fs: dodge an atomic in putname if ref == 1
PATCH 1/1 fs: dodge an atomic in putname if ref == 1
While the structure is refcounted, the only consumer incrementing it is audit and even then the atomic operation is only needed when it interacts with io_uring.
If putname spots a count of 1, there is no legitimate way for anyone to bump it.
If audit is disabled, the count is guaranteed to be 1, which consistently elides the atomic for all path lookups. If audit is enabled, it still manages to elide the last decrement.
Note the patch does not do anything to prevent audit from suffering atomics. See [1] and [2] for a different approach.
Benchmarked on Sapphire Rapids issuing access() (ops/s):
before: 5106246 after: 5269678 (+3%)
Link 1: https://lore.kernel.org/linux-fsdevel/[email protected]/
Link 2: https://lore.kernel.org/linux-fsdevel/20250307164216.GI2023217@ZenIV/
---
This is an alternative to the patch I linked above.
I think the improved commit message should also cover the feedback Christian previously shared concerning it.
This is a trivial win until the atomic issue gets resolved, I don't see any reason to NOT include it. At the same time I don't have that much interest arguing about it either.
That is to say, here is a different take, if you don't like it, I'm dropping the subject.
cheers
fs/namei.c | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-)
fs/namei.c +9 -4
@@ -275,14 +275,19 @@ EXPORT_SYMBOL(getname_kernel);void putname (struct filename *name) { +int refcnt; +if (IS_ERR_OR_NULL (name))return ; -if (WARN_ON_ONCE (!atomic_read (&name->refcnt))) -return ; + refcnt =atomic_read (&name->refcnt); +if (refcnt !=1 ) { +if (WARN_ON_ONCE (!refcnt)) +return ; -if (!atomic_dec_and_test (&name->refcnt)) -return ; +if (!atomic_dec_and_test (&name->refcnt)) +return ; + }if (name->name != name->iname) {__putname (name->name);
Signature · 1 line
2.43.0
Quoted 8 lines · Mateusz Guzik
While the structure is refcounted, the only consumer incrementing it is audit and even then the atomic operation is only needed when it interacts with
io_uring.If putname spots a count of 1, there is no legitimate way for anyone to bump it.
[...]
Applied to the vfs-6.15.misc branch of the vfs/vfs.git tree. Patches in the vfs-6.15.misc branch should appear in linux-next soon.
Please report any outstanding bugs that were missed during review in a new review to the original patch series allowing us to drop it.
It's encouraged to provide Acked-bys and Reviewed-bys even though the patch has now been applied. If possible patch trailers will be updated.
Note that commit hashes shown below are subject to change due to rebase, trailer updates or similar. If in doubt, please check the listed branch.
tree: https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git branch: vfs-6.15.misc
[1/1] fs: dodge an atomic in putname if ref == 1
https://git.kernel.org/vfs/vfs/c/c93617c0f22c
Quoted 14 lines · Mateusz Guzik
While the structure is refcounted, the only consumer incrementing it is audit and even then the atomic operation is only needed when it interacts with
io_uring.If putname spots a count of 1, there is no legitimate way for anyone to bump it.
If audit is disabled, the count is guaranteed to be 1, which consistently elides the atomic for all path lookups. If audit is enabled, it still manages to elide the last decrement.
Note the patch does not do anything to prevent audit from suffering atomics. See [1] and [2] for a different approach.
Quoted 7 lines
Benchmarked on Sapphire Rapids issuing
access()(ops/s):before: 5106246 after: 5269678 (+3%)Link 1: https://lore.kernel.org/linux-fsdevel/[email protected]/
Link 2: https://lore.kernel.org/linux-fsdevel/20250307164216.GI2023217@ZenIV/Signed-off-by Mateusz Guzik <[email protected]>
Yeah, I like this much more than the original. Feel free to add:
Honza
Quoted 48 lines
This is an alternative to the patch I linked above.
I think the improved commit message should also cover the feedback Christian previously shared concerning it.
This is a trivial win until the atomic issue gets resolved, I don't see any reason to NOT include it. At the same time I don't have that much interest arguing about it either.
That is to say, here is a different take, if you don't like it, I'm dropping the subject.
cheers
fs/namei.c | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-)fs/namei.c +9
-4@@ -275,14 +275,19 @@ EXPORT_SYMBOL(getname_kernel);void putname (struct filename *name) { +int refcnt; +if (IS_ERR_OR_NULL (name))return ; -if (WARN_ON_ONCE (!atomic_read (&name->refcnt))) -return ; + refcnt =atomic_read (&name->refcnt); +if (refcnt !=1 ) { +if (WARN_ON_ONCE (!refcnt)) +return ; -if (!atomic_dec_and_test (&name->refcnt)) -return ; +if (!atomic_dec_and_test (&name->refcnt)) +return ; + }if (name->name != name->iname) {__putname (name->name);Signature · 1 line
2.43.0
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Quoted 15 lines · Mateusz Guzik
fs/namei.c +2
-2@@ -275,14 +275,19 @@ EXPORT_SYMBOL(getname_kernel);void putname (struct filename *name) { +int refcnt; +if (IS_ERR_OR_NULL (name))return ; -if (WARN_ON_ONCE (!atomic_read (&name->refcnt))) -return ;
fs/namei.c EXPORT_SYMBOL(getname_kernel); · line 290
Quoted 10 lines
fs/namei.c +7
-2+ refcnt =atomic_read (&name->refcnt); +if (refcnt !=1 ) { +if (WARN_ON_ONCE (!refcnt)) +return ; -if (!atomic_dec_and_test (&name->refcnt)) -return ; +if (!atomic_dec_and_test (&name->refcnt)) +return ; + }
Looks good to me, I use this trick with bitops too all the time, to avoid a RMW when possible.
Signature · 1 line
Jens Axboe