op->write)
lo_flags |= LO_FLAGS_READ_ONLY;diff -pur linux-2.6.15.4.orig/include/linux/fs.h linux-2.6.15.4.patched/include/linux/fs.h
--- linux-2.6.15.4.orig/include/linux/fs.h 2006-02-10
01:22:48.000000000 -0600
+++ linux-2.6.15.4.patched/include/linux/fs.h 2006-02-28
17:18:48.000000000 -0600
@@ -83,6 +83,7 @@ extern int dir_notify_enable;
/* public flags for file_system_type */
#define FS_REQUIRES_DEV 1
#define FS_BINARY_MOUNTDATA 2
+#define FS_REQUIRES_LOCKING 4 /* Filesystem requires locking */
#define FS_REVAL_DOT 16384 /* Check the paths ".", ".." for
staleness */
#define FS_ODD_RENAME 32768 /* Temporary stuff; will go away as soon
* as nfs_rename() will be cleaned up
Quoted 8 lines · Robert S Peterson
Hi All,
Below is a small patch to
loop.cI'd like to see in the kernel.This is an extension of Anton Altaparmakov's previous fix which allows
loop.cto use theaop->writerather than prepare_write/commit_write if prepare_write/commit_write aren't available.
There's a reason why we must use prepare_write/commit_write rather than ->write() and ow-ow-my-brain-hurts I can't remember what it was. Anton, do you recall?
Quoted 6 lines
Right now, the current
loop.cuses aop->prepare_write/commit_write unless there is no other option. However, due to special locking requirements, some backing filesystems may prefer the use ofaop->writerather than prepare_write/commit_write. Sinceloop.cdoes not have advisory locking, the backing fs should have a choice of which to use.
Quoted 8 lines
In the case of GFS, for example, loop.c's use of
aop->prepare_writecircumvents proper cluster locking and transaction building, so usingaop->writeis the right thing forloop.cto do.How the patch works:
If the backing filesystem has special locking requirements (new flag infs_flags)loop.cusesaop->writerather than prepare_write/commit_write.
I think you'll find that cryptoloop doesn't work (see "ow-ow", above).
Quoted 4 lines
---
linux-2.6.15.4.orig/drivers/block/loop.c2006-02-10
01:22:48.000000000 -0600
+++linux-2.6.15.4.patched/drivers/block/loop.c2006-03-01 09:38:48.000000000 -0600
(The patch is wordwrapped)
Quoted 10 lines · Andrew Morton
Quoted 5 lines · Robert S Peterson
Below is a small patch to
loop.cI'd like to see in the kernel.This is an extension of Anton Altaparmakov's previous fix which allows
loop.cto use theaop->writerather than prepare_write/commit_write if prepare_write/commit_write aren't available.There's a reason why we must use prepare_write/commit_write rather than ->write() and ow-ow-my-brain-hurts I can't remember what it was. Anton, do you recall?
It is simply that it is faster on all file systems that can use prepare_write/commit_write and the "performance hungry" would have complained if it were to be removed.
Quoted 10 lines
Quoted 10 lines
Right now, the current
loop.cuses aop->prepare_write/commit_write unless there is no other option. However, due to special locking requirements, some backing filesystems may prefer the use ofaop->writerather than prepare_write/commit_write. Sinceloop.cdoes not have advisory locking, the backing fs should have a choice of which to use.In the case of GFS, for example, loop.c's use of
aop->prepare_writecircumvents proper cluster locking and transaction building, so usingaop->writeis the right thing forloop.cto do.
Quoted 6 lines
Quoted 4 lines
How the patch works:
If the backing filesystem has special locking requirements (new flag infs_flags)loop.cusesaop->writerather than prepare_write/commit_write.I think you'll find that cryptoloop doesn't work (see "ow-ow", above).
*blink* Why should it not work?
cryptoloop is exactly the reason for the performance hit with using ->write rather than prepare_write/commit_write as we need to allocate a temporary page, do the transformation from the source page to the temporary page, then send the temporary page to ->write, and finally release the page. Ugly but the only way to allow transformations in-flight I could see when writing the ->write using code for loop.c...
Best regards,
Anton
Signature · 4 lines
Anton Altaparmakov <aia21 at cam.ac.uk> (replace at with @) Unix Support, Computing Service, University of Cambridge, CB2 3QH, UK Linux NTFS maintainer / IRC: #ntfs on irc.freenode.net WWW: http://linux-ntfs.sf.net/ & http://www-stu.christs.cam.ac.uk/~aia21/
Quoted 3 lines · Anton Altaparmakov
It is simply that it is faster on all file systems that can use prepare_write/commit_write and the "performance hungry" would have complained if it were to be removed.
(snip)
Quoted 1 line
Anton
Having heard no objections to my loop.c patch, when can I expect it to be integrated into the 2.6 kernel?
Regards,
Bob Peterson
Quoted 11 lines · Robert S Peterson
Quoted 3 lines · Anton Altaparmakov
It is simply that it is faster on all file systems that can use prepare_write/commit_write and the "performance hungry" would have complained if it were to be removed.
(snip)
Quoted 1 line
Anton
Having heard no objections to my
loop.cpatch, when can I expect it to be integrated into the 2.6 kernel?
When we've remembered what Al's statement meant in
http://marc.theaimsgroup.com/?t=102129995600002&r=1&w=2
I knew at the time, but I've forgotten. Maybe the issue went away..
Quoted 8 lines · Andrew Morton
Quoted 7 lines · Robert S Peterson
Quoted 3 lines · Anton Altaparmakov
It is simply that it is faster on all file systems that can use prepare_write/commit_write and the "performance hungry" would have complained if it were to be removed.
(snip)
Quoted 1 line
Anton
Quoted 7 lines
Quoted 2 lines
Having heard no objections to my
loop.cpatch, when can I expect it to be integrated into the 2.6 kernel?When we've remembered what Al's statement meant in
http://marc.theaimsgroup.com/?t=102129995600002&r=1&w=2I knew at the time, but I've forgotten. Maybe the issue went away..
The comment is that the patch put forward by Urban in the above thread replaced the use of readpage for reading with file ->read and prepare_write/commit_write for writing with file ->write without thinking about it. So he left out the entire transformation on the data which is what Al correctly complained about.
I.e. Urban's patch made loop only work for a "straight" loopback mount without any changes to the data (i.e. crypto) applied to it.
This is no longer relevant because my patch which is in current 2.6 kernels (can't remember when I wrote it/when it got put into 2.6) did the conversion from prepare_write/commit_write to file ->write correctly so the data transformation still happens so that crypto still works with the loop driver.
To get back to Robert's patch that he is requesting to be included. I think it is fine but the flag name could perhaps be better. Perhaps "FS_AOPS_PRIVATE" or "FS_AOPS_SPECIAL" or "FS_AOPS_NEED_LOCKING" or even "FS_AOPS_REQUIRE_LOCKING" or something. "FS_REQUIRES_LOCKING" just does not mean much and certainly would not suggest to me that no-one outside the file system should use the address space operations of the file system... But maybe I am just bein picky. (-:
Best regards,
Anton
Signature · 4 lines
Anton Altaparmakov <aia21 at cam.ac.uk> (replace at with @) Unix Support, Computing Service, University of Cambridge, CB2 3QH, UK Linux NTFS maintainer / IRC: #ntfs on irc.freenode.net WWW: http://linux-ntfs.sf.net/ & http://www-stu.christs.cam.ac.uk/~aia21/
PATCH 1/1 loop.c to use write ops for fs requiring special locking [try #2]
Quoted 9 lines · Anton Altaparmakov
Quoted 2 lines · Andrew Morton
When we've remembered what Al's statement meant in
http://marc.theaimsgroup.com/?t=102129995600002&r=1&w=2This is no longer relevant because my patch which is in current 2.6 kernels (can't remember when I wrote it/when it got put into 2.6) did the conversion from prepare_write/commit_write to file ->write correctly so the data transformation still happens so that crypto still works with the loop driver.
Quoted 7 lines
To get back to Robert's patch that he is requesting to be included. I think it is fine but the flag name could perhaps be better. Perhaps "
FS_AOPS_PRIVATE" or "FS_AOPS_SPECIAL" or "FS_AOPS_NEED_LOCKING" or even "FS_AOPS_REQUIRE_LOCKING" or something. "FS_REQUIRES_LOCKING" just does not mean much and certainly would not suggest to me that no-one outside the file system should use the address space operations of the file system... But maybe I am just bein picky. (-:
Here is a resubmission of my patch to loop.c, this time against the 2.6.16 kernel.
Andrew: Sounds like Anton answered your concerns. Anton: As per your suggestion, I changed the constant to your suggested
FS_AOPS_NEED_LOCKING.
Regards,
Bob Peterson
[email protected]
drivers/block/loop.c 2006-03-24 10:49:10.000000000 -0600 renamed from drivers/block/loop.c 2006-03-19 23:53:29.000000000 -0600 +8 -1
@@ -44,6 +44,11 @@ * backing filesystem. * Anton Altaparmakov, 16 Feb 2005 * + * Extension of Anton's idea: Use normal write file operations rather than + * prepare_write and commit_write when the backing filesystem requires + * special locking. + * Robert Peterson <[email protected]>, 01 Mar 2006 + * * Still To Fix: * - Advisory locking is ignored here. * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
@@ -74,6 +79,7 @@ #include <linux/completion.h> #include <linux/highmem.h> #include <linux/gfp.h> +#include <linux/mount.h> #include <asm/uaccess.h>
@@ -791,7 +797,8 @@ static int loop_set_fd(struct loop_devic */ if (!file->f_op->sendfile) goto out_putf; - if (aops->prepare_write && aops->commit_write) + if (!(file->f_vfsmnt->mnt_sb->s_type->fs_flags & FS_AOPS_NEED_LOCKING) && + aops->prepare_write && aops->commit_write) lo_flags |= LO_FLAGS_USE_AOPS; if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write) lo_flags |= LO_FLAGS_READ_ONLY;
include/linux/fs.h 2006-03-24 10:27:20.000000000 -0600 renamed from include/linux/fs.h 2006-03-19 23:53:29.000000000 -0600 +1 -0
@@ -83,6 +83,7 @@ extern int dir_notify_enable; /* public flags for file_system_type */ #define FS_REQUIRES_DEV 1 #define FS_BINARY_MOUNTDATA 2 +#define FS_AOPS_NEED_LOCKING 4 /* Filesystem aops have special locking needs */ #define FS_REVAL_DOT 16384 /* Check the paths ".", ".." for staleness */ #define FS_ODD_RENAME 32768 /* Temporary stuff; will go away as soon * as nfs_rename() will be cleaned up
Re: [patch 2.6.16] loop.c to use write ops for fs requiring special locking [try #2]
Quoted 18 lines · Robert S Peterson
Quoted 9 lines · Anton Altaparmakov
Quoted 2 lines · Andrew Morton
When we've remembered what Al's statement meant in
http://marc.theaimsgroup.com/?t=102129995600002&r=1&w=2This is no longer relevant because my patch which is in current 2.6 kernels (can't remember when I wrote it/when it got put into 2.6) did the conversion from prepare_write/commit_write to file ->write correctly so the data transformation still happens so that crypto still works with the loop driver.
Quoted 7 lines
To get back to Robert's patch that he is requesting to be included. I think it is fine but the flag name could perhaps be better. Perhaps "
FS_AOPS_PRIVATE" or "FS_AOPS_SPECIAL" or "FS_AOPS_NEED_LOCKING" or even "FS_AOPS_REQUIRE_LOCKING" or something. "FS_REQUIRES_LOCKING" just does not mean much and certainly would not suggest to me that no-one outside the file system should use the address space operations of the file system... But maybe I am just bein picky. (-:
Quoted 6 lines
Here is a resubmission of my patch to
loop.c, this time against the 2.6.16 kernel.Andrew: Sounds like Anton answered your concerns. Anton: As per your suggestion, I changed the constant to your suggested
FS_AOPS_NEED_LOCKING.
Looks good.
Best regards,
Anton
Signature · 4 lines
Anton Altaparmakov <aia21 at cam.ac.uk> (replace at with @) Unix Support, Computing Service, University of Cambridge, CB2 3QH, UK Linux NTFS maintainer / IRC: #ntfs on irc.freenode.net WWW: http://linux-ntfs.sf.net/ & http://www-stu.christs.cam.ac.uk/~aia21/